http://researchspace.auckland.ac.

nz

ResearchSpace@Auckland

Copyright Statement

The digital copy of this thesis is protected by the Copyright Act 1994 (New
Zealand).

This thesis may be consulted by you, provided you comply with the provisions of
the Act and the following conditions of use:

• Any use you make of these documents or images must be for research or
private study purposes only, and you may not make them available to any
other person.
• Authors control the copyright of their thesis. You will recognise the
author's right to be identified as the author of this thesis, and due
acknowledgement will be made to the author where appropriate.
• You will obtain the author's permission before publishing any material from
their thesis.

To request permissions please use the Feedback form on our webpage.

http://researchspace.auckland.ac.nz/feedback

General copyright and disclaimer

In addition to the above conditions, authors give their consent for the digital copy
of their work to be used subject to the conditions specified on the Library Thesis
Consent Form and Deposit Licence.
MODELLING THE FLUSHING MECHANISM OF
THIN FLEXIBLE SURFACE PAVEMENTS

Sachi Kodippily

Supervised by
Dr Theuns Henning
and
Professor Jason Ingham

A thesis submitted in partial fulfilment of the requirements for the degree of Doctor of
Philosophy in Civil Engineering, The University of Auckland, 2013
 ABSTRACT

The research presented in this thesis investigated the occurrence of flushing of chip
seal pavements. The research aimed to determine the effects of volumetric properties
of chip seal surfaces on flushing, focusing on the relationship between air voids and
flushing initiation. Additionally, the research aimed to develop a pavement
deterioration model to predict the initiation and progression of flushing on chip seal
pavements.

Mechanical testing was conducted on cores obtained from in-service, flushed chip
seal pavements from Auckland, Waikato, Christchurch and Dunedin regions of New
Zealand. The tests that were performed on the cores included wheel tracking, air voids
measurement, bitumen extractions and sieve analyses to determine aggregate grading
profiles. Measurements were made of the depth and pattern of deformation that had
developed on the cores during wheel tracking. Samples extracted from tested cores
were scanned using a computed tomography scanner and the scan images were
analysed using image analysis techniques to calculate the volume of air voids within
the samples. The reduction in air void volume during wheel tracking was compared to
the flushing that had occurred on the samples to establish the relationship between
flushing and air voids. Furthermore, data analysis was performed on pavement
condition data gathered from New Zealand’s long-term pavement performance
database to identify the combination of factors that provided the best prediction of
flushing, and regression analysis was performed to develop a model to predict the
initiation and progression of flushing.

The laboratory testing described above revealed that the thickness of a chip seal
surface has a direct correlation to flushing, where thicker surfaces tend to have more
severe flushing. Moreover, the reduction in air void volume that occurred in a chip
seal structure due to loading was directly related to the amount of flushing likely to
occur on that surface. The pattern of deformation of a chip seal structure provided an
indication of its state of stability, which in turn indicated the best method of treatment
for flushing. From the data analysis it was identified that the combination of factors
that provided the best indication of flushing were surface thickness, surface age, rut

i
depth and grade of aggregates. The flushing initiation model had an accuracy of 76%
and the flushing progression model was also statistically strong at predicting the
quantity of flushing. Using these research outcomes, a pavement condition assessment
guideline was developed to aid with managing flushed pavements. Overall, this
research has significantly increased the understanding of the mechanisms that lead to
flushing and established ways to better identify and manage flushing.

ii
 ACKNOWLEDGEMENTS

I would like to express my sincere thanks to my supervisors Dr. Theuns Henning and
Professor Jason Ingham, for providing invaluable input into this work over the years. I
am grateful for your guidance, research ideas and time.

I would like to acknowledge The University of Auckland doctoral scholarship

programme and New Zealand Transport Agency for providing financial assistance for
this research. A big thanks also goes to the following organisations for their technical
assistance and input into the research: Fulton Hogan Ltd Auckland branch and
Bitumen Laboratory, Fulton Hogan Ltd Dunedin, Fulton Hogan Ltd Christchurch,
New Zealand Transport Agency Dunedin branch, Christchurch City Council, OPUS
Ltd, Transfield Services Ltd and The University of Auckland Bioengineering
Institute.

Most importantly, I would like to thank my husband Bhagya, for his love, patience
and constant encouragement throughout this journey. To my parents and brother,
thank you for your loving support.

iii
 TABLE OF CONTENTS

Abstract ............................................................................................................................ i

Acknowledgements ........................................................................................................iii

Table of Contents ........................................................................................................... iv

List of Tables .................................................................................................................. x

List of Figures ................................................................................................................ xi

Glossary of Terms ......................................................................................................... xv

Chapter 1 - INTRODUCTION ....................................................................................1-1

1.1 Background to the Flushing Defect .....................................................1-1

1.2 Problem Statement ...............................................................................1-4

1.2.1 The Need for a Fundamental Understanding of Pavement
Material Behaviour Leading to Flushing .............................................1-4
1.2.2 The Need for a Model to Predict Flushing...............................1-4
1.2.3 The Use of Non-Destructive Visualisation Methods for
Assessing Chip Seal Structures ............................................................1-5

1.3 Research Objectives .............................................................................1-6

1.4 Scope and Structure of the Thesis ........................................................1-7

1.4.1 Outline of Thesis Chapters.......................................................1-8

1.5 Data Sources used for Modelling Flushing ........................................1-11

1.5.1 Laboratory-Based Data Obtained from Field Samples ..........1-11
1.5.2 The Long Term Pavement Performance Programme ............1-11

Chapter 2 - LITERATURE REVIEW .......................................................................2-13

2.1 General Pavement and Surfacing Technology ...................................2-13

2.1.1 Road Classification and Surfacing Options ...........................2-13
2.1.2 Construction ...........................................................................2-14
2.1.3 Maintenance ...........................................................................2-15

2.2 Surface and Pavement Failure Mechanisms ......................................2-15

2.2.1 Flushing..................................................................................2-15
2.2.2 Bleeding .................................................................................2-16
iv
 2.2.3 Deformation ...........................................................................2-17
2.2.4 Cracking .................................................................................2-19
2.2.5 Chip Loss ...............................................................................2-19

2.3 World-Wide Surfacing Issues ............................................................2-20

2.4 Currently Available Evidence on Flushing ........................................2-21

2.5 Mechanics of Bitumen and Asphalt ...................................................2-29

2.5.1 Comparison of Chip Seal Pavements and Asphalt Concrete
Pavements ..........................................................................................2-29
2.5.2 Modelling of Asphalt Surfaces ..............................................2-30

2.6 Interactions between Flushing, Shoving and Rutting ........................2-32

2.7 Pavement Defect Modelling ..............................................................2-33

2.7.1 The World Bank HDM-III and HDM-4 Pavement Deterioration
Models................................................................................................2-33
2.7.2 Pavement Deterioration Modelling in HDM-4 ......................2-34

2.8 Pavement Defect Prediction Models Developed for New Zealand

Pavements ..........................................................................................2-36
2.8.1 Crack Initiation Model ...........................................................2-36
2.8.2 Rut Progression Model ..........................................................2-37

2.9 Current Texture Model of NZdTIMS and Previously Developed

Flushing Model ..................................................................................2-38
2.9.1 The Existing Texture Model of NZdTIMS ............................2-38
2.9.2 Previously Developed Flushing Model ..................................2-42
2.9.3 Other Methods of Assessing Flushing ...................................2-43

2.10 Data Analysis and Modelling Approaches Used for Pavement

Deterioration Modelling ....................................................................2-45
2.10.1 Deterministic Models .............................................................2-46
2.10.2 Probabilistic Models ..............................................................2-48
2.10.3 Data Mining Approaches .......................................................2-49
2.10.4 Statistical Data Analysis Methods used in the Study.............2-51

2.11 Summary of Literature Review ..........................................................2-52

v
Chapter 3 - EXAMINING THE VOLUMETRIC BEHAVIOUR OF CHIP SEAL
MATERIALS .................................................................................................3-54

3.1 Introduction ........................................................................................3-54

3.2 Objectives ..........................................................................................3-55

3.3 Hypothesis of Relationship between Chip Seal Materials .................3-55

3.4 Sampling of Field Pavement Cores....................................................3-58

3.4.1 Assumptions Used in Sampling .............................................3-60

3.5 Laboratory Testing Methodology ......................................................3-61

3.5.1 Measurement of Volume of Air Voids ..................................3-63
3.5.2 Wheel Tracking ......................................................................3-65
3.5.3 Bitumen Extraction and Aggregate Grading Measurements .3-68
3.5.4 Aggregate Shape and Angularity Measurements ...................3-69

3.6 Volume of Air Void ...........................................................................3-70

3.7 Wheel Tracking Results .....................................................................3-72

3.7.1 Auckland Sites .......................................................................3-74
3.7.2 Waikato Sites .........................................................................3-76
3.7.3 Christchurch Sites ..................................................................3-78
3.7.4 Dunedin Sites .........................................................................3-79
3.7.5 Summary of Wheel Tracking Results ....................................3-81
3.7.6 The Role of Chip Seal Thickness on Flushing Patterns .........3-82

3.8 Binder Content Results ......................................................................3-84

3.9 Relationship between Aggregate Grading and Flushing ...................3-87

3.10 Discussion ..........................................................................................3-93

3.10.1 Availability of Standard Test Methods for Investigation of Chip
Seals ................................................................................................3-94
3.10.2 Comparison of Observed Chip Seal Behaviour to Previous
Studies ................................................................................................3-96
3.10.3 Availability of Chip Seal Samples for Testing ......................3-96

3.11 Conclusions ........................................................................................3-97

3.12 Recommendations for Further Investigations ....................................3-99

vi
Chapter 4 - USE OF IMAGE ANALYSIS TO QUANTIFY CHIP SEAL MATERIAL
INTERACTIONS .........................................................................................4-101

4.1 Introduction ......................................................................................4-101

4.2 Use of Image Analysis in Pavement Material Modelling ................4-102

4.3 Computed Tomography Scanning of Cores .....................................4-103

4.3.1 Calculation of Area and Volume of Air Voids ....................4-104
4.3.2 Three Dimensional Modelling of Air Voids ........................4-108

4.4 Air Void Volumes and Three Dimensional Reconstructions of Air

Void Distributions............................................................................4-109
4.4.1 Waikato Samples .................................................................4-109
4.4.2 Auckland Samples ...............................................................4-116
4.4.3 Dunedin Samples .................................................................4-118
4.4.4 Christchurch Samples ..........................................................4-122

4.5 Relationship of Air Voids Reduction to Flushing Trends ...............4-125

4.5.1 Inter-Relationship between Binder Volume, Air Void Volume
and Flushing .....................................................................................4-128

4.6 Discussion ........................................................................................4-130

4.6.1 Effectiveness of Computed Tomography Scanning ............4-130
4.6.2 Limitations of Equipment ....................................................4-131
4.6.3 Difficulties with Sample Extraction.....................................4-132

4.7 Conclusions ......................................................................................4-133

4.8 Recommendations and Further Research .........................................4-135

Chapter 5 - DEVELOPMENT OF A FLUSHING PREDICTION MODEL ..........5-137

5.1 Introduction ......................................................................................5-137

5.2 Data Set Description ........................................................................5-138

5.2.1 The Long-Term Pavement Performance Data Set ...............5-138
5.2.2 Study Methodology ..............................................................5-140

5.3 Data Analysis Results ......................................................................5-141

5.3.1 Factor Analysis of Variables Causing Flushing...................5-141
5.3.2 Correlation Analysis of Variables Causing Flushing...........5-143
5.3.3 Summary of Findings from Correlation Analyses ...............5-152
vii
 5.4 Predicting the Initiation of Flushing ................................................5-154
5.4.1 Model Development.............................................................5-155
5.4.2 Test of Logistic Model Accuracy ........................................5-160

5.5 Predicting the Progression of Flushing ............................................5-162

5.5.1 Testing of Flushing Progression Model ...............................5-167

5.6 Limitations of the Data Set Affecting Flushing Prediction .............5-169

5.6.1 Traffic Volume Data ............................................................5-169
5.6.2 Unavailability of Air Voids Data .........................................5-169

5.7 Conclusions ......................................................................................5-170

5.7.1 Recommended Model Formats for Predicting Flushing ......5-171

5.8 Recommendations ............................................................................5-172

5.8.1 The Need for Testing of Flushing Prediction Model ...........5-172
5.8.2 Use of Model Outcomes for Better Prediction of Flushing .5-175

Chapter 6 - SUMMARY AND RECOMMENDATIONS OF RESEARCH

FINDINGS ...................................................................................................6-176

6.1 Introduction ......................................................................................6-176

6.2 Implications of Research Findings...................................................6-177

6.2.1 Mechanical Testing of Pavement Samples ..........................6-178
6.2.2 Image Analysis of Pavement Samples .................................6-181
6.2.3 Data Analysis and Prediction Model Development .............6-183
6.2.4 Pavement Condition Assessment Guideline for Flushing ...6-184

6.3 Limitations of the Research .............................................................6-186

6.3.1 Industrial Applications of Research Findings ......................6-186
6.3.2 Applicability of the Flushing Prediction Model ..................6-186

6.4 Recommendations ............................................................................6-187

6.4.1 Model Calibration and Data Requirements..........................6-187
6.4.2 Characterisation of Chip Seal Structures for Accurate Selection
of Maintenance Treatments..............................................................6-188
6.4.3 Assessing the Viability of CT Scanning in Practice ............6-188

Chapter 7 - CONCLUSIONS...................................................................................7-190

7.1 Research Conclusions ......................................................................7-190

viii
 7.2 Overall Understanding of Chip Seal Behaviour Developed Through the
Research Findings ............................................................................7-194

REFERENCES ........................................................................................................... 197

APPENDIX A – INVENTORY OF CORES.............................................................. 204

APPENDIX B – BITUMEN CONTENT, AGGREGATE GRADING, AND

AGGREGATE ANGULARITY MEASUREMENTS ................................... 206

APPENDIX C – AIR VOID AREA PLOTS OF SAMPLES ..................................... 212

APPENDIX D – AIR VOID VOLUME PLOTS OF SAMPLES ............................... 223

APPENDIX E – 3D RECONSTRUCTIONS OF CHIP SEAL SAMPLES ............... 227

APPENDIX F – 3D RECONSTRUCTIONS OF AIR VOIDS DISTRIBUTIONS ... 235

APPENDIX G – DATA ANALYSIS REFERENCES ............................................... 237

ix
 LIST OF TABLES

Table 2.1 Rankings of reasons for resealing (adapted from Oliver (1999)) .................... 2-22
Table 2.2 A summary of flushing research completed in New Zealand .......................... 2-25
Table 2.3 Pavement defects modelled in HDM-4 (Morosiuk and Kerali, 2001) ............. 2-35
Table 3.1 Field sampling design matrix ........................................................................... 3-60
Table 3.2 Test methods used in the study ........................................................................ 3-61
Table 3.3 The number of wheel passes used to track cores ............................................. 3-67
Table 3.4 Air void volume measurements ....................................................................... 3-71
Table 3.5 Binder contents of wheel tracked cores ........................................................... 3-85
Table 3.6 Main conclusions from laboratory testing ....................................................... 3-97
Table 4.1 Reduction in air void volumes calculated from CT scan images ................... 4-126
Table 4.2 Main conclusions of the study ....................................................................... 4-134
Table 5.1 Factors likely to affect flushing ..................................................................... 5-139
Table 5.2 Correlation matrix for pavement composition and traffic-related
factors – LTPP data set .................................................................................. 5-142
Table 5.3 Correlation matrix for climatic factors........................................................... 5-143
Table 5.4 Model coefficients from logistic regression to model initiation of
flushing .......................................................................................................... 5-159
Table 5.5 Prediction accuracy of the logistic model for flushing initiation ................... 5-161
Table 5.6 Linear regression model coefficients for first generation seals...................... 5-164
Table 5.7 Linear regression model coefficients for second and higher generation
seals ................................................................................................................ 5-165
Table 5.8 Data-driven models to predict flushing .......................................................... 5-171
Table A.1 Inventory of cores used for wheel tracking ........................................................204
Table B.1 Bitumen content measurements for chip seal cores ...........................................206
Table B.2 Aggregate grading measurements ......................................................................208
Table B.3 Aggregate angularity measurements of chip seal cores .....................................210
Table G.1 Variable Selection table .....................................................................................239
Table G.2 Variable Significance table ................................................................................239
Table G.3 Parameter Estimates table for logistic regression model ...................................239
Table G.4 Classifications table for logistic regression model ............................................241
Table G.5 Coefficients table from linear regression ...........................................................243
Table G.6 Excluded variables table from linear regression ................................................244
Table G.7 Model summary table of linear regression .........................................................244

x
 LIST OF FIGURES

Figure 1.1 A chip seal pavement with flushing on the wheel path ...................................... 1-1
Figure 1.2 Structure of the thesis......................................................................................... 1-8
Figure 2.1 Types of pavement deformation (Transit New Zealand et al., 2005) .............. 2-18
Figure 2.2 Types of cracking on bituminous surfaces (Austroads, 2006b) ....................... 2-19
Figure 2.3 Most common distress modes on chip seal surfaces (adapted from
Gransberg and James (2005)) .......................................................................... 2-20
Figure 2.4 Reasons for resealing of chip seals in 2002-2003 (adapted from Ball
et al. (2004)) ..................................................................................................... 2-21
Figure 2.5 Reasons for resurfacing chip seal surfaces in 1992-1993 (adapted
from Ball et al. (2004)) .................................................................................... 2-23
Figure 2.6 Material components and volumes in an asphalt pavement layer
(adapted from Krishnan and Rao (2000)) ........................................................ 2-31
Figure 2.7 Probability of cracking in a given year – output from the logit model
(Henning, 2008) ............................................................................................... 2-37
Figure 2.8 Output from the logistic model predicting the initiation of accelerated
rut progression (Henning et al., 2009) ............................................................. 2-38
Figure 2.9 Texture depth model from the HDM format and simplified format
(from Henning et al. (2006)) ............................................................................ 2-39
Figure 2.10 Plots of the fitted model (based on the original data) compared to the
calibration data (Henning et al., 2006) ............................................................. 2-41
Figure 2.11 Pavement macrotexture measuring devices; Circular Track Meter and
Outflow Meter (Martino, 2007) ....................................................................... 2-44
Figure 2.12 Illustration of MPD of a pavement surface (Flintsch et al., 2003) ................... 2-44
Figure 2.13 The process of knowledge discovery in databases (adopted from Tan
et al. (2005)) ..................................................................................................... 2-49
Figure 3.1 Constituents in a chip seal layer ....................................................................... 3-56
Figure 3.2 Map of sampling locations across New Zealand.............................................. 3-59
Figure 3.3 Classifications of sampled sites ....................................................................... 3-59
Figure 3.4 Illustration of sampling locations at a site ....................................................... 3-60
Figure 3.5 Top and side views of a core from Christchurch site 1 .................................... 3-63
Figure 3.6 Setup of scale with water bath ......................................................................... 3-64
Figure 3.7 Setup of vacuum sealing apparatus .................................................................. 3-65
Figure 3.8 Setup of the wheel tracking machine ............................................................... 3-66
Figure 3.9 A core set in the wooden test mould and prepared to be wheel tracked .......... 3-66
Figure 3.10 Setup of light table and camera ........................................................................ 3-69
Figure 3.11 Aggregate shape analysis images ..................................................................... 3-70
Figure 3.12 Schematic showing the development of the stable rutting pattern and
the corresponding state of aggregate layers in the core ................................... 3-73
Figure 3.13 Schematic showing the development of the continuous rutting pattern
and the corresponding state of aggregate layers in the core............................. 3-74
Figure 3.14 Wheel tracking results for Auckland cores ...................................................... 3-75
Figure 3.15 Wheel tracking results for Waikato cores ........................................................ 3-77
Figure 3.16 Wheel tracking results for Christchurch cores ................................................. 3-79
Figure 3.17 Wheel tracking results for Dunedin cores ........................................................ 3-81
Figure 3.18 Relationship between core thickness and flushing on wheel tracked
cores ................................................................................................................. 3-83
Figure 3.19 Initial rut rate measured from wheel tracking test against the
thickness of core .............................................................................................. 3-84
Figure 3.20 Correlation between flushing and the bitumen-to-aggregate ratio ................... 3-87
Figure 3.21 Grading curves for wheel tracked cores classified into the extent of
flushing ............................................................................................................ 3-88
xi
Figure 3.22 Aggregate grading curves of wheel tracked cores classified into
stability of the core........................................................................................... 3-90
Figure 3.23 Angularity of aggregates classified into the deformation pattern of the
cores ................................................................................................................. 3-92
Figure 4.1 Details of CT scanning................................................................................... 4-104
Figure 4.2 A sample that was used for scanning ............................................................. 4-104
Figure 4.3 CT scan images of a pavement sample showing the original image
and the binary converted image ..................................................................... 4-105
Figure 4.4 Illustration of air void volume analysis of samples ....................................... 4-106
Figure 4.5 Method used for selecting threshold values for binary conversion of
scan images of samples of Core 1 from Waikato site 1 ................................. 4-107
Figure 4.6 CT scan images of the samples taken from Core 1 from Waikato site
1 (5,000 wp) ................................................................................................... 4-109
Figure 4.7 Variation in air void volume along the depth of the tracked samples
and the untracked samples from Waikato site 1 cores ................................... 4-112
Figure 4.8 3D reconstruction of the untracked sample and the tracked sample of
Core 1 from Waikato site 1 ............................................................................ 4-113
Figure 4.9 3D reconstruction of the air voids distribution in the untracked
sample and the tracked sample of Core 2 from Waikato site 1 ...................... 4-114
Figure 4.10 Volume of air voids along the depth of the tracked sample and the
untracked sample of Core 3 from Waikato site 1........................................... 4-115
Figure 4.11 3D reconstruction of the air voids distribution in the untracked
sample and the tracked sample of Core 3 from Waikato site 1 ...................... 4-115
Figure 4.12 Volume of air voids along the depth of the tracked samples and the
untracked samples of cores from Auckland site 1 ......................................... 4-117
Figure 4.13 3D reconstruction of air voids distribution in the untracked sample
and the tracked sample of Core 2 from Auckland site 1 ................................ 4-118
Figure 4.14 3D reconstruction of air voids distribution in the untracked sample
and the tracked sample of Core 3 from Auckland site 1 ................................ 4-118
Figure 4.15 Volume of air voids along the depth of the tracked samples and the
untracked samples of cores from Dunedin site 1 ........................................... 4-120
Figure 4.16 3D reconstruction of the air voids distribution in the untracked
sample and the tracked sample of Core 1 from Dunedin site 1 ..................... 4-121
Figure 4.17 3D reconstruction of the air voids distribution in the untracked
sample and the tracked sample of Core 2 from Dunedin site 1 ..................... 4-121
Figure 4.18 3D reconstruction of the air voids distribution in the untracked
sample and the tracked sample of Core 3 from Dunedin site 1 ..................... 4-122
Figure 4.19 Volume of air voids along the depth of the tracked samples and the
untracked samples of cores from Christchurch site 2 .................................... 4-124
Figure 4.20 Correlation between flushing measured after wheel tracking tests and
the air void volume reduction calculated from CT scan images .................... 4-127
Figure 4.21 Correlation between flushing and the ratio of bitumen volume to air
voids volume .................................................................................................. 4-129
Figure 5.1 Illustration of flushing ratings used in the LTPP database ............................ 5-139
Figure 5.2 Correlation between flushing and pavement surface characteristics ............. 5-144
Figure 5.3 The amount of flushing with respect to seal generation ................................ 5-145
Figure 5.4 The relationship between flushing and the grade of aggregates .................... 5-147
Figure 5.5 Relationship between flushing and traffic volume......................................... 5-148
Figure 5.6 Correlation between flushing and mean air temperature ............................... 5-149
Figure 5.7 Relationship between flushing and rutting..................................................... 5-151
Figure 5.8 Deterioration phases of flushing .................................................................... 5-155
Figure 5.9 Relationship between flushing and surface texture ........................................ 5-157
Figure 5.10 Effect of independent variables on flushing initiation ................................... 5-158
Figure 5.11 Model outcome to predict the probability of flushing initiation .................... 5-160

xii
Figure 5.12 Predicted probabilities for flushing initiating on LTPP pavement
sections........................................................................................................... 5-162
Figure 5.13 Distribution of original dependant variable TotalFlushing data and the
transformed data............................................................................................. 5-163
Figure 5.14 Evaluation of residuals from linear regression for first generation
seals ................................................................................................................ 5-164
Figure 5.15 Evaluation of residuals from linear regression for second and higher
generation seals .............................................................................................. 5-166
Figure 5.16 Model predictions for flushing progression on second and higher
generation seals .............................................................................................. 5-167
Figure 5.17 Locations of moisture measurements ............................................................. 5-173
Figure 5.18 Correlation between flushing and a) soil water content and b) soil dry
density ............................................................................................................ 5-174
Figure 6.1 Assessment of flushing of chip seal pavements ............................................. 6-185
Figure 7.1 Main methodological components of the research......................................... 7-191
Figure C.1 Area of air voids in the tracked samples and the untracked samples of
cores from Auckland site 1 ................................................................................212
Figure C.2 Area of air voids in the tracked sample and the untracked sample of
Core 2 from Auckland site 2 ..............................................................................213
Figure C.3 Area of air voids in the tracked samples and the untracked samples of
cores from Waikato site 1 ..................................................................................215
Figure C.4 Area of air voids in the tracked sample and the untracked sample of
Core 1 from Christchurch site 1 .........................................................................216
Figure C.5 Area of air voids in the tracked samples and the untracked samples of
cores from Christchurch site 2 ...........................................................................218
Figure C.6 Area of air voids in the tracked samples and the untracked samples of
cores from Dunedin site 1 ..................................................................................220
Figure C.7 Area of air voids in the tracked samples and the untracked samples of
cores from Dunedin site 2 ..................................................................................222
Figure D.1 Volume of air voids along the depth of the tracked sample and the
untracked sample of Core 2 from Auckland site 2 .............................................223
Figure D.2 Volume of air voids along the depth of the tracked sample and the
untracked sample of Core 1 from Christchurch site 1........................................224
Figure D.3 Volume of air voids along the depth of the tracked samples and the
untracked samples of cores from Dunedin site 2 ...............................................226
Figure E.1 3D reconstruction of the untracked sample and the tracked sample of
Core 2 from Auckland site 1 ..............................................................................227
Figure E.2 3D reconstruction of the untracked sample and the tracked sample of
Core 3 from Auckland site 1 ..............................................................................228
Figure E.3 3D reconstruction of the untracked sample and the tracked sample of
Core 2 from Auckland site 2 ..............................................................................229
Figure E.4 3D reconstruction of the untracked sample and the tracked sample of
Core 2 from Waikato site 1 ................................................................................230
Figure E.5 3D reconstruction of the untracked sample and the tracked sample of
Core 3 from Waikato site 1 ................................................................................230
Figure E.6 3D reconstruction of the untracked sample and the tracked sample of
Core 1 from Dunedin site 1 ................................................................................231
Figure E.7 3D reconstruction of the untracked sample and the tracked sample of
Core 2 from Dunedin site 1 ................................................................................231
Figure E.8 3D reconstruction of the untracked sample and the tracked sample of
Core 3 from Dunedin site 1 ................................................................................232
Figure E.9 3D reconstruction of the untracked sample and the tracked sample of
Core 1 from Dunedin site 2 ................................................................................233
Figure E.10 3D reconstruction of the untracked sample and the tracked sample of
Core 2 from Dunedin site 2 ................................................................................233
xiii
Figure E.11 3D reconstruction of the untracked sample and the tracked sample of
Core 3 from Dunedin site 2 ................................................................................234
Figure F.1 3D reconstruction of air voids distribution in the untracked sample
and the tracked sample of Core 2 from Auckland site 2 ....................................235
Figure F.2 3D reconstruction of air voids distribution in the untracked sample
and the tracked sample of Core 1 from Dunedin site 2 ......................................236
Figure F.3 3D reconstruction of air voids distribution in the untracked sample
and the tracked sample of Core 2 from Dunedin site 2 ......................................236
Figure F.4 3D reconstruction of air voids distribution in the untracked sample
and the tracked sample of Core 3 from Dunedin site 2 ......................................236

xiv
 GLOSSARY OF TERMS

Average Annual Daily Total volume of traffic passing a point on a road over the
Traffic (AADT) period of a calendar year, divided by the number of days
in that year.
Base course / base A base course is a sub-layer in a pavement structure that is
used for load distribution and to provide a stable
foundation for the surfacing layers. A base course is
typically constructed of durable aggregates and can be
either stabilised or unstabilised.
Ball penetrometer test A test method which is used to assess the penetration
resistance of a road surface.
CAPTIF Canterbury Accelerated Pavement Indoor Testing Facility
CT scanning Computed Tomography scanning
dTIMS Deighton's Total Infrastructure Management System
Equivalent Standard Number of standard axle loads that are equivalent in
Axles (ESA) damaging effect on a pavement to a given vehicle or axle
loading.
Fog seal A fog seal is a light application of bitumen that is applied
to protect and rejuvenate an existing pavement surface by
waterproofing the surface and preventing aggregate loss
and ravelling.
Geotextile seal A method of rehabilitating a pavement, where a geotextile
is placed under the surface layer to act as a stress-relieving
layer. A geotextile often reduces and prevents reflective
cracking of a pavement surface, and also prevents water
from entering the pavement structure.
HDM-III Highway Design and Maintenance Standards Model
HDM-4 Highway Development and Management Model
Heavy Commercial A commercial vehicle that has a gross vehicle mass
Vehicle (HCV) exceeding 12 tonnes.

xv
Hot Mix Asphalt A material commonly used for constructing pavement
(HMA) surfaces. It is produced by mixing hot bitumen with
aggregates and other additives such as cement, filler and
polymers.
International Measurement of the roughness of a pavement surface. It is
Roughness Index (IRI) found by measuring the response of a specific vehicle
(quarter-car) as it is driven along the pavement (m/km or
in/mi).
Macrotexture Texture on the pavement surface that is determined by the
size, shape and spacing of the aggregates in the seal.
Mean Profile Depth A mean measure of macrotexture, where the pavement
(MPD) surface profile is divided into small segments of a given
base length, typically 100 mm, which are analysed and the
mean segment depths averaged to give an average value
for a specific length of profile.
Medium Commercial A commercial vehicle that has a gross vehicle mass
Vehicle (MCV) between 3.5 tonnes and 12 tonnes.
Microtexture Roughness on the surface of individual pieces of
aggregate in the seal.
Multicollinearity When two or more independent variables in a multiple
regression model are highly correlated.
Newtonian fluid A fluid whose viscous stress at each point is linearly
proportional to its strain rate at that point. Common
Newtonian fluids include liquids, such as water,
hydrocarbons and oils, and all gases.
Pavement Management A pavement management system is a planning tool that
System (PMS) aids with decision making for the entire range of activities
involved in providing and maintaining pavements.
Polymer Modified A type of sealing binder that is made by mixing polymers
Binder (PMB) with binder. Polymer modified binders enhance the
performance of a pavement surface by improving shear
resistance and, reducing reflective cracking and rutting.

xvi
Racked-in seal A type of chip seal that is constructed by applying a layer
of binder which is followed by applying a layer of large
aggregates that are widely spaced. A layer of smaller sized
aggregates is then applied to cover the spaces between the
larger aggregates.
RAMM database Road Assessment and Maintenance Management
database.
Ravelling Loosening of aggregate particles from a pavement surface.
Ravelling often occurs due to a lack of adhesion between
the binder and aggregates and can occur when the
pavement surface is constructed in cooler temperatures.
Sandwich / Bicouche A sandwich seal is a chip seal that consists of a single
Seal layer of large aggregates that is spread directly on the
pavement surface, and onto which a layer of binder is
applied. This is followed by the application of a layer of
smaller sized aggregates.
Sideways Force The SCRIM measures the wet skidding resistance of a
Coefficient Routine pavement surface.
Investigation Machine
(SCRIM)
Single coat seal A type of chip seal that is constructed by applying a single
coat of binder and on top of which a single layer of
aggregates is applied and rolled into place.
Skid resistance Skid resistance is the frictional force between a pavement
surface and a vehicle tyre. The skid resistance of a
pavement surface depends on its microtexture and
macrotexture.
Subbase A sub-layer in a pavement structure that is commonly
placed between the subgrade and the base course. The
subbase is a load-bearing layer within the pavement
structure and helps to spread a load evenly over the
subgrade.

xvii
Subgrade The subgrade refers to the underlying ground on which a
pavement is constructed. The subgrade is often compacted
prior to construction of a pavement and it can also be
stabilised to improve its load-bearing strength.
Texturising seal A type of seal that is used to reduce texture variation of an
existing surface prior to being resurfaced.
Two-coat seal A type of chip seal that is constructed by applying a coat
of binder and followed immediately by applying a layer of
large sized aggregates. A second coat of binder is applied
on top of the aggregate layer and a second layer of smaller
sized aggregates is applied and rolled.
Vehicles Per Day (vpd) The number of vehicles passing a point on a road in both
directions for 24 hours.
Void fill seal A seal that is constructed by spraying a thin layer of
binder and applying a single layer of small sized
aggregates to fit into the texture of the existing chip seal
surface.
Voids in mineral In a phase diagram of a pavement surfacing mixture,
aggregate (VMA) VMA refers to the sum of the volume of air and the
volume of effective binder expressed as a percentage of
the total volume of the mixture.
Ultra High Pressure A type of pavement treatment that uses a high-pressure jet
(UHP) Water cutting of water to remove excess binder from a pavement
surface.

xviii
 Graduate Centre
ClockTower – East Wing
22 Princes Street, Auckland
Co-Authorship Form Phone: +64 9 373 7599 ext 81321
Fax: +64 9 373 7610
Email: [email protected]
www.postgrad.auckland.ac.nz

This form is to accompany the submission of any PhD that contains research reported in published or
unpublished co-authored work. Please include one copy of this form for each co-authored work.
Completed forms should be included in all copies of your thesis submitted for examination and library
deposit (including digital deposit), following your thesis Abstract.

Please indicate the chapter/section/pages of this thesis that are extracted from a co-authored work and give the title
and publication details or details of submission of the co-authored work.
Publication details:
KODIPPILY, S., HENNING, T. F. P. & INGHAM, J. M. (2012) Detecting flushing of thin sprayed seal pavements using
pavement management data. Journal of Transportation Engineering, 138, 665-673.

Sections of thesis related to this publication:

Chapter 5.
Nature of contribution
Lead Author
by PhD candidate

Extent of contribution
95%
by PhD candidate (%)

CO-AUTHORS
Name Nature of Contribution

Dr. Theunis F.P. Henning Involved in the design phase and reviewing phase of the publication

Prof. Jason M. Ingham Involved in the design phase and reviewing phase of the publication

Certification by Co-Authors
The undersigned hereby certify that:
 the above statement correctly reflects the nature and extent of the PhD candidate’s contribution to this
work, and the nature of the contribution of each of the co-authors; and
 in cases where the PhD candidate was the lead author of the work that the candidate wrote the text.

Name Signature Date

Dr. Theunis F.P. Henning 7/03/2013

Prof. Jason M. Ingham 7/03/2013

 Graduate Centre
ClockTower – East Wing
22 Princes Street, Auckland
Co-Authorship Form Phone: +64 9 373 7599 ext 81321
Fax: +64 9 373 7610
Email: [email protected]
www.postgrad.auckland.ac.nz

This form is to accompany the submission of any PhD that contains research reported in published or
unpublished co-authored work. Please include one copy of this form for each co-authored work.
Completed forms should be included in all copies of your thesis submitted for examination and library
deposit (including digital deposit), following your thesis Abstract.

Please indicate the chapter/section/pages of this thesis that are extracted from a co-authored work and give the title
and publication details or details of submission of the co-authored work.
Publication details:
KODIPPILY, S., HENNING, T. F. P., INGHAM, J. M. & HOLLERAN, G. Computed Tomography Scanning for Quantifying
Chipseal Material Volumetrics. Journal of Computing in Civil Engineering, [In press].

Sections of thesis related to this publication:

Chapter 3
Chapter 4, section 4.3 .
Nature of contribution
Lead Author
by PhD candidate

Extent of contribution
95%
by PhD candidate (%)

CO-AUTHORS
Name Nature of Contribution

Dr. Theunis F.P. Henning Involved in the design phase and reviewing phase of the publication

Prof. Jason M. Ingham Involved in the design phase and reviewing phase of the publication

Mr. Glynn Holleran Involved in the reviewing phase of the publication

Certification by Co-Authors
The undersigned hereby certify that:
 the above statement correctly reflects the nature and extent of the PhD candidate’s contribution to this
work, and the nature of the contribution of each of the co-authors; and
 in cases where the PhD candidate was the lead author of the work that the candidate wrote the text.

Name Signature Date

Dr. Theunis F.P. Henning 7/03/2013

Prof. Jason M. Ingham 7/03/2013

Mr. Glynn Holleran 6/03/2013

 Graduate Centre
ClockTower – East Wing
22 Princes Street, Auckland
Co-Authorship Form Phone: +64 9 373 7599 ext 81321
Fax: +64 9 373 7610
Email: [email protected]
www.postgrad.auckland.ac.nz

This form is to accompany the submission of any PhD that contains research reported in published or
unpublished co-authored work. Please include one copy of this form for each co-authored work.
Completed forms should be included in all copies of your thesis submitted for examination and library
deposit (including digital deposit), following your thesis Abstract.

Please indicate the chapter/section/pages of this thesis that are extracted from a co-authored work and give the title
and publication details or details of submission of the co-authored work.
Submission details:
KODIPPILY, S., HENNING, T. F. P. & INGHAM, J. M. Using a Multi-Phase Model to Predict Flushing of Sprayed Seal
Pavements. International Journal of Pavement Engineering, [In Press].

Sections of thesis related to this publication:

Chapter 5.
Nature of contribution
Lead Author
by PhD candidate

Extent of contribution
95%
by PhD candidate (%)

CO-AUTHORS
Name Nature of Contribution

Dr. Theunis F.P. Henning Involved in the design phase and reviewing phase of the publication

Prof. Jason M. Ingham Involved in the design phase and reviewing phase of the publication

Certification by Co-Authors
The undersigned hereby certify that:
 the above statement correctly reflects the nature and extent of the PhD candidate’s contribution to this
work, and the nature of the contribution of each of the co-authors; and
 in cases where the PhD candidate was the lead author of the work that the candidate wrote the text.

Name Signature Date

Dr. Theunis F.P. Henning 7/03/2013

Prof. Jason M. Ingham 7/03/2013

CHAPTER 1 - INTRODUCTION

1.1 Background to the Flushing Defect

Flushing is a defect which has a damaging effect on the functional performance of a
chip seal pavement. Flushing is characterised by a smooth area on the pavement
surface having low skid resistance, as shown in Figure ‎1.1, and it occurs as a result of
upward migration of bitumen which causes full or partial covering of seal aggregates.
Chip seal surface failure due to flushing has become an ongoing problem on state
highways and urban roads in New Zealand. Approximately 30% of chip seal
pavements on the state highway network of New Zealand have failed due to flushing
before reaching half their expected design life. Furthermore, 50% of the network with
chip seals incurred major reductions in design lives due to flushing (Ball and Patrick,
2005).

A flushed
wheel path

a) Flushed wheel path of a chip seal pavement b) Close-up of a flushed chip seal

Figure ‎1.1 A chip seal pavement with flushing on the wheel path

1-1
Flushing can cause major problems in terms of both safety, by reducing skid
resistance, and structural performance of a pavement surface. Unfortunately, flushed
surfaces cannot always be repaired by resurfacing, and they often require a full
rehabilitation treatment. As a result, flushing of chip seal surfaces has been identified
as one of the main reasons for resealing state highways. Approximately $45 million
was spent annually repairing New Zealand pavements due mainly to flushing. This
amounted to over 50% of the‎ NZ‎ Transport‎ Agency’s‎ (NZTA) expenditure on
pavement maintenance (Transit New Zealand et al., 2005).

The majority of areas that have flushing as the primary cause for resurfacing
experience high traffic levels, and as the quantity and load of traffic are constantly
increasing this surface defect needs to be better understood. Additionally, important
factors that affect the design life of pavement seals such as materials, construction
practices, traffic volumes and environmental conditions vary widely even between
different regions of the country. This variability makes it extremely difficult to apply
a single model to predict deterioration of all pavement surfaces. As increased
importance is given to achieving good texture, sealing practices have been altered to
reduce binder application rates to avoid flushing. In the process this updated
construction procedure has increased the occurrence of chip loss, which has then
become another cause of resurfacing (Lawson and Senadheera, 2009).
Conventionally, chip sealing practitioners have selected binder grades and binder
application rates based on their experience. However, for future roading practitioners
these variables will have to be selected by trial and error. Although there are
guidelines available for the selection of binders (Waters, 2008), these guidelines do
not show the future performance of the binders and seals. Hence modelling pavement
surface defects such as flushing is essential to understanding the future condition of
road assets.

The two other pavement defects that have a significant effect on the accurate
modelling of flushing are rutting and shoving. There is a strong interaction between
these defects, and thus identifying the effect of these three defects on each other is the
key to developing accurate pavement performance forecasting models. As pavements
are resurfaced repeatedly, the chip seal layers deepen (between 40 to 100 mm) and the
bitumen content increases, and these thick chip seal layers often have high binder-to-
1-2
stone ratios (greater than 20% by mass). The top layers become unstable, and act
similar to poorly graded asphalt layers. This instability causes shearing between the
layers (shoving), and the high bitumen content leads to flushing (Gray and Hart,
2003). If the causes of seal failure are not understood properly, the current surfacing
practices may be leading to major seal failures. The seals that are deteriorating due to
defects such as flushing or shoving follow a catastrophic failure, unlike the more
gradual decrease in seal life over a number of seals (Ball et al., 2005). By modelling
flushing, construction practices can be improved to minimise catastrophic seal failures
and allow cost savings in the future.

In addition to the external factors that affect flushing, such as the volume and the type
of traffic loading applied on the pavement, climatic effects, and characteristics of the
pavement such as thickness, age and materials used in construction (Alderson, 2008;
Kodippily et al., 2012; Lawson and Senadheera, 2009), the micromechanical
interactions that take place between materials internally in a chip seal layer, such as
air voids reduction, bitumen absorption by aggregates and interlocking of aggregates,
are also important for the occurrence of flushing. The effects of the internal factors
that lead to flushing are difficult to identify as establishing these effects requires
investigating the interactions between the constituents that make up a chip seal layer.
Currently there is limited published research that has investigated the
micromechanical changes that occur in chip seals that are prone to flushing. A number
of studies have investigated the volumetric behaviour of asphalt concrete pavements
(Krishnan and Rao, 2001; Masad et al., 1999b), and even though asphalt concrete
pavements are different to chip seal pavements by the manner in which the pavement
is constructed, knowledge about the volumetric behaviour of materials in asphalt
concrete pavements can be utilised to understand the volumetric behaviour of chip
seal materials and the micromechanical changes that take place within a chip seal
layer.

1-3
1.2 Problem Statement
1.2.1 The Need for a Fundamental Understanding of Pavement
Material Behaviour Leading to Flushing
The main constituents that make up a chip seal layer are aggregates, bitumen and air
voids, and the volumetric balance of these constituents is important to the safe
structural performance of a chip seal. Surface defects such as flushing and bleeding
are likely to be caused by an imbalance in the volumetric composition of the
materials. Many of the currently published literature that have investigated flushing
have used empirical results to determine the causes of flushing, and they have not
investigated the effect of volumetrics on flushing occurrence. The volumetric balance
of a chip seal layer is significantly affected by the presence of air voids, which
provide void spaces for bitumen movement and allow compaction and densification of
a layer when subjected to compressive loading without failure occurring. Over time,
the volume of air voids decrease and with this decrease the availability of void spaces
for bitumen movement also decreases. This occurrence leads to bitumen being exuded
to the pavement surface, resulting in flushing. There is limited published research that
has investigated the reduction in the volume of air voids in chip seal layers with
respect to flushing (Ball and Patrick, 2008) and the only available literature on this
subject is on the behaviour of asphalt concrete pavements (Krishnan and Rao, 2001;
Masad et al., 1999b). In asphalt pavements the reduction in the volume of air voids is
affected by the magnitude of loading and the initial volume of air voids present, and
this understanding can be used to examine when failure due to flushing begins in a
chip seal pavement layer.

1.2.2 The Need for a Model to Predict Flushing

Mathematical modelling of flushing can have a significant impact in terms of
predicting future maintenance needs, expenditure, and performance of a pavement.
However, this modelling can be extremely difficult due to the complex nature of
flushing and the wide variety of materials, construction practices and environmental
conditions involved. At present, there is no model incorporated in the World‎Bank’s‎
Highway Development and Management Model (HDM-4) that can model flushing of
chip seal surfaces. The only model used that can predict surface texture deterioration
1-4
is a texture model (Parkman and Rainsford, 2005), but, this model is unable to predict
the accelerated phase of texture loss, which is flushing. As a result, flushing is often
not modelled in pavement management systems. Although previous research has
investigated the use of a model to predict the initiation of flushing, it was unsuccessful
at producing a model that was statistically robust or that could accurately forecast the
quantity of flushing (Ul-Islam, 2006). This model has not been adopted due to its
limitations.

When modelling flushing, an understanding needs to be gained about the different

phases of flushing development before the flushing reaches a stage that affects the
safe and functional performance of a chip seal surface. Some pavements could have a
small amount of flushing that still allows that pavement to be functional by having
sufficient texture depth for safe use by vehicles. However, on other pavements the
flushing could develop rapidly and hinder the functional performance of the
pavement. Thus, it is important that a model for flushing be able to differentiate
between the severities of flushing to indicate whether a pavement needs to be
monitored for future flushing.

1.2.3 The Use of Non-Destructive Visualisation Methods for

Assessing Chip Seal Structures
Measuring the volumetric properties of pavement surfacing layers is conventionally
performed using indirect or destructive methods. The deficiency of these indirect
methods is that they lack accuracy and fail to show the properties of a pavement layer
in its insitu state. Particularly when examining the air voids within a chip seal layer,
the destructive methods fail to show how the air voids are distributed within a chip
seal layer. Knowledge of the air voids distribution is important when attempting to
determine whether a chip seal is likely to fail through flushing due to an inadequate
distribution of air voids. To overcome this deficiency in conventional test methods,
non-destructive assessment methods in the form of x-ray scanning can be used. X-ray
scanning methods, such as Computed Tomography (CT) scanning, are widely used in
the field of medical imaging to produce images of internal organs and other objects
(Masad and Sivakumar, 2004), and this method in combination with image analysis

1-5
techniques can be an effective method of producing three-dimensional models of an
object. Literature has been published on the use of CT scanning and image analysis
techniques for evaluating internal structures of asphalt pavement layers (Masad et al.,
1999a; Brzezicki and Kasperkiewicz, 1999; Tashman et al., 2007), however, the use
of these techniques for chip seal layer assessment has not previously been
documented. Imaging techniques can provide valuable information about the internal
structure of chip seal layers, and particularly how the changes in the volume and
distribution of air voids affect the occurrence of flushing. The ability of these imaging
techniques to isolate the constituents in a chip seal layer can show how well-
performing seals differ from poorly-performing seals and this understanding can aid
with improving chip seal design and construction practices.

1.3 Research Objectives

There were three primary objectives for this research. These objectives were to:
a) Develop a mechanistic understanding of the processes involved with the flushing
defect, including:
 Identifying the main causes of flushing of chip seal pavements;
 Investigating the relationship between flushing and variables known to cause
flushing;
 Understanding the inter-effects of other defects such as rutting on flushing;
and
 Providing guidance on ways to improve the maintenance methods of flushed
pavements.

b) Understand the micromechanical behaviour and interactions that occur between

the materials that make up a chip seal surface, including:
 Determining how loading affects the distribution and reduction of air voids
within a chip seal pavement;
 Determining the effect of air voids reduction on flushing occurrence;
 Determining the effects that the aggregate grading profile of a chip seal
surface has on its flushing potential; and

1-6
  Investigating the effectiveness of using image analysis techniques to quantify
the volumetric changes that occur in a chip seal layer due to loading.

c) Develop a pavement deterioration model to predict the progression of flushing on

chip seal pavements, including:
 Identifying the combination of factors that provide the best indication of
flushing potential of a chip seal surface;
 Determining a threshold to identify when a pavement surface is flushed; and
 Developing a data-driven model to predict the initiation and progression of
flushing on a chip seal pavement.

1.4 Scope and Structure of the Thesis

In the research that is presented in this thesis flushing was investigated using three
main methodological components, which were laboratory testing of chip seal samples,
image analysis of chip seal samples, and analysis of pavement performance data.
These three methodological components provided an understanding of flushing at a
different scale. The purposes of conducting the laboratory testing and image analysis
tasks were to understand the fundamental behaviour of chip seal layers that lead to the
development of flushing, particularly the volumetric and micromechanical changes
that occur within a chip seal layer that results in flushing. The purpose of the data
analysis task was to identify from network-level pavement performance data the
factors that can be used to indicate flushing or flushing potential of chip seal sites, and
the identified factors were used to develop a model to predict flushing on field chip
seal pavement sites. The laboratory testing and image analysis components of the
research were unrelated to the data analysis and model development component of the
research. While the three methodological components investigated flushing
differently, the combination of the research outcomes resulted in gaining a thorough
understanding of the processes that lead to flushing. Presented in Figure ‎1.2 is the
structure of this thesis, and an outline of each chapter is presented in Section 1.4.1.

1-7
 Chapter 1 - Inroduction

Chapter 2 - Literature Review

Chapter 3 - Examining the Chapter 4 - Use of image

Chapter 5 - Development of a
volumetric behaviour of chip analysis to quantify chip seal
flushing prediction model
seal materials material interactions

Chapter 6 - Summary and

recommendations of
research findings

Chapter 7 - Conclusions of
the Research

Figure ‎1.2 Structure of the thesis

1.4.1 Outline of Thesis Chapters

Chapter One – Introduction
Chapter One presents an introduction to the thesis and the research that is presented in
the thesis. This chapter gives an overview of the flushing defect and describes the
context of the flushing defect. The chapter also identifies the issues that surround the
understanding of the flushing defect and sets out the main objectives of the presented
research. This chapter presents an outline of the thesis structure and also introduces
the data sets that were used in the research.

1-8
Chapter Two – Literature review
Chapter Two presents an extensive review of literature that is relevant to the presented
research. The topics that are discussed in the literature review include:
 Background to the behaviour of materials that make up a pavement layer, and
the relationship between chip seal materials and how they interact with respect
to flushing;
 Previous research that was undertaken to determine the nature of the flushing
mechanism on chip seal pavements. This section includes a summary of the
methodologies followed in earlier research on flushing as well as the outcomes
and recommendations that were reached.
 Background of the HDM-III and HDM-4 models and their limitations with
respect to modelling the flushing mechanism;
 Existing texture models and texture measuring techniques used worldwide and
in New Zealand, including the previously developed flushing model for New
Zealand pavements; and
 Data analysis techniques that are used in pavement defect modelling.

Chapter Three – Examining the volumetric behaviour of chip seal materials

Chapter Three focuses on the volumetric aspects of chip seal materials and how they
relate to the occurrence of flushing. The outcomes presented in this chapter were
achieved by conducting laboratory testing on chip seal samples obtained from in-
service pavements. The main objectives of this chapter were to determine how the
flushing behaviour of chip seal structures was affected by the characteristics of the
aggregate grading profile, bitumen content and the effects of loading. This chapter
describes the laboratory test methods that were conducted as part of this research,
which included wheel tracking, binder extraction tests and aggregate grading tests,
and the results of these laboratory tests are presented.

Chapter Four – Use of image analysis to quantify chip seal material interactions
Chapter Four extends from the laboratory results presented in Chapter Three and aims
to determine the micromechanical changes that occur within chip seal structures and
how these changes impact on flushing occurrence. This chapter describes the use of
computed tomography scanning and image analysis for chip seal structure analysis.

1-9
The chip seal samples that were tested in Chapter Three were analysed using
computed tomography scanning to examine how the air voids volume and distribution
were altered and how the reduction in air voids affected the flushing patterns observed
on the samples. Results of this chapter included plots of air void volumes within the
samples and images of three-dimensional reconstructions of the air void structures
within the samples.

Chapter Five – Development of a flushing prediction model

Chapter Five presents the development of a data-driven model for predicting the
initiation and progression of flushing on chip seal pavements. The model that is
presented in this chapter was developed using pavement condition data from the New
Zealand long-term pavement performance database. The chapter presents
investigations into the correlations between flushing and various factors that influence
its occurrence, such as surfacing and pavement characteristics, traffic characteristics
and the contribution of other pavement defects. The statistical data analysis
techniques that were used in this chapter were factor analysis, correlations analysis
and regression.

Chapter Six – Summary and recommendations of research findings

Chapter Six combines the findings from Chapters Three, Four and Five to
demonstrate how these findings can be used for better assessment of flushed chip seal
sites. This chapter presents an assessment guideline for efficiently identifying flushed
sites in a pavement network and sets out steps to take in order to conduct in-depth
analysis on identified flushed sites. A discussion of the implications of the
documented research and recommendations for further research are also presented.

Chapter Seven – Conclusions

Chapter Seven presents the main conclusions that were reached from the documented
research.

1-10
1.5 Data Sources used for Modelling Flushing
The data that was analysed to understand the flushing mechanism and to develop the
prediction model was obtained from two sources. Firstly, data was gathered from
laboratory testing of pavement samples obtained from in-service flushed chip seal
pavements in New Zealand. Secondly, pavement condition data was collected from
the New Zealand Long-Term Pavement Performance (LTPP) database. The following
sections provide an overview of these two data sources. Detailed information relating
to the methodology that was used in the collection of the data can be found within the
relevant chapters in the thesis.

1.5.1 Laboratory-Based Data Obtained from Field Samples

In order to gain a fundamental understanding of the mechanisms that cause flushing it
was necessary to perform laboratory testing that simulated flushing as it happens in
the field. Laboratory testing was performed on pavement samples obtained from in-
service chip seal pavements from four regions in New Zealand, namely, Auckland,
Waikato, Christchurch and Dunedin. The field samples were subjected to mechanical
testing to determine the volumetric properties of the chip seal layers, such as volume
of air voids, deformation behaviour, bitumen content and aggregate grading profiles.
In addition to the mechanical tests, the samples were analysed using non-destructive
methods such as x-ray scanning to quantify the volumetric changes. By performing
these tests, a data set was developed which was used to investigate how flushing was
affected by the changes in the volumetric properties of chip seal surfaces.

1.5.2 The Long Term Pavement Performance Programme

Pavement deterioration models make up an essential part of a pavement management
system and are used to predict the future performance, maintenance needs and
expenditure of road pavements. In the late 1990s, New Zealand adopted the software
application dTIMS (NZdTIMS) for the management of pavements. As a starting
point, the NZdTIMS system used the World Bank HDM-III and HDM-4 pavement
condition deterioration models to forecast the long-term maintenance needs of its
pavements. These models needed to be calibrated to suit the local conditions, which
1-11
led to the development of new pavement deterioration models and the establishment
of the LTPP programme. The New Zealand LTPP programme began in year 2000,
and initially included 63 LTPP sections on the State Highway network that were
established by Transit New Zealand. This sample set was extended further in 2003 by
Land Transport New Zealand and local authorities to include 82 LTPP sections on
both the urban and rural road network (Transit New Zealand, 2007). The sections
included in the LTPP programme have been selected to represent a wide range of
environmental and soil conditions, traffic volumes, pavement ages and strengths, and
construction types. The data collected for the LTPP programme includes variables
that indicate the structural condition and performance of the pavement. These
variables include visual recordings of surface defects such as cracking, potholes and
surface damage, measurements of wheel track rutting, road roughness and surface
texture, and records of maintenance conducted on the sections, such as crack sealing
and pothole patching (Henning et al., 2004).

The data collected as part of the LTPP programme have been primarily used in the
calibration of the dTIMS and HDM deterioration models to suit New Zealand
conditions. Recently, research has been conducted to develop completely new model
forms from this data. Such models include the crack initiation model, the rut
progression model, the asphalt ravelling model and the texture model (Henning, 2008,
Henning and Roux, 2008, Parkman and Rainsford, 2005). The data have also been
used in other research projects carried out to determine strength variations over time
on pavements and to develop new test methods to measure pavement strength, as well
as to calibrate equipment used in pavement surveys.

1-12
CHAPTER 2 - LITERATURE
REVIEW

2.1 General Pavement and Surfacing Technology

Pavements are categorised into two groups: flexible pavements and rigid pavements.
Flexible pavements are typically surfaced using chip seals, slurry seals or thin
asphaltic concrete, and these pavements are relatively low cost and commonly used in
New Zealand. In contrast, rigid pavements are surfaced using high strength concrete
or layers of dense asphaltic concrete and these pavements in contrast to flexible
pavements are high cost and used in areas that require high strength and durability
(Transit New Zealand et al., 2005). The research discussed in this thesis is primarily
focused on the performance of flexible pavements with chip seal surfaces, and
therefore only chip seal surfaces will be discussed hereafter.

2.1.1 Road Classification and Surfacing Options

The type of chip seal surface used for a particular pavement depends on a number of
factors such as the function of the road, expected traffic volume, availability of
construction materials, environmental conditions, and the expected design life of the
pavement. A number of different types of chip seal surfaces exist, such as single coat
seals, multicoat seals or specialist surfacings (Austroads, 2004; Transit New Zealand
et al., 2005). Single coat seals are the most common type of chip seal and they consist
of a single layer of sprayed binder onto which a single layer of aggregates is spread
and rolled. Single coat seals are low cost and suitable for both low and high traffic
volumes, and they are widely used in New Zealand. Multicoat seals consist of more
than one application of binder and/or aggregates applied during a single sealing
session. These seals differ from multiple seal coats that result from resealing over a
number of years. Types of multicoat seals commonly used in practice include two-
2-13
coat seals, racked-in seals, void fill seals, sandwich seals, wet lock seals, dry lock
seals, and Stress Absorbing Membrane (SAM) seals (Read and Whiteoak, 2003). The
type of multicoat seal used depends on the specific purpose of a pavement, for
example in high-stress areas such as intersections and haul roads, where
waterproofing or snow resistance is required, to minimise traffic noise or to restore
texture on flushed surfaces. Specialist surfacings are mainly used to meet specific
demands of a pavement. Some of the commonly used specialist surfacings include fog
coats, geotextile seals, high skid-resistant surfaces and coloured surfacing for bus and
cycle lanes.

2.1.2 Construction
The construction processes used for chip seals contribute significantly to the
performance of the seals. Accurately designed and constructed seals perform
significantly better and can achieve the full design life in contrast to seals that are not
designed well. The sequence of steps followed in chip seal construction varies
between the different seal types. Nevertheless, any surface that needs chip sealing is
required to be cleaned well prior to sealing to remove dust and other fine material.
This cleaning process prevents the development of premature distresses in the seal
from foreign material. The seal is constructed by first spraying a layer of bitumen on
the existing surface and this is followed by spreading a layer of aggregates on top of
the bitumen. The aggregate layer is rolled to allow for compaction. Depending on the
type of seal, such as in two-coat seals, the above process may be repeated on top of
the initial seal. Variations in aggregate and bitumen spread rates greatly affect the
final outcome of the seal. For example, on surfaces with low texture a low bitumen
spray rate is used to avoid excess bitumen. Conversely, a high bitumen spray rate is
used on high-textured surfaces to allow sufficient coating of the existing seal.

In addition to the construction methods used, the environmental conditions during the
construction of a chip seal surface also impact on its performance. Construction of a
chip seal during wet weather or cold weather can prevent the aggregates from
adhering to the binder properly, resulting in the aggregates being stripped off when
the chip seal is subjected to traffic loading. This stripping of aggregates often results

2-14
in chip loss on the pavement surface. Therefore, when chip seals are constructed it is
imperative that the environmental conditions are taken into account to achieve a well
performing chip seal.

2.1.3 Maintenance
The amount of maintenance required on a chip seal surface depends largely on the
quality of construction of the seal. Maintenance treatments on chip seal surfaces are
carried out in two ways (Transit New Zealand et al., 2005):
 Localised treatments which are used on small areas of the pavement, such as
high-pressure water treatment and small digouts; and
 Area-wide treatments which are used on major seal failures, such as overlays,
seal replacing and recycling.

The maintenance methods that are used for a pavement vary depending on the
perceived importance of the pavement, cost, required life of the pavement and the
type and severity of the distresses. Surface defects such as minor chip loss can be
treated with localised texture restoration while severe flushing or cracking may
require full pavement rehabilitation.

2.2 Surface and Pavement Failure Mechanisms

Failure of pavements and chip seal surfaces can occur in a number of ways. Some of
the important failure mechanisms are flushing and bleeding on the seal surface,
structural and surface deformation, cracking and chip loss. These mechanisms affect
the performance and functionality of the pavement surfaces significantly, and
determine the type of rehabilitation that is needed. The following sections outline
these failure mechanisms in detail.

2.2.1 Flushing
Flushing on a pavement occurs when the seal aggregate is covered, partially or fully,
due to the upward movement of bitumen. This occurrence forms a smooth, black
2-15
surface with low texture depth. There are many causes for flushed surfaces and some
of these include (Transit New Zealand et al., 2005):
 Embedment of the seal aggregates in soft substrate caused by traffic loading;
 Excessive binder in the seal resulting from a build-up of successive seals
layers;
 Binder rising due to water vapour; or
 Old age of the surfacing layer.

Flushing can occur in both well-designed chip seals and in seals with construction
faults. Flushed surfaces cannot always be repaired by resurfacing, and they often
require a full rehabilitation treatment. As a result, flushing of chip seal surfaces is one
of the main causes for resurfacing and maintenance on state highways (Austroads,
2008). Flushing has a direct impact on the skid resistance of a pavement, especially in
wet conditions, which in turn affects the risk of vehicle accidents. Flushed surfaces
have been found to have a lower friction coefficient, of approximately 0.11 at a speed
of 60 km/h, than similar un-flushed surfaces. As a result, the skid resistance of these
flushed surfaces were found to decrease by 0.01 units with a slip-speed increase of
20 km/h (Henderson et al., 2011).

2.2.2 Bleeding
During hot weather periods the binder in chip seals becomes soft and can be picked up
by vehicle tyres. This occurrence is known as bleeding. Bleeding usually occurs as a
secondary effect from flushing which reduces the macrotexture of a pavement surface
and makes it easier for tyres to reach into the spaces between the aggregates and pick
up the binder. The mechanism of flushing is not always required for a surface to bleed
and similarly, a flushed surface may not always bleed provided that the temperature is
sufficiently cold or the binder is sufficiently hard. The definition of the terms
“flushing”‎ and‎ “bleeding”‎ varies‎ between‎ different‎ countries,‎ for‎ example,‎ in the
United States of America (USA), flushing and bleeding are used interchangeably to
describe the mechanism of bitumen rising in chip seal surfaces, while in South Africa,
bleeding is used to describe both flushing and bleeding.

2-16
Bleeding can also occur as a result of other defects, such as in rutted pavements where
binder application during resurfacing can allow binder to accumulate in the ruts
(Gransberg and James, 2005). Once the binder has softened enough it can cause
bleeding. Binder that adheres to vehicle tyres usually gets spread on the surrounding
surface, which is known as tracking. The process of tracking coats the surfaces of the
seal aggregates with binder which then reduces the microtexture of the pavement
surface. Tracking is undesirable, especially in urban areas where microtexture
provides most of the skid resistance. Furthermore, binder that adheres to vehicle tyres
from surfaces with severe bleeding can strip-out the seal aggregates, leading to
pavement delamination (Lawson and Senadheera, 2009). Surveys of highway and
road agencies in the USA, Canada, Australia, New Zealand, United Kingdom (UK)
and South Africa have found that bleeding is one of the most widespread surface
defects on chip seal pavements (Gransberg and James, 2005). As a result, it is critical
that methods to minimise bleeding on chip seal surfaces are found.

2.2.3 Deformation
Deformation on chip seal pavements can be divided into four categories: rutting,
shoving, depressions and corrugation. These defects usually occur as a result of
deformation in the subgrade or pavement layers caused by heavy traffic, inadequate
structural strength, or environmental conditions. In the sections below two important
types of deformation are discussed: rutting and shoving.

Rutting
Rutting on chip seal surfaces can be distinguished as vertical deformations occurring
in the wheel paths. Rutting can be caused by a number of reasons including traffic
loading, permanent deformation within pavement layers or in the subgrade due to
weak structural layers from water ingress, or failure of underground pipes or services
(Transit New Zealand et al., 2005). Rutting on pavements can cause problems in
terms of both safety of motorists and integrity of the structural layers. Rutting allows
water ponding, which can be hazardous for road users by causing aquaplaning,
freezing, or impaired visibility from splashing in high speed areas. Surface rutting that

2-17
hold water can also allow water ingress into the structural layers of the pavement,
which can in turn weaken the base and subbase layers leading to further rutting. The
infiltrated water may also cause cracking of the pavement layers and increase the rate
of deterioration. Figure ‎2.1 illustrates the different types of surface deformation that
can be found on chip seal surfaces.

Figure ‎2.1 Types of pavement deformation (Transit New Zealand et al., 2005)

Shoving
Shoving is a type of deformation that occurs on the surface of pavements where the
surfacing or the layers close to the surface deform and distort. Shoving is also called
shallow shear, and in contrast to rutting which usually occurs perpendicular to the
pavement surface, shoving generally occurs parallel to the surface. Shoving on chip
seal surfaces is characterised by bulges of displaced material near depressed areas and
commonly occurs on the outside of the carriageways (Austroads, 2004). Shoving can
occur as a result of a number of factors, such as:
 Accumulation of moisture under the upper pavement layers due to poor
drainage or water ingress through cracking or rutting;
 Large shear stresses exerted on the surface layers by slow moving, turning or
heavy traffic; or
 Weakening of base course layers caused by repetitive loading.

2-18
2.2.4 Cracking
Cracking on chip seal surfaces can be divided into a number of categories (as shown
in Figure ‎2.2). These categories include block, alligator, longitudinal, transverse,
meandering, diagonal and crescent-shaped cracks. Cracking is an important pavement
performance indicator and can be a measure of the amount of deterioration within the
pavement. Cracking exposes the structural layers of the pavement, allowing water
ingress. This process can subsequently increase the rate of deterioration by allowing
secondary defects, such as rutting or potholes, to occur. The type of crack on a
pavement determines the maintenance procedures required, and it is necessary to
identify the cracks accurately in order to select the best form of treatment for the
pavement. The main types of cracks recorded in the Road Assessment and
Maintenance Management (RAMM) manual include crocodile, block, and
longitudinal and transverse cracks (Transfund New Zealand, 1997).

Figure ‎2.2 Types of cracking on bituminous surfaces (Austroads, 2006b)

2.2.5 Chip Loss

Chip loss on chip seal surfaces is caused by three mechanisms:
 Scabbing, where aggregates are lost from patches in the seal;
 Stripping, where aggregates are lost in long strips along the wheel paths; and
 Attrition, where the aggregates are worn off due to friction between the seal
and vehicle tyres (Transit New Zealand et al., 2005).

2-19
Chip loss occurs either if there is not enough binder in the seal or if the aggregates are
not adhered to the binder due to moisture or dirt on the aggregate surfaces. Chip loss
is common in hot climates where dry seals are used in order to reduce the occurrence
of bleeding or flushing. Premature chip loss can be a result of design faults such as
incorrect seal type, construction faults, or cold or wet weather. When chip loss occurs,
it is required that the surface be repaired prior to applying a reseal, and failure to
repair the surface may result in either further chip loss or, flushing or roughness due
to reduced macrotexture.

2.3 World-Wide Surfacing Issues

An international survey carried out by Gransberg and James (2005) looked at the most
common surfacing issues observed on chip seals in New Zealand, Australia, UK,
South Africa, Canada and the USA. The study was based on chip sealing practices of
highway authorities of these countries. The results of this survey showed that overall
the most common surfacing issue identified was bleeding (including flushing),
followed by ravelling (Figure ‎2.3).

Figure ‎2.3 Most common distress modes on chip seal surfaces (adapted from
Gransberg and James (2005))

2-20
A closer investigation of the observations by Gransberg and James (2005) showed
that for chip seal pavements in the USA and Canada the most common short-term
surfacing problem identified was loss of aggregate, and flushing was identified as the
most common long-term surfacing problem. This is different from the problems seen
by countries such as Australia and New Zealand, where premature flushing was the
main surfacing problem overall. These observations may be related to the different
environmental conditions in those countries. The freeze/thaw action of the colder
climate in Canada and the USA is likely to contribute significantly to the loss of
aggregates, while the higher temperatures in Australia and in some regions of New
Zealand are the cause of the flushing and bleeding seen in those countries.

2.4 Currently Available Evidence on Flushing

From surveys conducted on chip seal pavements in New Zealand it has been found
that the most common indications for resealing result from problems relating to
texture, cracking and Sideways Force Coefficient Routine Investigation Machine
(SCRIM)-measured skid resistance. Figure ‎2.4 shows these surfacing issues in
graphical form for resealing carried out in the 2002-2003 season.

Figure ‎2.4 Reasons for resealing of chip seals in 2002-2003 (adapted from Ball
et al. (2004))
2-21
In addition to the study by Ball et al. (2004), Oliver (1999) also performed a survey to
determine the main issues that lead to resurfacing on chip seal pavements. Oliver
(1999) surveyed Transit New Zealand, Australian State Roading Authorities (SRA),
and a number of Local Governments (LG) in Australia and New Zealand as part of the
study and the survey results (presented in Table ‎2.1) showed that cracking was the
main issue leading to resurfacing on chip seal pavements, and loss of texture
(flushing) was ranked third. Oliver’s (1999) observations did not agree with the
results found by Ball et al. (2004), and the reason for this difference may be related to
the traffic volumes observed on New Zealand roads. An earlier study by Ball (1997)
found that on roads with traffic volumes lower than 2000 vehicles per day (vpd),
cracking is the main cause of resurfacing, while on roads that carry more than 2000
vpd, flushing is the main reason for resealing. The majority of the local road network
in‎ New‎ Zealand‎ carries‎ low‎ traffic‎ volumes‎ and‎ this‎ may‎ explain‎ Oliver’s‎ (1999)
findings.

Table ‎2.1 Rankings of reasons for resealing (adapted from Oliver (1999))

Reason for resealing Aust. SRAs Transit NZ Aust. LG NZ LG

Cracking 1 1 1 1
Pavement deterioration 5 4 2 2
Age 2 5 4 6
Loss of aggregate 3 6 3 3
Loss of texture 4 3 5 5
Skid resistance 6 2 6 4

Resurfacing trends in 1992-1993 show that surfacing issues observed in 2002-2003

were different from those observed a decade ago. As shown in Figure ‎2.5, the main
surfacing issue identified in the 1992-1993 pavement survey by Ball et al. (2004) was
similar to the result obtained by Oliver (1999). This survey also identified cracking as
the predominant reason for resealing, and flushing and scabbing were identified as the
next most important issues. Figure ‎2.4 and Figure ‎2.5 show that there was a
significant increase in the percentage of pavements resealed due to flushing in 2002-
2-22
2003 compared to those that were resealed ten years earlier. This difference was
found to be due to new criteria on the critical Mean Profile Depth (MPD), where
surfaces are treated for flushing when there are appearances of flushing on the surface
(Ball et al., 2004). This new resurfacing criteria is different from the earlier
resurfacing‎criteria‎of‎MPD‎≤‎0.5‎mm‎before‎treatment‎was‎implemented.‎

Figure ‎2.5 Reasons for resurfacing chip seal surfaces in 1992-1993 (adapted
from Ball et al. (2004))

All of the studies referred to above assessed the whole pavement network in New
Zealand. It is equally important to look at pavement networks from various regions of
the country to identify the variations in surface defects that are most prominent
between different regions. Low friction is the most common reason for resealing at
the top of the country in Northland and Auckland regions and at the bottom of the
country in Otago and Southland regions, while low texture is the most common
reason for resealing in the Bay of Plenty and Hawke’s Bay regions (Ball and Patrick,
2005).

From the studies described above, it can be seen that texture loss caused by flushing is
a major problem that needs to be investigated further to improve the performance of
chip seal surfaces in New Zealand. Consequently, a number of studies have already
2-23
been conducted to identify the main causes of flushing on chip seals in New Zealand.
The main objectives, methodologies, findings and recommendations of these studies
are summarised in Table ‎2.2. Although these studies were successful at identifying
some of the mechanisms and factors leading to flushing, none have developed a
model to predict the initiation and progression of flushing.

2-24
 Table ‎2.2 A summary of flushing research completed in New Zealand

Study / Research Objectives Methodology Followed Outcomes / Findings / Recommendations

BALL, G. F. A. & Effects of traffic and temperature on Flushed and unflushed seal samples A linear relationship existed between the number of
PATRICK, J. E. (1998) the occurrence of flushing were were obtained from local roads in wheel passes and texture loss. This relationship differed
Flushing Processes in tested. The study aimed to: Wellington city and state highway from field observations (Patrick and Donbavand, 1996).
Chipseals: Effects of  Identify the importance of traffic sites in Wairarapa. The unflushed The test temperature was an important factor in
Trafficking. Transfund on the rate of texture loss; samples were tracked using a determining the flushing potential of a seal sample.
New Zealand Research  Determine the effects of weighted wheel at varying Harder bitumen grades, such as 80/100 or 130/150
Report 122. different pavement materials temperatures. Texture depth changes delayed flushing and increased seal lives considerably
and environments on the over time were measured using a more than 180/200 grade bitumen. Variations in viscosity
occurrence of flushing; and laser profilometer. resulting from bitumen ageing had no significant impact
 Identify ways to minimise on the rate of texture loss.
flushing in order to extend the 
Bitumen from flushed samples were Field trials, with a range of seal and bitumen types, were
life of chip seal surfaces. tested for rheological properties. recommended to allow higher amounts of wheel tracking
on seals and identify further the factors contributing to
flushing.

BALL, G. F. A., Effects of moisture in chip seals Seal samples with varying degrees of Both mechanisms of moisture-induced flushing were
LOGAN, T. C. & were investigated in order to flushing were cored. Pulses of water, supported by tests.
PATRICK, J. E. (1999) minimise moisture-induced flushing. at pressures similar to truck tyres, The main requirement for bitumen flow through the seal
Flushing Processes in Two mechanisms of moisture- were applied to the samples and the was the availability of passages. An effective diameter of
Chipseals: Effects of induced flushing were identified. rate of water ingress was measured. only 0.1 mm is needed for flushing.

2-25
Water. Transfund New They were: Moisture ingress under static water Water ingress under traffic is likely on all seals, not
Zealand Research  Water vapour rising up from the pressures was measured for flushed limited to flushed seals or wheel paths. Water ingress was
Report 156. base layer of pavement and and unflushed seal samples. not confined to large pressures, but also occurred at
pushing bitumen to the seal significantly lower pressures, such as the pressure of car
surface; and Flow through a capillary was used to tyres.
 Moisture entering the seal model bitumen flow in the seal. Under large tyre pressures the discontinuities on the seals
through discontinuities on the Temperatures and water pressures can be enlarged, allowing further moisture ingress.
surface under wheel loads, then that are usually observed on chip Further testing is needed to identify and improve site
vaporising to create bubbles of seals were used to calculate the factors that contribute to flushing, in particular, whether
bitumen on the pavement likely quantity of bitumen. reducing the water level beneath the seal leads to
surface. decreased flushing in areas prone to flushing.

HENDERSON, R., A study was conducted to identify a A statistical study was conducted to Statistical analysis of the study concluded that successful
CENEK, P., statistic to detect flushing from two investigate the degree of correlation identification of flushing from laser measured texture
JAMIESON, N. & dimensional laser profiles of roads. of various measures of surface profiles needs to be complemented by other information
WILSON, D. J. (2011) This study aimed to quantify the texture with visual ratings of such as skid resistance or surface reflectance.
The Influence of Binder reduction in surface friction due to flushing. The Northland state The presence of flushing on a chip seal surface has a
Rise in Reducing Tire- flushing as well as to quantify the highway network was used for the detrimental impact on its level of skid resistance. The
Road Friction. NZ reduction in braking performance on statistical study. Texture profile and skid resistance of a wet flushed surface reduced by 0.1
Transport Agency flushed surfaces under dry and wet friction measurements were units at a slip-speed of 40km/h compared to a similar
Research Report 447. conditions. performed on chip seal sites. unflushed surface. The presence of extreme flushing
46pp. increased the breaking distances considerably when
compared to unflushed surfaces.
Statistical analysis is recommended to detect flushing

2-26
 using‎ the‎ latest‎ data‎ extracted‎ from‎ NZTA’s‎ RAMM‎
database in order to confirm findings of this study.

WHITEHEAD, D., The aims of this study were to Surface texture data calculated from Texture thresholds, in terms of Sensor Measured Texture
DONBAVAND, J. & develop an objective measure to the SCRIM and skid resistance data Depth (SMTD) and MPD, were recommended to identify
MITCHELL, J. (2011) identify the point when flushing measured at 113 sites within New the flushed status of a pavement. A texture threshold of
Developing an starts to affect the skid resistance of Zealand were used in the study. SMTD‎ ≤‎ 0.50‎ and‎ MPD‎ ≤‎ 1.0‎ was‎ recommended‎ to‎
Objective Measure for a pavement surface and to establish Statistical regression was conducted identify‎areas‎starting‎to‎flush‎and‎a‎threshold‎of‎SMTD‎≤‎
Flushing. 3rd texture values that could be set as a to determine the correlations 0.40‎and‎MPD‎≤‎0.70‎was‎recommended‎to‎identify‎fully‎
International Surface level of service to maintain skid between skid resistance and SCRIM- flushed areas. The authors recommended validating the
Friction Conference, resistance and safety on a highway. measured texture data. texture thresholds by applying them to New Zealand state
Safer Road Surfaces - highway network sites.
Saving Lives. Gold
Coast, Australia.
BALL, G. F. A. & A study was carried out to determine Artificial chip seals were constructed Fine solid material in a multiple chip seal layer fill a
PATRICK, J. E. (2008) the change in void volume under on trays and hand-rolled to compact. significant proportion of the void volume otherwise
Resealing Strategies to trafficking in multiple chip seals. A number of seal combinations with available for bitumen. This process can lead to flushing.
Increase Seal Life and The study aimed to produce varying grades of chips were Fines content increases over time, making it difficult to
Prevent Seal Layer guidelines for the selection of reseals constructed. Water was added to the predict the available void content in a seal.
Instability. NZ on flushed sites to minimise future samples to measure the void content. Likely sources of fines in chip seals include aggregate
Transport Agency flushing. wear under traffic, upward migration of base course
Research Report 372. Chip seal samples were cored from material, wind-blown matter, vehicular material and
The study also aimed to predict the roads with varying traffic levels and residue, residue from iced road treatments or inadequate

2-27
 maximum compaction possible on a examined for void and fines sweeping prior to chip sealing. The study recommended
seal in order to predict the likelihood contents. monitoring new chip seals to examine the development of
of flushing. Sample layers were separated to fines over time and to determine the sources of fines
determine the origins of fine observed in chip seals.
materials.

BALL, G. F. A., The causes of premature flushing Seal samples from flushed and No correlation existed between ball penetrometer tests
PATRICK, J. E. & and instability on chip seal surfaces unflushed sites were cored from Bay and risk of a seal to flush.
HERRINGTON, P. R. were investigated. The study aimed of‎ Plenty,‎ Hawke’s‎ Bay and Mid Shear tests were effective at predicting the likelihood of
(2005) Factors to determine the combination of seal Canterbury. Ball penetrometer tests premature flushing on seals.
Affecting Multiple types that lead to early flushing. were carried out at each sample site. Flushing is more likely on unstable seals that have been
Chipseal Layer Shear forces similar to truck tyre built up over time than on stable seals. Sites exhibiting
Instability. Land Improvements that can be made to pressures were applied to the seal premature flushing had seen catastrophic seal failures
Transport New Zealand the construction processes in order to samples. earlier.
Research Report 278. reduce the occurrence of surface High bitumen content alone does not cause flushing. The
instability on these chip seals were ratio of bitumen to available volume in an optimum-
also investigated. packed seal layer is the most important factor leading to
flushing.

2-28
2.5 Mechanics of Bitumen and Asphalt
As described in the sections above, a number of studies have been carried out in New
Zealand to investigate the causes of flushing of chip seal pavements. While many of
these studies have identified particular factors that lead to flushing, such as traffic
loading, moisture effects or pavement characteristics, none have investigated this
defect from a fundamental perspective, specifically, the micromechanical interactions
that occur between chip seal materials that initiate flushing. As mentioned earlier in
the thesis, there is limited literature available that have looked at chip seal material
interactions and much of the available literature that investigates pavement material
micromechanics focuses on asphalt concrete pavements. Although asphalt concrete
pavements are different to chip seal pavements due to the constituents included in
them and the method in which they are constructed, it is still possible to use
knowledge of asphalt concrete pavement behaviour to understand chip seal behaviour.
The following sections detail the behaviour of asphalt pavement materials and in
particular the methods that have been used to model volumetric changes in the
materials.

2.5.1 Comparison of Chip Seal Pavements and Asphalt Concrete

Pavements
A chip seal surface consists of layers of bitumen and aggregates, spread singularly or
alternatively on the base of a pavement. In contrast asphalt surfaces, for example Hot
Mix Asphalt (HMA), consist of a mixture of bitumen, aggregates and modifiers such
as filler in the form of cement, crumb rubber and polymers. One of the main
differences between these two surface types is that in chip seal surfaces the bitumen
and the aggregates are laid separately and rolled when a pavement surface is
constructed, whereas in asphalt surfaces all the constituents are mixed together prior
to application. Although the methods used to construct chip seal surfaces are different
from the methods used to construct asphalt surfaces, the constituents included in these
two surface types and their functions are similar. In chip seal surfaces the main
purpose of bitumen is to provide adhesion to the aggregate matrix and waterproof the
pavement, and the aggregates provide texture and strength to the pavement. Likewise,
in asphalt surfaces the bitumen and aggregates have the same purpose as in chip seal
2-29
surfaces. The extra additions in asphalt surfaces, namely cement, crumb rubber and
polymers, make asphalt surfaces behave in a more complicated manner than chip seal
surfaces.

One of the most common methods of treating defects of a chip seal surface is
resealing. The build-up of binder and aggregate layers resulting from multiple reseals
can cause chip seal surfaces to behave similarly to asphalt surface layers, where the
aggregates and bitumen mix together to create a structure that is similar to an un-
designed asphalt layer. Thus, examining the behaviour of asphalt surface layers,
particularly the micromechanical interactions between materials, can aid with
understanding the behaviour of constituents in chip seal surfaces.

2.5.2 Modelling of Asphalt Surfaces

The different types of materials that form an asphalt surface layer make it difficult to
understand and model the mechanistic behaviour of the layer as a whole. One way to
understand the behaviour of asphalt materials is to first understand the effects of and
the interactions between the different constituents in the layer. The structure of an
asphalt layer can be separated into three main components: aggregates, bitumen and
air voids. From these three components the aggregates make up the most volume in an
asphalt layer and therefore have the largest contribution to the behaviour of the layer.
In typical HMA surfaces the aggregates make up approximately 85% of the volume.
The component structure of an asphalt layer can be more easily represented by the
phase diagram shown in Figure ‎2.6.

When modelling the micromechanical behaviour of asphalt layers it is necessary to

firstly understand the mechanical behaviour of bitumen. A number of studies have
explored bitumen modelling and many of these studies have recognised the complex
nature of different chemical species that form bitumen (Krishnan and Rao, 2001;
Schweyer, 1973; Taherkhani and Collop, 2008; Wang et al., 2004; Buttlar et al.,
1999). Due to the varying nature of bitumen at different temperatures, the behaviour
of bitumen is modelled for a certain temperature range (Krishnan and Rajagopal,
2005). In most studies that modelled bitumen it was concluded that bitumen behaves

2-30
as a Newtonian fluid at temperature ranges of approximately 65°C - 150°C and
exhibits viscoelastic behaviour at temperature ranges of 25°C - 65°C (Storm et al.,
1996; Coree, 1999). The mechanical behaviour of bitumen varies depending on the
polymers and modifiers added to it, and these additives are intended to make bitumen
more stable with varying temperatures.

Figure ‎2.6 Material components and volumes in an asphalt pavement layer

(adapted from Krishnan and Rao (2000))

One of the methods of modelling the mechanical behaviour of asphalt concrete is to

use the theory of mixture, which is a method of modelling the behaviour of fluid
mixtures and granular flow properties based on the properties of the mixture
constituents (Bowen, 1976; Bear, 1972). Mixture theory has also been used in
modelling the properties of asphalt concrete, particularly to model the point at which
bleeding begins (Wang et al., 2004; Krishnan and Rao, 2001). In‎Krishnan‎and‎Rao’s‎
(2001) study asphalt concrete was modelled as a three-constituent mechanical system
consisting of aggregates, bitumen and air voids, and it was found that a reduction in
the air voids content in the structure was the trigger for bleeding. The parameter Voids
Filled with Asphalt (VFA) was used to identify the point of bleeding initiation. The

2-31
percentage of VFA can be expressed as shown in Equation ‎2.1 using the notations in
Figure ‎2.6. When VFA reaches 100% there are no more air voids present in the
asphalt mixture and at this point bleeding is expected to occur.

( )

Equation ‎2.1

Where, Vma is the volume of voids in mineral aggregate; and

Va is volume of air voids.

2.6 Interactions between Flushing, Shoving and Rutting

Most failures seen on pavement surfaces arise as a result of a number of defects
occurring simultaneously, and identifying the interactions between certain defects is
imperative to developing accurate and functional pavement deterioration models. In
the case of pavement failures caused by flushing, the effects of rutting and shoving are
particularly important. Shoving often occurs due to the shear forces exerted on the
pavement surface by turning movements of vehicles and these movements can cause
aggregates to move and pack together more closely. This process reduces the volume
of air voids present in the seal that is usually filled by bitumen. When the volume of
air voids is reduced, the bitumen compresses and exudes to the pavement surface,
resulting in the surface being flushed. Shear testing performed on flushed pavement
sites have shown that the rate of compaction was greater on the area outside the wheel
path compared to the wheel path itself (Ball et al., 2005). Conversely, on unflushed
pavement sites no difference was observed in the compaction rates between the wheel
path and the area outside of the wheel path. The observations made by Ball et al.
(2005) were likely affected by the way in which the aggregates in the chip seal
structures were arranged, where the aggregate packing on areas with premature
flushing was likely to be not at its optimum, allowing the seal to shear easily during
testing. These seals will likely flush quickly under traffic loading, and evidence of
significant shearing of a seal can be an indication of the likelihood of premature
flushing.

2-32
Rutting on the pavement can also be a major factor leading to high bitumen-to-stone
ratios on chip seals. During pavement resurfacings bitumen is able to accumulate in
the ruts, which then causes the high bitumen-stone ratios leading to premature
flushing on pavements. Rutting can be one of the contributing factors to the process of
moisture ingress into the seal layer as described earlier in the literature review. As
water is able to accumulate easily in surface ruts, there can be a higher chance of
water being forced into the seal through surface cracks under traffic. Once the
moisture begins to vaporise it gives rise to flushing, similar to the processes identified
by Ball et al. (1999).

2.7 Pavement Defect Modelling

2.7.1 The World Bank HDM-III and HDM-4 Pavement
Deterioration Models
The‎ World‎ Bank’s‎ Highway Design and Maintenance Standards Model (HDM-III)
was released in 1987, and has been used throughout the world to assess the technical
and economical outcomes of road investments and to develop pavement managements
systems. HDM-III contains performance models for pavements which are used by
road controlling authorities to assess the effects of traffic and maintenance activities
on a pavement over time, as well as the effects of pavement characteristics on the road
user costs (vehicle operating costs, VOC). When the full extent of the costs of road
projects are identified, consisting of construction, maintenance and operational costs,
the economic viability of each project can be assessed, allowing for comparisons
between different projects. The HDM-III model was based on four primary research
studies that were carried out in Kenya, India, the Caribbean and Brazil (Bennett,
1996). Long-term pavement performance studies, controlled experiments and road
user surveys were carried out to develop the fundamental relationships present in the
model and these relationships were verified using data collected from a range of
countries worldwide. Although many of the relationships and road deterioration
models included in the HDM-III model are relevant to pavement performance even
today, there was a need to incorporate additional models to represent a wider range of
pavements and traffic conditions. In 1993 an international collaborative study was
conducted involving research teams based in Chile, Malaysia, Japan, Sweden and the
2-33
UK (Bennett, 1996). This study, named the International Study of Highway
Development and Management (ISOHDM), extended the scope of the existing HDM-
III model and included a wider range of models (Kerali, 2000) to account for:
 Traffic congestion effects;
 Cold climate effects;
 Wider range of pavement surfaces and structures;
 Road safety; and
 Environmental effects, including energy consumption, vehicle emission and
noise.

The result, released in 2000, was the new Highway Development and Management
Tool (HDM-4).

2.7.2 Pavement Deterioration Modelling in HDM-4

Technical analysis of pavements is carried out using four models within HDM-4
(Morosiuk and Kerali, 2001). These models are:
 Road Deterioration (RD) model, which predicts the deterioration of pavements
based on three surface types, namely bituminous surfaces, concrete surfaces
and unsealed surfaces;
 Works Effect (WE) model, which models the effects of road maintenance on
the condition of pavements and the resulting costs;
 Road User Effects (RUE) model, which predicts the costs of vehicle operation,
travel times and accidents; and
 Social and Environmental Effects (SEE) model, which models the effects of
vehicle emissions and energy consumption.

The deterioration of pavements modelled within the RD model is divided into three
classes depending on the type of surface, and the relationships for modelling different
types of defects are included for each type of pavement. Table ‎2.3 shows the types of
defects modelled by HDM-4 for each surface type.

2-34
Table ‎2.3 Pavement defects modelled in HDM-4 (Morosiuk and Kerali, 2001)

Bituminous Concrete Unsealed

Drainage Cracking Gravel loss

Cracking Joint spalling Roughness
Ravelling Faulting
Potholing Failures

Edge break Serviceability loss

Rutting Roughness

Roughness
Texture depth
Skid resistance

The successful use of the deterioration models provided in HDM-4 depends on two
factors (Morosiuk and Kerali, 2001):
 How well the data used in the models represent the current condition of the
pavements and influencing factors, such as climatic and traffic conditions; and
 How well the predicted behaviour represents the observed behaviour of the
pavement.

Application of the deterioration models therefore needs to include identification of the

requirements and quality of the input data that is appropriate for the required
reliability of the predictions. Adjustments must also be made to the model parameters
to represent the observed behaviour of the pavements that are modelled, taking into
account the local conditions observed on the pavement over time. In cases where the
predicted behaviour does not correlate with the observed behaviour, completely new
models need to be developed to suit the area under consideration, as is the case with
surface defects such as flushing or bleeding. At present, the only model in HDM that
comes close to modelling flushing is a texture model (Morosiuk and Kerali, 2001).
The current texture model has many shortfalls when it is used to model flushing, and
these are explained in the subsequent sections of this chapter. As flushing is a

2-35
significant defect observed on chip seal pavements in New Zealand, there is a need to
undertake full development of this model.

2.8 Pavement Defect Prediction Models Developed for New

Zealand Pavements
Recent research carried out to model pavement defects in New Zealand has developed
two deterioration models. These models are the crack initiation model and the rut
progression model. The following sections outline the main features and functions of
these two models.

2.8.1 Crack Initiation Model

A crack initiation model was developed by Henning (2008) for New Zealand
pavements using network survey data from the LTPP programme and cracking data
from the RAMM database. The development of this new model format was the result
of detailed analyses undertaken to calibrate the existing cracking model in the HDM-4
model format to suit New Zealand pavements. The development of the crack initiation
model was undertaken in three steps. Initially, the climatic calibration coefficients of
the HDM-4 cracking model were adjusted to suit the climatic conditions of New
Zealand. Following calibration of climatic coefficients, analysis was undertaken to
calibrate the HDM-4 cracking model coefficients to predict cracking of chip seal
pavements in New Zealand. Crack initiation predictions of this calibrated HDM-4
model were improved as a result of the use of the new model coefficients. In order to
further improve the accuracy of the calibrated model a new model format for cracking
prediction was developed. Henning (2008) developed a linear regression model which
predicted the crack initiation time on thin flexible pavements in years after the
construction of the surface. This model was further developed into a logit model to
predict the probability of a pavement section cracking in a given year depending on its
existing condition, being either cracked or un-cracked. The resulting output from this
logit model is shown in Figure ‎2.7.

2-36
 Figure ‎2.7 Probability of cracking in a given year – output from the logit
model (Henning, 2008)

2.8.2 Rut Progression Model

A rut progression model was developed by Henning et al. (2009) for flexible,
unbound pavements. The development of this model was also based on earlier work
carried out to calibrate the existing rut progression model format of HDM-4. HDM-4
uses two separate models to model rut progression in pavements: one for the initial
densification stage, which represents the period immediately after construction, and
another for the rut progression stage where rutting progresses at a relatively constant
rate (NDLI, 1995). The HDM-4 model was used to predict the initial densification and
rut progression rates on LTPP pavements. This process revealed large variations in
predictions between the observed values and the predicted values, which led to the
development of a new model format. In the new rutting model a linear model was
used to predict the initial densification stage and the rut progression stage was
separated further into two phases, namely the stable rut progression stage and the
accelerated rutting stage. Two linear models, one for thin pavements (thickness less
than 150 mm) and one for thick pavements (thickness more than 150 mm), were
developed for the stable rut progression stage. A logistic model was used to predict
the probability of accelerated rutting occurrence on a pavement (logistic model output
shown in Figure ‎2.8). All of the new models were developed using data sourced from

2-37
the Canterbury Accelerated Pavement Indoor Testing Facility (CAPTIF) in
Christchurch, New Zealand.

Figure ‎2.8 Output from the logistic model predicting the initiation of
accelerated rut progression (Henning et al., 2009)

2.9 Current Texture Model of NZdTIMS and Previously

Developed Flushing Model
2.9.1 The Existing Texture Model of NZdTIMS
The original surface texture depth model included in the HDM format uses two
distinct stages to model texture loss. Initially, there is a rapid decrease in surface
texture, to account for texture loss due to aggregate orientation and embedment, and
this rapid decrease is followed by a slow and steady decrease in texture caused by the
effects of traffic (shown in Figure ‎2.9). Due to the complex nature of texture
deterioration, the texture model in HDM is sensitive to the input variables, and thus a
more simplified model format was developed (Transit New Zealand, 2003).

2-38
Figure ‎2.9 Texture depth model from the HDM format and simplified format
(from Henning et al. (2006))

The texture deterioration model that is presently used to predict macrotexture on chip
seal surfaces in New Zealand takes the following form (Equation ‎2.2) (Parkman and
Rainsford, 2005):

[( ) ( )]
{ ( )}

Equation ‎2.2

Where,
ΔTD is the incremental change in sand patch derived texture depth during analysis
year, in mm;
ITD is the initial texture depth at construction of surfacing, in mm;
TDa is the texture depth at the beginning of the analysis year, in mm;
ΔNELV is the Number of Equivalent Light Vehicle passes during the analysis year (one
heavy truck or heavy bus equals to 10 NELV; one light vehicle equals to 1 NELV);
a0 is the regression coefficient; and
Ktd is the calibration factor for texture depth.

There are standard default values provided for the ITD and a0 depending on the traffic
loading and aggregate grade of the surface (Parkman and Rainsford, 2005). However,
2-39
these default coefficients were found to provide unrealistic predictions when applied
to New Zealand pavements. Due to the vast range of values available and the complex
form of the model, macrotexture is not modelled widely on the New Zealand state
highway network. Investigations were carried out to look at the different variations of
macrotexture models that are used in New Zealand, and three methods of model
calibration were proposed for this task. These methods included (Transit New
Zealand, 2003):
 Method 1 – Detailed section calibration of model coefficients using the current
HDM-4 incremental macrotexture model;
 Method 2 – National section calibration of model coefficients using the
current HDM-4 incremental macrotexture model; and
 Method 3 – Regional calibration of macrotexture performance using a
simplified surface age macrotexture model, based on absolute HDM-4
macrotexture model.

The accuracy and suitability of the calibrated models were tested by comparing the
predicted macrotexture values against field observations made through network
surveys. Although the new model forms showed good correlations with the measured
macrotexture values during the formulation stage, the models were found to have poor
predictive ability when used for performance modelling on the state highway network.
The main issues identified for this shortfall included loss of macrotexture due to
premature flushing resulting from a build-up of surface layers, bleeding of wheel
paths during summer months affecting the macrotexture of adjacent road sections and
over estimation of the predicted macrotexture values resulting in unnecessary early
interventions. Surface flushing was identified as a potential surface defect to be
modelled in order to answer surface problems related to macrotexture (Parkman and
Rainsford, 2005). Parkman and Rainsford (2005) recognised that the output of the
existing macrotexture model needed to be improved to provide a more useful
prediction. The existing macrotexture model predicts the average MPD, although a
better form of output that was suggested was the variation of macrotexture
measurements. Therefore, there is a need to develop a model format which is more
applicable to the decision making process involved in pavement surfacing.

2-40
Building on this earlier model calibration carried out by Transit New Zealand (2003),
Henning et al. (2006) investigated another form of the texture prediction model,
which was a statistical model. For this study network data was obtained for MPD and
NELV for six regions and seven surfacing types. Stepwise regression was carried out
on the MDP data and a model of the following form was chosen (Equation ‎2.3)
(Henning et al., 2006):

( )
Equation ‎2.3

Where, a0 and a1 are chip size specific coefficients

The predictions of the MPD values were then compared with calibration data obtained
from LTPP surveys. The model was found to predict MPDs relatively close to the
observed values (shown in Figure ‎2.10), however one of the shortfalls identified with
this statistical model was its inability to predict the flushing phase of texture
deterioration. As a result, it was recommended that further investigations are carried
out to develop a model form that can predict the flushing phase (Henning et al., 2006).

Figure ‎2.10 Plots of the fitted model (based on the original data) compared to
the calibration data (Henning et al., 2006)

2-41
2.9.2 Previously Developed Flushing Model
Based on previous work by Parkman and Rainsford (2005), Henning et al. (2006), and
Furlong et al. (2005), modelling of the flushing mechanism on New Zealand
pavements was undertaken by Ul-Islam (2006). This research attempted to model
flushing using data from low volume local authority roads from a number of networks
around the country. Ul-Islam (2006) used a probabilistic modelling approach to
develop a model to predict the initiation of flushing at the selected sites. The
progression of flushing of the same sites was modelled using a regression modelling
approach and this model is shown in Equation ‎2.4.

( ( ) ( ) ( ) ( )√

Equation ‎2.4

Where, Age is the surfacing age in years;

Heavy is the number of HCVs; and
TSD is the Total Surface Depth (in mm).

The goodness-of-fit of this regression model was found to be satisfactory, and the
study concluded that this model can be utilised for predicting the progression of
flushing of chip seals for short to medium term. However, the probabilistic model was
found to have poor predictive power. The usefulness of the regression model was
dependent on the accuracy of the probabilistic model, and due to the inaccuracies of
the probabilistic model, the regression model was also unreliable. The only
explanatory variables used in the development of this flushing model included the
surfacing age, number of heavy commercial vehicles, depth of pavement surface and
the binder application rate. The use of further key variables, such as SNP, binder and
aggregate properties and the construction quality indicator, was recommended in
order to achieve a more robust model form. As a result, the data collected for the
current research needs to include important variables mentioned above in order to
develop a reliable model to forecast the long-term behaviour of flushing. Furthermore,
in the above presented model the effects of climate on flushing initiation and
2-42
progression was only partly explored through the use of data from the North and
South Islands of New Zealand even though there are large variations in climate, such
as rainfall volumes, pavement surface temperatures and snow conditions, present even
within the islands. A better approach for accurate modelling of flushing would be to
identify regions within the country with different climates to investigate the effects of
climate on flushing. Due to the complex nature of flushing and the uncertainties
associated with the progression of this defect, a different modelling approach needs to
be used in order to account for the varying materials, construction practices and
environments involved.

2.9.3 Other Methods of Assessing Flushing

A study in the USA, conducted by Weissmann and Martino (2009), looked at
developing a method to evaluate the performance of chip seal surfaces (sealcoats)
affected by flushing and ravelling. In order to avoid the subjectiveness associated with
measurements of surface distresses such as flushing or bleeding, Weissmann and
Martino’s‎(2009) approach used the macrotexture depth to provide a failure criterion
for the seal. The study used a Circular Track Meter (CTM) (Figure ‎2.11a) and an
Outflow Meter (OFM) (Figure ‎2.11b) to predict the macrotexture in terms of MPD
(illustrated in Figure ‎2.12) of the chip seal surface. The CTM is a device that uses a
laser-displacement sensor to measure the macrotexture at a location on the pavement
surface, while the OFM uses water to measure the drainage capability of the chip seal
surface by measuring the time it takes for a known amount of water to drain through
the voids on the pavement surface. The results obtained from the OFM can be used to
indicate the aquaplaning potential of a surface by mimicking the rate at which water is
displaced under a vehicle tyre. The faster the water drains from the OFM the higher
the macrotexture on the surface.

2-43
 (a) Circular Track Meter (b) Outflow Meter

Figure ‎2.11 Pavement macrotexture measuring devices; Circular Track Meter

and Outflow Meter (Martino, 2007)

Figure ‎2.12 Illustration of MPD of a pavement surface (Flintsch et al., 2003)

The measurements of MPD obtained from the circular track meter and the outflow
meter showed good correlations, where the profile depth decreased as the outflow
time increased. This behaviour was expected as it would take a long time to displace
water under a tyre on a surface with low texture. From these results a threshold for
failure was found. An MPD of 0.46 mm or an equivalent outflow time of 14.5 seconds
or higher was concluded to be the failure criteria for the surface when flushed. The
relationship between the MPD and outflow time are shown in Equation ‎2.5.

2-44
 ( )

Equation ‎2.5

Where, MPD was in millimetres; and

Outflow was in seconds.

Weissmann‎ and‎ Martino’s‎ (2009) study contained 558 measurements of MPD and
outflow times, with 71 locations being tested on both directions of traffic. Due to the
large number of measurements included the study is likely to be representative of
most chip seal surfaces that form the pavement network. It should be noted that the
above study did not include variables such as traffic growth rates, vehicle
classifications and axle loads due to them being unavailable, and as a result, the above
relationship needed to be verified under different traffic conditions. The aim of
Weissmann‎ and‎ Martino’s‎ (2009) study was to develop an objective technique to
assess the performance of chip seal surfaces when exhibiting flushing and bleeding,
and this aim was achieved by developing failure criteria for the MPD of chip seal
surfaces. This study however did not develop a model to predict the initiation or the
quantity of flushing. Furthermore, due to the dependence of the MPD failure criteria
on the results of the outflow meter the usefulness of these failure criteria would
depend on the availability of such equipment to conduct seal quality assessment.
Similar texture measuring techniques can be used on chip seal pavements in New
Zealand to supplement mathematical modelling approaches needed to develop a
flushing model.

2.10 Data Analysis and Modelling Approaches Used for

Pavement Deterioration Modelling
Accurate modelling of pavement performance is essential for the successful operation
of pavement management systems (PMS), and there is an ever-increasing need to find
modelling techniques that can efficiently achieve this task. There are many types of
modelling approaches that are being used for the task of performance modelling and
the type of model used determines the accuracy and practicality of the predictions.
2-45
The main categories of modelling approaches commonly used in pavement
performance predictions include deterministic models, probabilistic models and data
mining applications such as artificial neural networks. The following sections describe
these model types in detail.

2.10.1 Deterministic Models

The deterministic approach to pavement performance modelling uses statistical
relationships between the independent and dependant variables to predict the future
condition of a pavement. Deterministic models can be divided into a number of sub-
models based on their derivation, and three of the common sub-models are empirical
models, mechanistic models and mechanistic-empirical models.

Empirical Models
Empirical models are one of the most common forms of deterministic models used to
predict future pavement performance, and they are developed by performing
regression analyses on observed data. Empirical models are fundamentally used when
the mechanisms that influence pavement performance are not known, although the
applicability of empirical models is limited to the range of data that is used to develop
the models (Lytton, 1987). Linear regression is the simplest form of regression carried
out in empirical models. A typical linear regression takes the following form shown in
Equation ‎2.6 (Ozbay and Laub, 2001):

Equation ‎2.6

Where, y is the dependant variable, eg, pavement condition rating;

x1, x2... are the independent variables, eg, pavement age; and
12...k are the regression parameters.

2-46
An example of a regression-based empirical deterministic model is the Road
Transport Investment Model (RTIM), which was developed to predict the
performance of pavements in developing countries over one life cycle. The RTIM
model was based on data collected from sealed bituminous roads consisting of
crushed rock and cement-stabilised bases, and this model resulted in a number of
equations being formed, such as for roughness prediction, the amount of cracking and
patch treatment (Austroads, 2009).

Mechanistic Models
Mechanistic models use the analysis of pavement response parameters, such as stress,
strain and deflection, to predict pavement performance. These models typically
require data in the form of large amounts of laboratory testing or accurate field
measurements in order to model the pavement behaviour mechanistically.
Furthermore, most factors that influence the performance of pavements are usually
difficult to model using mechanistic methods. As a result, the use of mechanistic
models in pavement performance modelling is limited.

Mechanistic-Empirical Models
Mechanistic-Empirical (M-E) models use a combination of both mechanistic and
empirical models to calibrate theoretical knowledge using data from field
observations. Mechanistic-empirical models are more flexible than empirical models
or mechanistic models as they are able to incorporate interactions between different
types of pavement distress, such as the effects of rutting on the progression of
cracking. As a result, the outputs from M-E models provide better accuracy and
reliability compared to empirical or mechanistic models alone. Furthermore, provided
that they are developed correctly, M-E models can be used to model the performance
of pavements even with limited data availability.

2-47
2.10.2 Probabilistic Models
The probabilistic approach to pavement performance modelling uses a probability that
is assigned to the occurrence of the dependant variable (the future condition of the
pavement). This recognises the variability of the independent variables and the errors
commonly observed in deterministic relationships with the dependant variable
(Austroads, 2009). The output of a probabilistic model is a probability distribution of
the condition of the dependent variable at a point in time. The probability distribution
is expressed as a vector based on the classification of the dependent variable, such as
a descriptive category of the condition or a numerical value, and the vectors are
defined by the probability of the dependant variable falling into a certain
classification. The probability distribution of the current condition of the dependant
variable gives the starting vector (Weninger-Vycudil et al., 2008).

The most common types of probabilistic modelling approaches include the Markov,
semi-Markov and Continuous probabilistic approaches (Barbu and Limnios, 2008;
Ross, 1999). The application of Markov and semi-Markov probabilistic approaches
uses probability transition matrices (PTM) to define the transition of the pavement
condition from one state to the next. The derivations of these PTMs are achieved in
two ways. In PMSs that include a sizable database, the observed data is used to derive
the PTM. When there is limited data available, particularly at the initial stages of a
PMS, expert knowledge is utilised to obtain a stationary transition matrix. Each
maintenance and rehabilitation treatment requires a separate PTM to define the
pavement condition transition. The appropriateness of using PTMs for the prediction
of long-term pavement behaviour is debatable (Symons, 1985) as most PTMs are
derived using past pavement behaviour which can affect the accuracy of the long-term
predictions. As a result, PTMs can only be used with confidence for the range of
observation of the variables. Beyond this range, the PTMs need to be reviewed and
updated using observed pavement performance.

The continuous probabilistic modelling approach uses a continuous failure

distribution curve to forecast the likelihood of pavement failure. Some of the
examples of continuous modelling approaches include Bayesian models, which use a
combination of expert knowledge and observed data for predicting failure (Berger,

2-48
1985; Bernardo and Smith, 1994; Costello et al., 2012; Ortiz-Garcia et al., 2006). In
Bayesian models a probability distribution is assigned for a hypothesis and as
additional evidence is acquired, the prior probability distribution is refined. Another
example of continuous probabilistic models is logit models, which have been used for
predicting crack initiation and accelerated rutting on New Zealand pavements
(Henning et al., 2009).

2.10.3 Data Mining Approaches

Data mining is a process used to discover useful information from vast amounts of
data. Data mining techniques use software to recognise relationships and patterns in
the data which are then used to predict an outcome. By using an automatic pattern
recognition process, it is possible to find relationships that are otherwise difficult to
identify from simple regression. Data mining is an important component of the
process of Knowledge Discovery in Databases (KDD). The KDD process consists of a
number of steps that are used to convert data from its raw form to a useful output.
Figure ‎2.13 shows the steps involved in the KDD process.

Data Postprocessing
Preprocessing •Filtering Patterns
Input •Feature Selection •Visualisation
•Dimentionality Data Mining •Pattern Information
Data
Reduction Interpretation
•Normalisation
•Data Subsetting

Figure ‎2.13 The process of knowledge discovery in databases (adopted from

Tan et al. (2005))

Data mining has been successfully used in businesses where retailers use sales data
collected at checkout counters to predict customer shopping needs and patterns. In a
similar way, data mining is being used in the fields of science and engineering to
better‎ analyse‎ and‎ understand‎ the‎ Earth’s‎ climatic‎ data‎ (Tan et al., 2005). Several
studies have utilised data mining processes to predict the performance of pavements

2-49
with accurate results (Byrne et al., 2005; Yang et al., 2003). Of the different types of
data mining processes available, one process that is relevant to the present research is
the process of Neural Networks (NN). The following section details the uses of neural
networks.

Neural Networks
Neural networks, or Artificial Neural Networks (ANN), consist of a network of
interconnected‎ “neurons”‎ that‎ are‎ organised‎ into‎ layers‎ for‎ processing‎ information.‎
Each neuron represents a single cell and networks of these neurons are capable of
mimicking the learning process of the human brain. The neurons are connected to
other neurons using direct links, each of which has an assigned weight. Each input to
a neuron is multiplied by the weight on the link which is used to find the total input
into a neuron. By adjusting the weights on each link, ANNs can be trained to convert
inputs into outputs. The training process gives ANNs the capability to make educated
guesses by utilising a collection of examples of the desired conversion, eliminating
the need to have vast amounts of data.

The use of ANNs in pavement performance studies has ranged from predicting
condition ratings, modelling roughness progression, creep compliance of asphalt
pavements, and the performance of thick asphalt pavements (Owusu-Ababio, 1998).
Artificial neural networks were used by Yang et al. (2003) to predict pavement
surface conditions of flexible and rigid pavements by developing ANN models for
crack, ride and rut indices. Comparisons of the R2 value and the Root Mean Square
Error (RMSE) of the ANN model to a similar regression model showed that for a
2-year forecast the ANN model yielded a higher R 2 value and a lower RMSE than the
regression model. The study concluded that ANN models were capable of achieving
more accurate and effective predictions than were regression models, particularly for
long forecasting periods (such as 5-year forecasts) and high rates of deterioration.

2-50
2.10.4 Statistical Data Analysis Methods used in the Study
Factor Analysis
Factor analysis is a data reduction technique that is commonly used to summarise
large amounts of data in order to allow for easy identification of relationships between
a set of variables (Pallant, 2007; Hajek and Haas, 1987). Factor analysis uses
correlation matrices to display correlation coefficients between the variables included
in the analysis. Outputs from factor analysis are often used to identify correlations
between variables in a data set that merit detailed analysis, or to identify
multicollinearity between independent variables.

Correlation Analysis
Correlation analysis is a data analysis method that is used for detailed investigation of
the nature of correlations between variables (Smith and Diefenderfer, 2008; De Farias
and De Souza, 2009; Pallant, 2007). The output of a correlation analysis is the
correlation coefficient, r. The correlation coefficient measures the strength of the
linear relationship between two continuous variables, in other words the extent to
which the changes in one variable affect the changes in another variable. In addition
to the strength of the relationship between two variables, correlation analysis also
indicates whether the relationship is positive or negative. The correlation coefficient
takes a value between -1 and 1, where -1 indicates a perfect negative linear correlation
between two variables and 1 indicates a perfect positive linear correlation. A
correlation coefficient of 0 indicates that no linear relationship exists between the two
variables under observation. Performing correlation analyses allows identification of
the independent variables that have the strongest effect on the dependant variable in a
data set that are suitable for use in regression analyses.

In addition to basic correlation analysis, partial correlations are also used to assess
correlations between variables in a data set. Partial correlation analysis is a specific
type of correlation analysis which is used to measure the correlation between two
variables while controlling the effects of other variables (Raghuraj Rao and
Lakshminarayanan, 2007). Partial correlations are useful when investigating the inter-
relationship between independent variables. When investigating discrete variables,

2-51
Analysis of Variance (ANOVA) is used to compare the differences in means of a
variable between different groups (Fuentes et al., 2010).

Linear Regression
Linear regression is a statistical technique that is used to model the relationship
between a continuous dependant (target) variable and a number of independent
(predictor) variables (Kendall and Stuart, 1969; Salama et al., 2008). Linear
regression uses linear functions in its modelling, and the parameters for the
independent variables are determined from the data. In addition to the parameters of
the independent variable, the regression also produces a constant value which
accounts for the error value of the linear model. Based on the parameters of
independent variables and the constant value, the value of a dependent variable can be
predicted. The general form of the model that is produced from linear regression is as
shown in Equation ‎2.7 where x denotes the independent variable and y denotes the
dependant variable. The parameter of the independent variable is given by a, and b
denotes the constant value.

Equation ‎2.7

2.11 Summary of Literature Review

Resurfacing trends over the last 20 years have shown that flushing and texture loss
have become the most prominent causes for resurfacing on chip seal pavements in
New Zealand. These defects were investigated in a number of studies. Past research
on flushing and texture loss has led to the following conclusions:
 A linear relationship exists between texture loss and wheel passes in the
laboratory;
 Water‎vapour‎pressure‎causes‎flushing‎by‎creating‎‘bubbles’‎of‎bitumen‎on‎the‎
seal surface; and
 Fine solid material in a chip seal structure reduces the void volume available
for bitumen, thus contributing to flushing.

2-52
Although these studies were successful at identifying some of the factors leading to
flushing, none have developed a model to predict the initiation or the progression of
flushing. Moreover, currently there is limited understanding of the fundamental
behaviour of chip seal structures, and particularly how the micromechanical
interactions between materials contribute to the occurrence of flushing. The behaviour
of asphalt pavements suggests that flushing can be attributed to the decrease in air
voids within a pavement surfacing layer and this mechanism needs to be explored
further in order to correctly identify how flushing develops in chip seals. The present
research can use outcomes from the studies summarised above as a starting point in
order to identify the mechanisms that are involved in the initiation of flushing on
pavements, which can then aid the gaining of a fundamental understanding of the
behaviour of seals that are prone to flushing. Identifying the mechanisms involved in
flushing development will also aid the development of a model to efficiently and
accurately predict the initiation and progression of flushing which can be used for
better management of chip seal pavements.

The complex nature of flushing means there are many uncertainties associated with
modelling the progression of this mechanism, and the previously developed flushing
model, which has not been adopted in practice, has been unable to incorporate all
these different uncertainties. The present research needs to adopt a modelling process
that can produce a model that is flexible enough to account for the varying materials,
construction practices and environments involved, and at the same time accurately
predict flushing. The availability of data from the LTPP database will be invaluable
for the model development task. Additionally, literature showed that there was no
evidence of any previous studies that had investigated the relationship between
flushing and rutting. One of the directions that the present research can take is to
conduct laboratory testing on pavement samples to determine the relationship between
flushing and rutting in order to investigate the inter-relationships between these two
defects.

2-53
CHAPTER 3 - EXAMINING
THE VOLUMETRIC
BEHAVIOUR OF CHIP SEAL
MATERIALS

3.1 Introduction
Chip seal pavement layers are made up of three main constituents, namely the
aggregate, bitumen and air voids. The relative proportions of these constituents play a
vital role in the structural performance of a chip seal layer. The importance of material
volumetrics on the performance of bituminous pavement mixtures has been
recognised and well documented in the literature (Coree, 1999; Krishnan and Rao,
2001; Buttlar et al., 1999). In particular, surface failure mechanisms such as flushing
and bleeding are likely triggered by the volumetric behaviour of the materials.
Currently published literature that have investigated flushing have relied on empirical
results to identify the causes of flushing, and not explored the effects of constituent
volumetrics to model this defect.

There are many studies that have investigated the volumetric behaviour of asphalt
concrete pavements (Krishnan and Rao, 2001; Masad et al., 1999b). Though an
asphalt concrete pavement structure differs from a chip seal pavement structure by the
manner in which the constituents are mixed together, knowledge about the volumetric
behaviour of an asphalt concrete pavement can be applied to understand the behaviour
of a chip seal pavement. By investigating the volumetric behaviour of chip seal layer
constituents, the effects of micromechanical activities that occur within a chip seal

3-54
layer, such as air voids reduction, bitumen absorption by aggregate and interlocking
of aggregate, can be thoroughly understood.

3.2 Objectives
The purpose of this chapter is to provide an overview into the materials that are
included in a chip seal layer and to understand the micromechanical behaviour and
interactions that occur between the materials that make up a chip seal. This chapter
describes the laboratory testing that was performed on field pavement samples in
order to determine the volumetric changes that occur within chip seal layers during
loading. The main objectives of the laboratory testing were to:
 Determine the effect of loading on the flushing behaviour and deformation
behaviour of chip seal structures;
 Determine the relationship between flushing and the deformation behaviour of
chip seal structures; and
 Determine the effects of aggregate grading profile and bitumen content of a
chip seal structure on its flushing potential.

In addition to the objectives stated above, the results from the laboratory testing were
used to gain a general understanding of the air void distributions within chip seal
samples. This chapter presents the preliminary analysis of the air voids results, while
in-depth analyses of the air voids results are presented in Chapter Four.

The laboratory testing that was performed as part of this study focused on the flushing
behaviour of resurfaced chip seal pavements that contained multiple chip seal layers,
rather than addressing newly constructed chip seal pavements.

3.3 Hypothesis of Relationship between Chip Seal

Materials
A parameter that has a significant effect on the performance of a chip seal is the
volume of air voids, Va, contained within the layer (Figure ‎3.1). The presence of air

3-55
voids within a chip seal layer allows that layer to compact under compressive loading
without failure occurring. Over the lifetime of a chip seal layer the volume of air
voids decreases due to continuous compaction caused by compressive loads from
traffic (Alderson, 2008), and with this decrease in air voids the void spaces that are
available for bitumen movement within a chip seal layer also decrease. This
occurrence causes the bitumen to exude to the pavement surface, resulting in the
initiation of flushing (Krishnan and Rao, 2000). Thus, examining the trend in air voids
volume reduction can indicate when failure begins in a chip seal pavement layer.

Chip seal layer

Total volume
of air voids, Va
Base course Effective volume of
Total volume bitumen, Vbe
of bitumen, Vb Volume of bitumen
absorbed by
Subbase aggregate, Vba

Effective
Total volume of
volume of
aggregate, Vs
aggregate, Vse
Subgrade

a) Typical cross section of a chip b) Phase diagram of constituents in a chip seal layer
seal pavement

Figure ‎3.1 Constituents in a chip seal layer

Studies investigating the behaviour of asphalt concrete surface layers indicate that the
reduction in the volume of air voids is affected by the magnitude of loading and the
initial volume of air voids present within the layer (Krishnan and Rao, 2001). In
addition to the air voids, the characteristics and volume of bituminous binder can also
impact on the volumetric balance of a chip seal layer. In some situations, pavements
that consist of numerous chip seal layers can have a high proportion of binder with
respect to aggregates, and this excess binder, coupled with compressive loading, can
increase the binder movement that normally takes places within a chip seal layer,
3-56
which is a trigger for flushing. Furthermore, the characteristics of the aggregates that
form a chip seal layer play an important role in the behaviour of the binder. The
aggregates within a chip seal layer typically absorb a quantity of bitumen (denoted as
Vba in Figure ‎3.1), and this occurrence affects the amount of bitumen that is free to
migrate. Depending on the type of aggregates that makes up a chip seal layer, the
amount‎of‎“free”‎binder‎within‎a‎seal‎can‎vary,‎and‎when‎there‎is‎a‎small‎amount‎of‎
“free”‎ binder‎ in‎ the‎ chip‎ seal‎ layer,‎ the‎ possibility‎ for‎ flushing‎ occurring‎ is‎ low.‎
Another characteristic of the aggregates that can affect the volumetric balance of a
chip seal layer is the crushing and abrasion resistance of the aggregates (Dahir, 1979;
Abdul-Malak et al., 1996). Aggregates that crush or abrade easily under the wearing
action of vehicle tyres produce fine solid material (fines) which can reduce the
amount of void volume that is otherwise occupied by binder and cause binder to
exude to the surface of a chip seal layer. Furthermore, aggregates that abrade easily
often have less shear resistance and tend to slide past each other when loaded. This
occurrence reduces the probability of chip seal structures with multiple chip layers
reaching a stable state due to the lack of shear resistance between the chip layers.

Aside from the characteristics of the individual aggregates, the characteristics of the
aggregate layer as a whole is also important to the behaviour of a chip seal structure.
Aggregate embedment is an observable development that typically occurs on chip seal
surfaces. Aggregate embedment can occur when a new chip seal layer is created on
top of an existing chip seal layer (Alderson, 2008), and in this case if the binder of the
previous layer is soft, as is typical in the case of flushed surfaces, then the aggregates
from the new layer embed into the underlying binder. This occurrence reduces the
texture and void spaces that are naturally created when a chip seal layer is
constructed, which in turn impacts on flushing occurrence. Additionally, aggregate
embedment can also develop when a new chip seal surface is constructed on soft
substrate. Thus it is essential that the changes in the aggregate layers with respect to
loading are explored.

Distinguishing the specific mechanism that causes flushing at a particular chip seal
site, such as the mechanisms that were identified above, is essential for selecting the
best treatment option, and presented in the following sections and subsequent chapters
are techniques for identifying these flushing-causing mechanisms. The research that is
3-57
presented in this thesis concentrated on investigating the mechanism of air void
volume reduction that results in flushing.

3.4 Sampling of Field Pavement Cores

The testing that was performed as part of the documented research consisted of two
components, which were mechanical testing of pavement samples and image analysis
of pavement samples. The present chapter discusses the mechanical testing
component and the corresponding results and the image analysis component of the
research is discussed in Chapter Four.

Pavement samples (cores) were obtained from in-service chip seal pavement sites
from the Auckland, Waikato, Christchurch and Dunedin regions of New Zealand
(map of sampling locations is shown in Figure ‎3.2). Two sites were sampled from
each region, with a total of eight sites being sampled. The four regions that were
sampled were selected to cover a range of surfacing types and designs that are
typically used on chip seal pavements in New Zealand. The sites that were selected
for sampling consisted of four to five chip layers and displayed flushing on the wheel
paths. The sampled sites consisted of a combination of single coat seals (1 CHIP) and
two-coat seals (2 CHIP) containing aggregates of two grades, and the seals were
constructed with binder grades of either 80/100 or 180/200. Figure ‎3.3 shows the
classifications that were used for identification of the sampled sites and Table ‎3.1
shows the design matrix that was utilised for field sampling.

Twelve cores were taken from each site that was sampled and the cores were taken
from the between-wheel path area of the pavement that had flushing on the adjacent
wheel path. By taking cores from the between-wheel path area it was possible to
obtain cores that were prone to flushing yet still had no significant flushing present
and retained the original surface texture. The cores were taken at most 0.5 m apart to
ensure uniform seal characteristics. Figure ‎3.4 details the locations where pavement
sampling was performed. The cores were extracted using 200 – 250 mm diameter
drills.

3-58
 Auckland

Waikato

Christchurch

Dunedin

Figure ‎3.2 Map of sampling locations across New Zealand

Site sampling

Auckland (A) Waikato (H) Christchurch (C) Dunedin (D)

Site 1 Site 2 Site 1 Site 2 Site 1 Site 2 Site 1 Site 2

(S1) (S2) (S1) (S2) (S1) (S2) (S1) (S2)

12 cores 12 cores 12 cores 12 cores 12 cores 12 cores 12 cores 12 cores

Figure ‎3.3 Classifications of sampled sites

3-59
 Table ‎3.1 Field sampling design matrix

Age of surface
0 - 10 >10
(years)
Number of layers ≤‎4 >4 ≤‎4 >4

Binder grade 80/100 180/200 80/100 180/200 80/100 180/200 80/100 180/200
Surface AADT
type (vpd)
≤‎2500 H.S2 C.S1
1 CHIP
> 2500 D.S1

≤‎2500 C.S2 H.S1

2 CHIP
> 2500 D.S2 A.S2 A.S1

Flushed area Unflushed area

Wheel path

Between-Wheel path

Wheel path

Pavement centre line

Flushed area on wheel path

Pavement samples

Figure ‎3.4 Illustration of sampling locations at a site

3.4.1 Assumptions Used in Sampling

Some assumptions were made when conducting field sampling. It was assumed that
the structural characteristics of the between-wheel path area of the pavement, such as
chip seal thickness and age, binder grade, aggregate grade and chip seal type, were
identical to those of the wheel path. Also, it was assumed that the texture on the

3-60
between-wheel path when sampling took place was a reliable representation of the
texture on the wheel paths immediately after construction and before trafficking had
taken place. These assumptions were similar to the assumptions made in previously
documented studies that have conducted field sampling (Alderson, 2008).

It is recognised that on some field chip seal sites there may be a difference in the
binder contents and the binder application rates that are used on the wheel path area in
comparison to the between-wheel path area, and this variability in binder application
rates has the potential to affect the proportion of air voids within the chip seal that is
filled with binder. However, in the presented research one of the objectives was to
observe the changes that occurred to a chip seal structure as flushing developed, and it
was necessary to obtain field pavement samples that were not flushed and exhibited
the original aggregate packing structure and the original texture of the chip seal
surfacing when it was constructed. Therefore, the between-wheel path areas of the
pavements were sampled instead of the wheel path areas.

3.5 Laboratory Testing Methodology

The mechanical tests that were performed on the chip seal cores included
measurement of the volume of air voids, wheel tracking of cores, extraction of
bitumen from cores, and determination of aggregate grading profiles of cores. The
specific tests that were conducted in order to obtain the above noted measurements are
summarised in Table ‎3.2.

Table ‎3.2 Test methods used in the study

Measurement Test method used

Measurement of air void volume:

Bulk specific gravity ASTM D2726 - 05a Standard Test Method for Bulk Specific
Gravity and Density of Non-Absorptive Compacted Bituminous
Mixtures.
Theoretical maximum specific ASTM D2041 - 95 Standard Test Method for Theoretical Maximum
gravity Specific Gravity and Density of Bituminous Paving Mixtures.

3-61
Average volume of air voids ASTM D3203 – 05 - Standard Test Method for Percent Air Voids in
Compacted Dense and Open Bituminous Paving Mixtures.

Wheel tracking of cores:

Wheel tracking AGPT/T231 - Deformation Resistance of Asphalt Mixtures by the

wheel tracking test.

Binder and aggregate measurements:

Binder extraction ASTM D2172M – 11 Standard Test Methods for Quantitative

Extraction of Bitumen From Bituminous Paving Mixtures.
Aggregate grading profile ASTM C136 - 06 Standard Test Method for Sieve Analysis of Fine
and Coarse Aggregate.
Aggregate shape analysis A test method developed specifically for this study was used for
aggregate shape analysis.

Prior to conducting the tests, the cores in their original state were measured for
thickness and diameter. The cores were then trimmed to obtain just the chip seal
section, and these trimmed cores were photographed from the top and sides (as shown
in Figure ‎3.5). General observations of the state of the cores were noted, such as the
presence of flushing spots on the surface. Most cores displayed a few spots of flushing
on the surfaces, similar to the core in Figure ‎3.5a. The diameters of the cores ranged
from 193.4 mm to 243.8 mm and the thicknesses of the cores ranged from 26.3 mm to
49.5 mm. A full inventory of the cores is shown in Table A.1 in Appendix A.

3-62
 Spots of
flushing

a) Top view of a core b) Side view of a trimmed core

Figure ‎3.5 Top and side views of a core from Christchurch site 1

3.5.1 Measurement of Volume of Air Voids

The cores were tested to measure the undisturbed volume of air voids using the test
method set out in ASTM D3203 – 05 (ASTM International, 2005b). The ASTM
D3203 – 05 test method is generally used to measure the volume of air voids in
asphalt concrete cores. As the cores that were used in the documented research
consisted of a number of chip seal layers, the thicknesses of the cores were large
enough (26.3 mm to 49.5 mm) to treat them similarly to dense graded asphalt cores
and apply the ASTM D3203 – 05 test. Determining the air voids in the cores required
measuring the density and the bulk specific gravity using the test method set out in
ASTM D2726 - 05a (ASTM International, 2005a) and the theoretical specific gravity
using the test method set out in ASTM D2041 - 95 (ASTM International, 1995). The
bulk specific gravity of the cores was found by measuring the mass of the cores in air
(Massair), in water (Masswater) and when saturated and dried (MassSSD). Firstly, the
mass of cores in water was found by placing each core in a wire basket which was
attached to a scale and submerged in a tub of water (setup of scale is shown in Figure
‎3.6) at a water temperature of 25oC. The core was left in the wire basket for three
minutes to allow for fluctuations in the mass measurement to stabilise and then the
result was recorded as Masswater. The core was then taken out of the submerged wire
basket, dried quickly to remove excess moisture and weighed to measure the saturated
surface dry mass of the core (MassSSD). Finally, the core was left to dry completely

3-63
and weighed to measure the mass of the core in air (Massair). From these mass
measurements the density and bulk specific gravity were calculated. The mass
measurements made for each individual core are shown in Table A.1 in Appendix A.

Scale

Water bath

Wire
basket

Figure ‎3.6 Setup of scale with water bath

In order to find the theoretical specific gravity, the cores were softened in an oven
until the aggregates were able to be separated. The separated core mixture was then
left to dry, and once dried a weighted amount of the mixture was put into a metal
vacuum container (setup shown in Figure ‎3.7) and filled with water until the mixture
was completely submerged. A vacuum was applied to the container for 10 minutes. At
the end of the 10 minutes, the vacuum was released and the container was filled
completely with water and the mass of the filled container was measured. From these
weight measurements the theoretical specific gravity of the bitumen and aggregate
mix was calculated. The average air void volumes of the cores were then calculated
using the bulk specific gravity and theoretical specific gravity measurements. The
theoretical specific gravity measurements and air voids measurements for the cores
are shown in Table ‎3.4. As the test for theoretical specific gravity was destructive, the
cores were tested for theoretical specific gravity after wheel tracking was completed.

3-64
 Vacuum pump

Lid with
vacuum
attached

Vacuum
container

Figure ‎3.7 Setup of vacuum sealing apparatus

3.5.2 Wheel Tracking

The cores that were obtained from field sampling were subjected to lateral cyclic
loading using a wheel tracking test device in order to examine the micromechanical
changes that occur within a chip seal layer when subjected to loading. The wheel
tracking machine, shown in Figure ‎3.8, consisted of a steel wheel with a non-threaded
rubber bearing rim which was positioned above a moving table (Austroads, 2006a).
The function of the moving table was to hold a core and oscillate back and forth
laterally under the wheel to simulate the movement of a tyre. The outside diameter of
the wheel was 200 – 205 mm and the width was 50 ± 1 mm. The wheel was loaded
with a 700-Newton load. The wheel tracking machine included a vertical
displacement measuring device which was used to measure the vertical deformation
(rut depth) on the surface of the core during wheel tracking. The wheel tracking
device was housed in a temperature controlled cabinet that had sensors to measure the
temperature during testing. Each core that was wheel tracked was set in a plaster
casing within a wooden test mould to hold it steadily during wheel tracking, as shown
in Figure ‎3.9, and the mould with the core was placed on the moving table. Prior to
starting the wheel tracking test, the core was conditioned in the temperature controlled
cabinet at a constant temperature of 60°C for 16 hours. The wheel tracking was then
performed, also at a constant temperature of 60°C.

3-65
 Vertical
displacement
measuring
device

Rubber wheel
Load

Core in plaster casing

Moving table Temperature
sensors

Figure ‎3.8 Setup of the wheel tracking machine

Plaster
casing

Core Wooden
test mould

Figure ‎3.9 A core set in the wooden test mould and prepared to be wheel
tracked

One core from each site that was sampled was used as a trial core to determine the
number of wheel passes (wp) to use for the other cores from that site. Each trial core
was wheel tracked to 20,000 wp and the rut depth on the surface of the core was
3-66
recorded against the number of wheel passes that were used. The rutting pattern on the
trial core was used to select the number of wheel passes to use for the rest of the cores
from the corresponding site. Three wheel pass values were chosen for each site from
the trial core and three test cores were tracked using the selected number of wheel
passes. The number of wheel passes used on the test cores for each site is shown in
Table ‎3.3.

Table ‎3.3 The number of wheel passes used to track cores

No. of wheel passes

Region Site
Core 1 Core 2 Core 3

Auckland 1 2,000 5,000 10,000

Auckland 2 2,500 5,000 10,000
Waikato 1 5,000 10,000 20,000
Waikato 2 1,000 2,500 5,000
Christchurch 1 2,500 5,000 10,000
Christchurch 2 2,500 5,000 10,000

Dunedin 1 2,500 5,000 10,000

Dunedin 2 2,500 5,000 10,000

During the wheel tracking test the rut depth on each core was measured at the centre
of the core and at locations of ± 7.5 mm, ± 22.5 mm and ± 37.5 mm from the centre of
the core along the wheel path, and the rut depth measurements recorded at these
points were averaged to find the final rut depth on a tracked core. The accuracy of rut
depth measurements is typically 0.3 mm for field samples, and the rut depth
measurements that were obtained from the wheel tracking test were rounded to the
nearest 0.1 mm. In addition to rut depth, measurements were taken of the area of
flushing that was displayed on the wheel tracked area of each core at the conclusion of
the wheel tracking test. The flushed area on a core was measured manually using a
ruler, and the measured flushed area was divided by the wheel tracked area of the core
to obtain the percentage of flushing for the core. On each core, multiple measurements
were taken of the flushed area and the measurements were averaged to ensure the
measured flushed areas were accurate. Each wheel tracked core was then
3-67
photographed and these post-tracking photographs were compared to pre-tracking
photographs of the cores to identify aggregate breakdown that may have occurred
during the wheel tracking test.

3.5.3 Bitumen Extraction and Aggregate Grading Measurements

The wheel tracked cores were tested to determine the binder content and aggregate
grading. The binder extraction and aggregate grading tests were performed as part of
an undergraduate research project conducted at‎ The‎ University‎ of‎ Auckland’s‎
Department of Civil and Environmental‎Engineering,‎titled‎“Investigating‎the‎impact‎
of‎grading‎distribution‎on‎flushing‎of‎chip‎sealed‎roads”.‎This undergraduate project
was carried out to supplement the outcomes presented in this thesis and to add to the
understanding of the volumetric behaviour of chip seals contributing to flushing.

The binder extraction and aggregate grading tests were performed on all the cores that
were wheel tracked. The binder content of the cores was measured according to the
test method set out in ASTM D2172M – 11 (ASTM International, 2011). Each wheel
tracked core was melted in an oven to soften the core and separate the core mix for
binder extraction. The separated mix was heated in a solvent solution of
Trichloroethylene to extract the binder. Once the binder was extracted, the core mix
was cleaned with more solvent to remove excess binder from the aggregates. The
aggregates were then sieved to determine the aggregate grading curve of the mix
according to the test method set out in ASTM C136 – 06 (ASTM International, 2006).
When performing the aggregate grading tests it was assumed that wheel tracking did
not cause aggregate degradation to an extent that affected the original grading curve
of the core mix. Thus, the behaviour of the cores during the wheel tracking test was
due purely to the original grading distribution and the shape characteristics of the
aggregates. A complete inventory of the measurements recorded during the bitumen
extraction and aggregate grading tests is presented in Table B.1 and Table B.2 in
Appendix B.

3-68
3.5.4 Aggregate Shape and Angularity Measurements
Once the aggregate grading curves were determined for each core a small sample of
aggregates was taken from each core mix to carry out aggregate shape analysis. Shape
analysis was carried out by photographing the aggregates and then using an image
analysis software to analyse the shape of each aggregate. The method of performing
aggregate shape analysis using digital images has been well documented in the
literature (Masad, 2003; Mahmoud et al., 2010; Henderson et al., 2011). In the
analysis that was conducted in the presented research the shape of the aggregates was
measured using the radius angularity measurement method (Masad, 2003). The
equipment setup for this task consisted of a table with a glass surface and under which
an 80-Watt lamp was positioned, as shown in Figure ‎3.10. A 9.1 megapixel camera
was used to photograph the aggregates and this camera was set up on a tripod and was
positioned parallel to the glass surface of the table. A white A4 size piece of paper
was laid on the glass surface to create a white surface and the aggregates were
positioned on top of the paper.

Camera set up
on tripod

Surface of table Aggregates on

the glass surface

Light

Figure ‎3.10 Setup of light table and camera

The aggregates that were used for photographing were selected from the portion of
each core mix that was larger than the 4.75 mm sieve. Each core mix was split until a
sample of approximately 15 aggregates was obtained. The aggregates were placed on
the surface of the light table and a photo was taken, as shown in Figure ‎3.11a, and the
3-69
obtained photo was read into the software programme MATLAB for analysis. A
screen shot of the MATLAB programme is shown in Figure ‎3.11b. An ellipse was
drawn around each aggregate outline and the perimeter of the ellipse and the outline
of the aggregate were measured. The ratio between the ellipse perimeter and the
aggregate outline was calculated to determine the angularity of each aggregate. The
aggregates that had very irregular outlines had a high angularity value. For each core
mix sample the angularity of each aggregate was determined and these values were
averaged across all 15 aggregates to obtain an average angularity for the sample. The
angularity results were then analysed against wheel tracking results to determine how
the aggregate angularity correlated with the flushing and rutting patterns that were
observed. A complete inventory of the aggregate angularity measurements that were
recorded during the aggregate shape analysis test is presented in Table B.3 in
Appendix B.

a) A photo of aggregates on the light table b) Screen shot of the MATLAB program
showing an aggregate and the ellipse

Figure ‎3.11 Aggregate shape analysis images

3.6 Volume of Air Void

In pavement surface mix design a particular volume of air voids is specified for each
mix to obtain satisfactory performance of the pavement surface. In asphalt concrete
pavement mixes the specification of this design volume of air voids is straightforward
3-70
as the manner in which the aggregates and bitumen are mixed together prior to
placement of the surface allows the volume of air voids to be controlled. In contrast,
on chip seals the bitumen and aggregate layers are laid separately, and therefore
specifying a design volume of air voids is not as simple because the constituents are
not mixed together prior to the chip seal being laid. When there are multiple chip seal
layers that are constructed on top of each other, such as on the sites that were sampled
for the reported research, the chip seal layers act as a single structure containing a
mixture of aggregates, bitumen and air voids. The composition of a thick chip seal
layer can be considered to be similar to an asphalt mix, where the asphalt mix also
consists of a mixture of aggregates, bitumen and air void. The difference between an
asphalt mix and a thick chip seal layer is that in the asphalt mix the proportions of the
aggregates, bitumen and air voids are determined prior to the mix being laid on the
pavement, and as such the asphalt‎ mix‎ can‎ be‎ considered‎ as‎ being‎ “designed”.‎ In‎
contrast, a thick chip seal layer is built up over time and therefore, the proportions of
aggregates, bitumen and air voids are not designed. Hence, a thick chip seal layer can
be expected to behave similarly to an un-designed asphalt layer. As there are no
guidelines to assess the volume of air voids of these multiple-layered chip seals, the
design air voids specification for thin asphalt mixtures was used. Table ‎3.4 shows the
average air void volumes that were measured for the sites sampled in this research.

Table ‎3.4 Air void volume measurements

Density Bulk Theoretical Volume

Massair Masswater MassSSD
Site at 25°C specific specific of air
(g) (g) (g)
(kg/m3) gravity gravity voids (%)
Auckland 1 3092.9 1731.7 3100.8 2.259 2252.3 2.363 4.4
Auckland 2 2424.9 1380.5 2438.9 2.291 2284.2 2.495 4.3

Waikato 1 1806.6 988.1 1814.1 2.187 2180.6 2.329 6.1

Waikato 2 2005.9 1094.6 2013.4 2.183 2176.6 2.309 5.4
Christchurch 1 2700.0 1499.2 2708.8 2.232 2225.4 2.302 3.0
Christchurch 2 2505.8 1390.1 2512.4 2.233 2226.0 2.310 3.3
Dunedin 1 3665.9 2098.2 3670.5 2.332 2324.6 2.446 4.7
Dunedin 2 3360.4 3370.0 1908.2 2.299 2291.9 2.406 4.4

3-71
The average volume of air voids in the cores varied in the range between 3.0% and
6.1%, with the cores from Waikato sites having the highest volume of air voids. Mix
specifications for thin asphalt mixtures recommend a design air void volume between
8.5% and 10% (Transit New Zealand, 2005), and when the volume of air voids of the
cores were compared to the asphalt mix specifications it was found that the measured
air voids of cores from all eight sites were below the specified air void volume values
for thin asphalt surfacing mixtures. Further analyses of the air void volumes of the
cores and an assessment of the effect of air void volume on the volumetric behaviour
of chip seal surfaces are presented in Chapter Four.

3.7 Wheel Tracking Results

The cores that were sampled as part of the documented research were subjected to
wheel tracking tests as outlined above. The rutting patterns on the cores differed
between the different regions as well as between sites. From all the cores that were
wheel tracked some common rutting patterns were identified and these patterns are
described in the paragraphs below. The observed rutting patterns were used to
understand the behaviour of the aggregate layers within the cores and the effect of that
behaviour on the flushing quantities that were present on the core surfaces. A detailed
description of the rutting patterns that were observed on the wheel tracked cores is
presented in subsequent sections.

One rutting pattern that was identified from the wheel tracked cores was a pattern
where rutting developed up to a point in the wheel tracking test and then reached a
stable state where no further rutting developed up until the termination of the tracking
test. This type of rutting pattern is depicted in Figure ‎3.12. The enlarged portions of
the rutting pattern in Figure ‎3.12 shows the likely state of the chip seal layers within
the core, where at the beginning portion of the tracking test when rutting is still
developing (at point A), there is spacing between the aggregate layers. As the core
continues to be tracked the spacing between the aggregate layers decreases and the
core reaches a stone-on-stone state where the aggregate layers are lying directly on
top of each other (at point B). When a flushed chip seal pavement displays the
stabilising rutting pattern then in this situation the flushing can be treated with Ultra

3-72
High Pressure (UHP) water cutting to remove the excess bitumen (Gransberg and
Pidwerbesky, 2007) and further flushing will not take place as bitumen migration due
to densification cannot occur.

Aggregate
layers

Figure ‎3.12 Schematic showing the development of the stable rutting pattern
and the corresponding state of aggregate layers in the core

The other rutting pattern that was identified from the wheel tracked cores was a
pattern where rutting continued to develop until the termination of the tracking test.
This type of rutting pattern is depicted in Figure ‎3.13. In this situation the aggregate
layers slip past each other and the chip seal structure continues to deform with
increasing loading. If a sample from a flushed site was showing this continuous
rutting pattern, then that chip seal structure had not reached a stone-on-stone state and
flushing will continue to develop due to continuous densification. In this case water
cutting or resurfacing will not remedy the flushing and pavement rehabilitation is
necessary (O'Flaherty, 2002).

3-73
 C
B

Aggregate
layers

Figure ‎3.13 Schematic showing the development of the continuous rutting

pattern and the corresponding state of aggregate layers in the core

3.7.1 Auckland Sites

The wheel tracking results for cores from Auckland sites are shown in Figure ‎3.14.
The graphs show the rut depth that was recorded for each core during the wheel
tracking test and the three lines on each graph show the three cores that were tracked
at the three wheel pass values selected from trials. For Auckland site 1 the cores were
wheel tracked at 2,000 wp, 5,000 wp and 10,000 wp and for Auckland site 2 the cores
were wheel tracked at 2,500 wp, 5,000 wp and 10,000 wp.

For Auckland site 1 (Figure ‎3.14a), the final rut depths on the three cores were
significantly different, where the final rut depths on Core 1 (2,000 wp) and on Core 2
(5,000 wp) were much higher than the final rut depth on Core 3 (10,000 wp). Rutting
on Core 3 (10,000 wp) stabilised after 3,000 wp and the wheel tracking test
terminated automatically at 6,600 wp as no change in rutting had occurred. Thus, the
rutting trend for Core 3 was extrapolated for the unperformed portion of the wheel
tracking test. Core 1 appeared to have potential for further rutting, which indicated
that the air voids were being extruded out from within the core. Measurements were
made at the conclusion of the wheel tracking tests of the amount of flushing that was

3-74
present on the wheel tracked area of the cores. All three cores from site 1 displayed
large areas of flushing, with 90% of the wheel tracked area being flushed on all three
cores.

a) Site 1

b) Site 2

Figure ‎3.14 Wheel tracking results for Auckland cores

3-75
For Auckland site 2 (Figure ‎3.14b), the final rut depths on the cores were much lower
than for the cores from site 1. The rutting trends on Core 1 (2,500 wp) and Core 3
(10,000 wp) were similar, with a slow increase in rutting taking place with the
increase in wheel passes. In contrast, the rutting trend on Core 2 (5,000 wp) was much
higher, reaching a final rut depth of 9.9 mm. The flushed area of the cores varied on
this site, with Core 1 displaying flushing on 30%, Core 2 on 80% and Core 3 on 90%
of the wheel tracked area. The lower rutting experienced by these cores was identified
as the reason for the varied flushed areas.

3.7.2 Waikato Sites

The wheel tracking results for cores from Waikato are shown in Figure ‎3.15. The
cores from Waikato site 1 were wheel tracked at 5,000 wp, 10,000 wp and 20,000 wp
and the cores from Waikato site 2 were wheel tracked at 1,000 wp, 2,500 wp and
5,000 wp.

For site 1 cores (Figure ‎3.15a), the trend in rutting varied significantly for the three
cores. Core 1 (5,000 wp) and Core 2 (10,000 wp) appeared to have reached a state
where the rutting stabilised after approximately 1,000 wp. Conversely, Core 3 (20,000
wp) had a slow increase in rutting until the termination of the wheel tracking test. This
contrast in rutting patterns between the three cores was due to the manner in which the
aggregate layers were behaving in the cores. For example, Core 1 and Core 2 had
reached a stone-on-stone state, thus a stabilised rutting pattern was present. In
contrast, the rutting behaviour in Core 3 was due to a continuous slippage between the
aggregate layers which allowed Core 3 to continue to rut until the termination of the
test. Furthermore, at the conclusion of the wheel tracking tests, all three cores from
site 1 displayed large areas of flushing with Core 1 and Core 2 displaying flushing on
70% and 75% of the wheel tracked areas respectively and Core 3 displaying flushing
on 80% of the wheel tracked area. This observation showed that flushing was able to
develop even at relatively small loading, for example at 5,000 wheel passes compared
to 20,000 wheel passes, and that flushing development was more dependent on the
reduction of air voids rather than on the number of repetitions of loading applied to
the core.

3-76
 a) Site 1

b) Site 2

Figure ‎3.15 Wheel tracking results for Waikato cores

For site 2 cores (Figure ‎3.15b), the trend in rutting was similar between all three cores
where rutting developed rapidly immediately after beginning the wheel tracking tests
and then reached a gradual increasing state. Although there is a difference in the final
rut depths reached in the three cores, it can be seen that extrapolating the rutting
3-77
trends of Core 1 and Core 2 to 10,000 wp would yield final rut depths close to the 7.7
mm rut depth reached on Core 3. The areas of flushing that were displayed on Core 1,
Core 2 and Core 3 were 50%, 70% and 75% respectively.

3.7.3 Christchurch Sites

The cores that were taken from site 1 in Christchurch were wheel tracked at 2,500 wp
5,000 wp and 10,000 wp and the wheel tracking results for these cores are shown in
Figure ‎3.16a. The cores from Christchurch site 2 were also wheel tracked at 2,500 wp,
5,000 wp and 10,000 wp and the wheel tracking results are shown in Figure ‎3.16b. On
site 1 cores, the trend in rutting was similar across all three cores, and the increase in
rutting was steady throughout the wheel tracking tests. Core 1 and Core 2 appeared to
have slightly higher rutting than Core 3, although this difference was small. Core 3
(10,000 wp) experienced the highest rutting across all the sites that were sampled.
This core reached a final rut depth of 14.7 mm and showed potential for further
rutting. Surprisingly, Core 3 was not completely flushed given the high rut depth that
was reached, and flushing was only present on 80% of the wheel tracked area. This
behaviour was due to the thickness of Core 3 being smaller compared to some of the
other cores that were tested from other sites, resulting in a lower amount of
bituminous binder that could have appeared as flushing. The areas of flushing that
were present on Core 1 and Core 2 were only 25% and 50% respectively even though
these two cores had reached final rut depths of 9.3 mm and 13.5 mm.

On site 2 (Figure ‎3.16b), the trend in rutting on all three cores was similar, with a
gradual development of rutting throughout the wheel tracking tests. This rutting trend
was different from the cores of site 1 where the magnitude of the rut development was
larger. The final rut depths on the three cores from site 2 were significantly lower than
for site 1 cores, with Core 3 (10,000 wp) from site 2 having a final rut depth of only
5.0 mm compared to Core 3 of site 1 which reached a rut depth of 13.5 mm. The areas
of flushing that were measured on Core 1, Core 2 and Core 3 of site 2 were 10%, 20%
and 60% respectively.

3-78
 a) Site 1

b) Site 2

Figure ‎3.16 Wheel tracking results for Christchurch cores

3.7.4 Dunedin Sites

The wheel tracking results for cores from Dunedin sites are shown in Figure ‎3.17. The
cores from site 1 in Dunedin were wheel tracked at 2,500 wp, 5,000 wp, and 10,000
3-79
wp, and the cores from site 2 were also wheel tracked at 2,500 wp, 5,000 wp and
10,000 wp. The pattern of rutting on site 1 cores (Figure ‎3.17a) was similar across all
three cores, with a gradual increase in rutting occurring with the increase in wheel
passes. However, the actual rut depth on the three cores at a given point in the wheel
tracking test was different. For example, at 2,000 wp Core 1 (2,500 wp) had 3.7 mm
of rutting while Core 3 (10,000 wp) had 6.0 mm of rutting.

On Dunedin site 2 (Figure ‎3.17b) the pattern of rutting was similar between Core 1
(2,500 wp) and Core 2 (5,000 wp), where there was a sudden increase in rutting at the
beginning of the wheel tracking tests. On Core 3 (10,000 wp) the increase in rutting
was more gradual than for the other two cores. The final rut depths that were reached
on Cores 1 and 2 were significantly higher than the rut depth reached on Core 3, with
Core 2 reaching a rut depth of 12.8 mm whereas Core 3 only reached a rut depth of
8.9 mm. This rutting pattern also showed the variation in deformation characteristics
at different areas of a chip seal which was attributed to the manner in which the layer
is composed. In addition to the rut depths, the areas of flushing on the wheel paths of
the wheel tracked cores were also measured after the wheel tracking tests were
completed. The flushed wheel path areas on site 1 cores were 15% for Core 1 (2,500
wp), 35% for Core 2 (5,000 wp) and 70% for Core 3 (10,000 wp). The flushing that
had occurred on the cores from site 1 correlated positively with the final rut depths
reached on the cores. In contrast, all site 2 cores exhibited flushing on 95% - 100% of
the wheel path area.

3-80
 a) Site 1

b) Site 2

Figure ‎3.17 Wheel tracking results for Dunedin cores

3.7.5 Summary of Wheel Tracking Results

The rutting patterns observed on the eight test sites showed that there was a wide
variation in the rutting behaviour of individual cores at a site, which indicated that
3-81
there were many irregularities present within the chip seal structures even at closely
spaced locations on a particular site. These irregularities help to explain why the
occurence of flushing is so varied even on different areas of one site. The variations in
rutting between cores of a given site was highest on sites from Auckland, Waikato and
Dunedin. On the sites sampled from Christchurch the rutting patterns were similar
between cores of the same site. The cores from site 1 and site 2 of Christchurch had
the lowest chip seal thicknesses of all the sites, and this lower thickness would have
played a role in the rutting patterns observed on Christchurch sites. This observation
indicated that as the thickness of the chip seal section of a pavement increases, its
rutting behaviour caused by the interaction between aggregate layers becomes
unpredictable.

As mentioned earlier in the chapter, examining the rutting patterns will be of

significant benefit when selecting maintenance treatments for flushed chip seal
pavements. In addition to this benefit, wheel tracking results can also be used for
better management of chip seal sites. When undertaking pavement performance
surveys for flushing detection, samples from problematic sites should be wheel
tracked in a similar way to the method used in this chapter to determine the
deformation behaviour of the chip seal structures and by observing the deformation
behaviour of a chip seal, early treatment methods could be selected. In asphalt
concrete pavements, cores are routinely taken and wheel tracked to assess the
functional performance of the surfacing mix, and adopting a similar sampling method
for chip seal pavements would also immensly enhance the assessment measures,
particularly when noticable levels of flushing are present. Analysing problematic sites
in this way would help to minimise catestropic failures due to flushing.

3.7.6 The Role of Chip Seal Thickness on Flushing Patterns

The areas of flushing that were measured on the wheel tracked cores varied
significantly between the different sites. The amount of flushing did not always
correlate with the number of wheel passes that were used, where Core 1 from
Auckland site 1 (2,000 wp) displayed flushing on 90% of the wheel tracked area while
Core 1 from Christchurch site 1 (2,500 wp) had only 25% flushing. One main

3-82
difference between these two cores was that the Christchurch core was significantly
thinner (30.5 mm) compared to the Auckland core (48.8 mm). In order to investigate
whether flushing on all other cores was also affected by the core thickness these two
measurements were plotted on Figure ‎3.18. As can be seen, there is a moderately
strong correlation (R2 = 0.459) between core thickness and the measured area of
flushing, where the thicker cores displayed higher amounts of flushing. The cause
behind the result in Figure ‎3.18 is that thicker cores are able to deform a lot more
which means that the amount of embedment of aggregates that occurs is greater for
these cores. Furthermore, thicker cores take a longer time to reach a stable stone-on-
stone state than thinner cores and so the amount of bitumen that migrates up during
the deformation process is higher.

Figure ‎3.18 Relationship between core thickness and flushing on wheel tracked
cores

To confirm the effects of core thickness on the rutting behaviour the initial rut rate
observed on each of the wheel tracked cores was plotted against the core thickness
(Figure ‎3.19). The initial rut rate was taken as the rut rate during the first 1,000 wheel
3-83
passes, which included the portion of the tracking test where most of the densification
occured. This portion therefore indicated whether there was instability in the chip seal
structure such that it would rut to a high level as the tracking test progressed. The
result in Figure ‎3.19 showed that the initial rut rate was higher on thicker cores, which
confirmed the supposition that instability caused by the build up of aggregate layers
does contribute to high initial rutting, which then gives rise to higher amounts of
flushing.

Figure ‎3.19 Initial rut rate measured from wheel tracking test against the
thickness of core

3.8 Binder Content Results

Once the cores were wheel tracked they were tested to extract binder in order to
determine the amount of binder that was present in the core mixes. The binder content
results of the cores are shown in Table ‎3.5. On average, Waikato sites had the highest
binder contents out of all the regions that were sampled, with the two sites having
average binder contents of 9.9% and 9.7%. When comparing the binder contents of
3-84
cores of a given site it is worth noting that on some sites, particularly Waikato site 1
and site 2, and Dunedin site 1, there was a notable variation in the binder contents
between the three cores. This variation in binder contents between cores is interesting
because given that the cores were taken very close together it was expected that the
binder contents would be similar between the three cores.

Table ‎3.5 Binder contents of wheel tracked cores

Binder Content (% by mass)

Region Site
Core 1 Core 2 Core 3 Average

1 8.2 8.8 8.4 8.5

Auckland
2 6.8 7.1 6.8 6.9

1 10.4 10.0 9.3 9.9

Waikato
2 9.0 11.1 9.1 9.7

1 8.9 8.5 9.7 9.0

Christchurch
2 8.3 8.3 7.9 8.2

1 5.1 7.5 7.4 6.7

Dunedin
2 7.5 7.6 7.6 7.6

When a chip seal pavement surface is constructed the design specifications

recommend an optimum binder quantity for each individual chip seal layer that makes
up the pavement surface. However, when a pavement surface consists of multiple chip
seal layers as a result of numerous reseals over time, there are no specifications to
assess the binder quantity of the entire chip seal structure. In the absence of such a
specification the asphalt mix specifications were used as a guide to assess the binder
contents found on the cores. As mentioned earlier in the chapter, the cores consisted
of multiple chip seal layers that were acting as a single structure and resulted in the
entire chip seal structure behaving similarly to an un-designed asphalt layer. Thus, the
asphalt mix specifications for thin asphalt surfacings were an appropriate guide to use.
The‎ mix‎ specifications‎ for‎ asphalt‎ concrete‎ issued‎ by‎ NZTA,‎ “TNZ M/10
Specification for Asphaltic Concrete” (Transit New Zealand, 2005), were used to
assess the binder contents of the cores. This specification recommends an effective
3-85
binder content of 5.0% - 6.5% by mass for thin surface mixes. Most of the tested cores
had binder contents above this specification, with the cores from site 1 and site 2 of
Waikato having binder contents between 9% and 11%, which was well above the
recommended limits.

Past research and knowledge of pavement mix behaviour suggest that the volumetric
balance between the binder content, the aggregate volume and the air void volume of
a chip seal layer is important to its functional performance. In particular, the ratio
between the binder volume and the aggregate volume (bitumen-to-aggregate ratio) is
one measure that can influence flushing. If the bitumen-to-aggregate ratio is high,
then that can make the surfacing layer unstable and quickly lead to flushing (Gray and
Hart, 2003). In order to explore the effect of binder content on flushing, the bitumen-
to-aggregate ratio of each wheel tracked core was plotted against the flushing that was
measured on that core. The results are presented in Figure ‎3.20, and as can be seen in
Figure ‎3.20 the correlation between flushing and the bitumen-to-aggregate ratio is
weak (R2 = 0.034). This result indicated that even though the binder contents within
the cores were higher than recommended limits, the ratios between the binder volume
and the aggregate volume of the cores were not significant enough to cause flushing.

The absence of a strong correlation between flushing and the bitumen-to-aggregate

ratio indicated that the bitumen-to-aggregate ratio of a chip seal layer is not a
significant contributor to flushing. In order to further explore the effect of bitumen
content on flushing development, investigations were conducted to identify other
factors that were impacting on the flushing/bitumen content relationship. Past research
had theorised that flushing was also affected by the inter-relationship between the
bitumen content and the air void volume in a chip seal structure (Krishnan and Rao,
2001), where the available volume of air voids in a chip seal layer determined the
amount of bitumen movement that could occur within the chip seal layer before
flushing developed. Therefore, an investigation was conducted to determine the
correlation between flushing, binder content and air void volume. The results of this
investigation are presented in Section ‎4.5.1.

3-86
 Figure ‎3.20 Correlation between flushing and the bitumen-to-aggregate ratio

3.9 Relationship between Aggregate Grading and Flushing

The cores that were wheel tracked were tested to determine the grading curves of the
aggregates. These grading curves are shown in Figure ‎3.21. The grading curves were
classified into three flushing categories, being low, medium and high, which were
based on the amount of flushing that was present on the wheel tracked area of each
core. These flushing severity categories were selected based on the clustering of the
data, where the measured flushing values were clustered around 30%, 60% and 90%.
The flushing classifications that were used were as follows:
 Low – flushing on 0% - 40% of the wheel tracked area;
 Medium – flushing on 41% - 79% of the wheel tracked area; and
 High – flushing on 80% - 100% of the wheel tracked area.

Previously conducted studies on the topic of flushing (Ball and Patrick, 2008) had
indicated that the development of fine solid material (fines) in a chip seal layer often
leads to flushing due to the fines occupying air void volume that would otherwise be

3-87
available for bitumen, thereby causing the bitumen to exude to the surface of the chip
seal. In order to investigate the role of fines on flushing development on the cores of
the presented study, the fines contents from the wheel tracked cores were investigated.
According to aggregate grading specifications provided by NZTA, the proportion of
aggregates that pass the 4.75 mm‎sieve‎is‎commonly‎classed‎as‎“fines”‎(Transit New
Zealand, 2005), and investigating the fines proportions in the cores can indicate if the
fines actually do have a significant effect on the flushing potential of a seal.

Figure ‎3.21 Grading curves for wheel tracked cores classified into the extent of
flushing

From the results shown in Figure ‎3.21 it can be seen that there is no significant
difference in the proportion of aggregates passing the 4.75 mm sieve between highly
flushed cores and lightly flushed cores. Additionally, when visually comparing cores
that were seen to have medium flushing against those cores that had high flushing
there appears to be no particular difference in the fines proportions between these two
categories of cores. These observations differ from conclusions reached in previous
3-88
studies investigating fines and its relation to flushing, where Ball and Patrick (2008)
found that fines were higher in flushed chip seal sites and concluded that the build-up
of fines over time accelerated flushing. It should be noted that the proportion of fines
that were found on these cores comprised only the fines that had developed in the
field, and wheel tracking did not cause degradation to the aggregates that would have
affected the grading distributions significantly. Thus, the presented results are an
accurate reproduction of the expected effect of fines on the flushing process of a chip
seal pavement. From these results it is difficult to say with certainty that fines in a
chip seal impact on its flushing development.

While the fines content was seen to not directly affect the amount of flushing on the
wheel tracked cores it is possible that the fines still have an effect on the overall
behaviour of the core. The results from wheel tracking revealed that the deformation
behaviour of the cores could be separated into two patterns, with one pattern being
where there was a continuous increase in rutting until the termination of the tracking
test and the other pattern being where there was an increase in rutting to a certain
point and then stabilisation of rutting. These deformation patterns were assumed to
have occurred due to the interaction of the aggregate layers in the cores, which
determined whether the core reached a stable stone-on-stone state or continued to rut.
In order to investigate the underlying cause of these two deformation patterns the
aggregate grading distributions of the cores were analysed. The cores were classified
into two categories, being either stable or unstable. The stable category consisted of
cores where rutting stabilised during the wheel tracking test, and the unstable category
consisted of cores where rutting continued to develop up to the end of the tracking
test. The grading curves were plotted based on this classification and the results are
shown in Figure ‎3.22. In the legend on Figure ‎3.22 the “U”‎denotes‎unstable‎cores‎and‎
the “S”‎denotes‎stable‎cores.

3-89
 Figure ‎3.22 Aggregate grading curves of wheel tracked cores classified into
stability of the core

When comparing the stable cores against the unstable cores it can be seen that the
unstable cores had a lot more fines than did the stable cores, and that the stable cores
had a larger proportion of aggregates that were greater than 4.75 mm. Statistical
analysis of the fines quantity data revealed that the difference between the fines
quantity of stable cores in comparison to the unstable cores was statistically
significant (p value = 0.001). The reason for this trend was that the fines that are
generated through aggregate break down over time affects the interlocking strength
between two aggregates. On the unstable cores the presence of a higher proportion of
fines would have prevented strong interlock between the aggregates. This lack of
interlocking strength between the aggregates had caused the aggregates to slide past
each other easily, resulting in a weaker aggregate matrix that continued to deform. In
contrast, on the stable cores the fines proportion was lower indicating that the fines
did not have a significant impact on the interlocking strength between the aggregates.
As a result, the aggregates in the stable cores had created a stronger matrix that was
able to resist further deformation after the stone-on-stone state was reached. There

3-90
were a few unstable cores, namely A.S2.1, H.S1.9, C.S1.12, C.S2.7 and C.S2.10, that
had aggregate grading curves that were similar to the stable cores. However, these
cores are likely to be exceptions because the other cores that were taken from the
same sites as these cores were seen to have stable rutting patterns. Also, these cores
were wheel tracked to lower amounts of wheel passes (2,500 wp and 5,000 wp) and if
the wheel tracking was left to continue for longer, for example for 10,000 wp or
20,000 wp, then the rutting pattern would have stabilised. The conclusion that can be
reached by analysing the above presented results is that the presence of fines affects
the stability of a core structure which in turn determines the effectiveness of a
maintenance treatment for flushing. As mentioned previously, flushing on a surface
that has a stable rutting pattern can be treated with water cutting while a surface with
an unstable rutting pattern would need to be rehabilitated. Thus, while the fines do not
directly contribute to the amount of flushing that develops on a chip seal, they do
affect the maintenance options and it is important that chip seals are designed to
minimise fines in order to make maintenance easier.

In order to further investigate the contribution of aggregate characteristics to the

stability of the cores, the effects of aggregate angularity were investigated. The
angularity measurements for each core in the data set as well as the average aggregate
angularity of the stable and unstable groups are presented in Figure ‎3.23. It can be
seen in Figure ‎3.23 that there is no significant difference in the angularity of
aggregates between the stable and unstable cores. This result is surprising as
knowledge of aggregate behaviour suggests that stability of a surfacing layer is
affected by the presence of highly angular aggregates which create strong interlock to
resist deformation (Pan et al., 2006). Further to the above observation, it was also
noted that all the stable cores were from North Island sites (Auckland and Waikato)
while the majority of unstable cores were from South Island sites (Christchurch and
Dunedin). In the North Island surfacing aggregates are mainly sourced from volcanic
sources, where the majority of aggregates are greywacke. In contrast, in the South
Island surfacing aggregates are mainly sourced from alluvial sources (Black, 2009;
Christie & Barker, 2007). This difference in aggregate morphology could be used to
explain the difference in deformation behaviour of the chip seal cores, where
greywacke aggregates are more angular than aggregates from alluvial sources, and the

3-91
presence of angular faces can provide better interlock between the individual
aggregates resulting in the chip seal core reaching a stable state.

Figure ‎3.23 Angularity of aggregates classified into the deformation pattern of

the cores

As part of the aggregate shape analysis, visual observations were conducted and
photographs were also taken of the wheel tracked cores to observe any aggregate
breakdown that may have occurred during the wheel tracking test. However, there
was no evidence of aggregate breakdown on the surfaces of the cores, and this
observation was used to conclude that there was no significant difference in the
angularity of aggregates between stable and unstable cores. It is recognised that visual
observations of the surface of the cores alone cannot indicate any aggregate
breakdown that may have occurred within cores, and this is a limitation of the
aggregate shape analysis task that was conducted as part of the presented research.
However, in order to achieve the objectives of the research and conduct further testing
3-92
of the cores, it was required to keep the cores in their original wheel tracked state. In
order to further explore the aggregate angularity effects on chip seal deformation
behaviour, it is recommended that sampling and testing is conducted on flushed sites
that contain aggregates with different morphological characteristics, such as steel slag
aggregates, as this would better demonstrate aggregate angularity effects.

It should be noted that when considering the effects of aggregate properties on the
stability of chip seal layers there are two characteristics of aggregates that can impact
on the deformation behaviour of a chip seal. These two factors are the grading of the
aggregates and the properties of individual aggregates. The aggregate grading effects
were investigated and presented earlier in the thesis. When considering the aggregate
properties, there are two properties of aggregates that can affect the deformation and
the subsequent flushing behaviour of a chip seal layer. These properties are the shape
of the aggregates and the coarseness of the aggregates. In the lab experiments that
were conducted in the presented research only aggregate shape effects were explored.
It is recognised that the coarseness of the individual aggregates also needs to be
explored in order to fully investigate the effects of aggregates, although detailed
investigation of the aggregate coarseness effects were outside of the scope of the
presented research. It is recommended that aggregate coarseness effects on the
deformation behaviour of chip seal layers are investigated in future studies on this
subject.

3.10 Discussion
The experiments that were conducted as part of this chapter provided valuable
information about the flushing behaviour of chip seal pavements. While much was
learned, there were some challenges that were met during the experimentation
process. These challenges were associated with the availability of relevant test
methods, and the practicalities of conducting certain tests to obtain necessary results.
Some of these issues are discussed in the following sections.

3-93
3.10.1 Availability of Standard Test Methods for Investigation of
Chip Seals
One of the biggest limitations that was encountered during the laboratory test
procedures presented in this chapter was the lack of availability of standard test
methods for testing chip seal cores. The tests that were performed on the chip seal
cores included measuring the volume of air voids, wheel tracking, bitumen extraction
and aggregate grading tests, and all the methods that were used to conduct these tests
were originally developed for asphalt concrete cores. As mentioned earlier in the
chapter, the thick, multiple-layered chip seal structures of the chip seal cores allowed
asphalt concrete test methodologies to be adopted, although some of the test criteria
had to be relaxed in order to obtain useful results. One such case was the use of the
wheel tracking test method, Austroads AGPT/T231 (Austroads, 2006a), which
recommends that cores that are wheel tracked meet a minimum thickness criterion of
35 mm. The wheel tracking test is commonly used to test the deformation
performance of asphalt mixes made in the laboratory which means that the
thicknesses of the resulting cores can be adjusted easily to meet the minimum
thickness criteria required for the wheel tracking machine. When testing field cores,
such as the chip seal cores that were tested in this research, it is difficult to ensure that
the minimum thickness criterion will be met in all cases. Most of the chip seal cores
that were wheel tracked met this 35 mm thickness criterion apart from the
Christchurch cores which were slightly lower in thickness, even though they still had
four chip seal layers and met the sampling criterion that was set for the research
experiments. Even though the thicknesses of the tested cores were not ideal, the
results that were obtained by wheel tracking the cores were useful.

As described in the methodology section of this chapter, when conducting the wheel
tracking tests, the cores were confined in a plaster casing within a wooden mould
which held the core in place during wheel tracking. One of the restrictions of the
wooden mould was that when the core was loaded the core material was prevented
from moving laterally and downwards. As a result, when the core was rutting, the
material heaved upwards around the sides of the wheel rather than moving sideways.
This type of material movement is somewhat different from what would be expected
to happen in the field, where in the field, the pavement is not confined in the lateral or

3-94
downwards directions, and therefore when the pavement surface is loaded some
material gets pushed downwards into the subbase and some material moves sideways.
The heaving up of core material was an unavoidable result of the wheel tracking test
because, in order to successfully perform the wheel tracking test the cores needed to
be held steadily in place. In order to improve the material behaviour to mimic field
conditions the way in which a core is confined for the wheel tracking test needs to be
altered. This can be achieved by enclosing the core, particularly on the underside, in a
softer material such as a casing made of modelling clay, which would mimic the
subbase layer of a pavement and allow downward movement of the core material.

Another limitation that was encountered during laboratory testing was that there was
no standard test method that was available to measure the volume of air voids in the
chip seal cores and the standard test method for air voids measurement in asphalt
concrete cores, the test method ASTM D3203 – 05 (ASTM International, 2005b), had
to be adopted for this research. The results that were found by using this test method
may have been affected by some of the characteristics of the chip seal cores. For
example, the cores from Christchurch were not very thick which meant that the
accuracy of the bulk specific gravity measurements may have been affected. Also,
another difficulty that was encountered during the air voids test was that when
separating the chip seal core mixture to perform the vacuum suction test for
theoretical maximum specific gravity the high amounts of binder in the core mixtures
meant it was difficult to fully separate the aggregates in the core. In asphalt concrete
cores separating the aggregates is not a major problem because of the lower amount of
binder that is present in the mixture. As the cores that were tested for this research
were richer in binder compared to typical asphalt mixes, the high binder content
created problems with the test method. When performing the vacuum suction test it is
necessary to make sure that all air voids within a core mixture had been expelled, but
given that it was difficult to separate the chip seal core mixtures fully it was not
possible to know whether all air voids were expelled when the vacuum suction test
was performed.

3-95
3.10.2 Comparison of Observed Chip Seal Behaviour to Previous
Studies
Aggregate grading profiles that were determined for each wheel tracked core were
plotted against the amount of flushing that was present on the cores after wheel
tracking to determine whether the aggregate grading profile had any impact on the
flushing potential of a chip seal. Previous research conducted on this subject indicated
that there was a direct relationship between the proportion of aggregates in a seal that
would be classed as fines and the development of flushing, where it was theorised that
the fines reduced the air void volume in a chip seal layer which then caused flushing
(Ball and Patrick, 2008). However, the results that were obtained from this research
indicated that there was no clear difference in the fines content between cores that
showed high flushing and cores that showed low flushing. To investigate the effects
of fines further, the aggregate grading profiles were plotted against the deformation
patterns observed on the cores, and the results from this investigation revealed that in
general, the cores that continued to deform with loading (unstable cores) had a higher
proportion of fines than the cores on which deformation stabilised after a certain
amount of cyclic loading (stable cores). Additionally, earlier research had also
indicated that one of the causes contributing to flushing was the presence of high
bitumen-to-aggregate ratios (Gray and Hart, 2003), and this factor was explored in the
context of the presented research. The research analysis revealed that there was no
significant correlation between the quantity of flushing on a given chip seal and its
bitumen-to-aggregate ratio. Thus, based on the findings from the laboratory analysis
of the presented research it was concluded that the characteristics of the aggregate
profile of a chip seal only have an indirect contribution to flushing by affecting a chip
seal’s‎ stability,‎ which‎ in turn affects the likelihood of flushing recurring after
maintenance.

3.10.3 Availability of Chip Seal Samples for Testing

A limitation that was encountered during the sampling of field cores was the limited
number of sites that were available for sampling. Because of the damage to the
pavement caused by the extraction of cores, only limited sites that were being
rehabilitated were available for sampling. As the regions that were sampled did not

3-96
always have the same number of sites that were being rehabilitated, it was not
possible to sample a wide range of sites from each region and only two sites were
sampled per region. Due to the above reason it was also not possible to select sites
based on the chip seal characteristics, such as sites that had similar chip seal
thicknesses, number of chip seal layers, binder grades and aggregates, and traffic
volumes, and so it was not possible to control some of these factors which would have
affected the final results found from laboratory experiments. Despite this limitation,
the sites that were sampled covered a range of traffic volumes, seal types and binder
grades that are typically used on New Zealand pavements. Therefore, the results that
were obtained from the laboratory tests were representative of the behaviour of chip
seal pavements in the field. Furthermore, the laboratory test results provided valuable
insights into the behaviour of flushed chip seal pavements, particularly how the
deformation behaviour affected the flushing potential of a chip seal. These
understandings are extremely useful when trying to prevent flushing or when
assessing flushed chip seal pavements.

3.11 Conclusions
The present chapter examined the micromechanical interactions that take place
between chip seal layer materials in order to determine their effect on the occurrence
of flushing. The main objectives of the chapter and the conclusions that were reached
are summarised in Table ‎3.6.

Table ‎3.6 Main conclusions from laboratory testing

Objective Conclusions
Determine the effect of  The amount of flushing that was displayed on a wheel
loading on the flushing tracked core was influenced by the thickness of the
behaviour and the core.
deformation behaviour of  The amount of flushing on a core was independent of
chip seal structures. the amount of loading that was used on the core.
 Two distinct deformation patterns were observed. One

3-97
 was a continuous deformation pattern and the other was
a stable deformation pattern.
 The initial rate of deformation on a core during the
tracking test was influenced by the thickness of the
core.

Determine the  The continuous deformation pattern indicated an

relationship between unstable chip seal structure that would continue to
flushing and the deform and flush when subjected to loading.
deformation behaviour of  The stable deformation pattern indicated a chip seal
chip seal structures. structure that would reach a stable stone-on-stone state
after a certain amount of loading.
 The deformation pattern of a chip seal core could be
used to select appropriate maintenance methods for
flushed chip seal pavements. If flushing is present on a
surface that has reached a stone-on-stone state then
UHP water cutting can be used to treat the flushing and
further flushing would not develop. In contrast, if a chip
seal surface has a continuous deformation pattern then
rehabilitation is needed to remedy the flushing.
 Comparison of wheel tracking results on various cores
revealed that there were many differences in the way
that the cores behaved between different sites and
regions. These results explained the variation in
flushing that was observed at these sites, which is
typically affected by deformation.

Determine the effects of  There was no significant correlation between the

aggregate grading profile amount of flushing that was present on a core and its
and bitumen content of a bitumen-to-aggregate ratio.
chip seal structure on its  The aggregate grading profiles of the cores revealed
flushing potential. that the proportion of fines in a chip seal structure did
not affect its flushing potential.
 The fines content was higher on cores that displayed an

3-98
 unstable deformation pattern when compared to cores
that displayed a stable deformation pattern.
 The effect of fines on flushing development was
indirect. The fines affected the stability of the chip seal
structure, which in turn determined its flushing
potential.
 Aggregate shape analysis revealed no difference in the
angularity of aggregates between cores that displayed a
stable deformation pattern and cores that displayed an
unstable deformation pattern.

3.12 Recommendations for Further Investigations

One of the recommendations that has eventuated from this research is the need for
further investigations into the effect of aggregates on the behaviour of flush-prone
chip seal structures. The results obtained in this research showed that chip seals that
exhibited a stable stone-on-stone state in their deformation behaviour (stable cores)
tended to have a lower proportion of fines when compared to chip seals that exhibited
a continuous deformation behaviour (unstable cores). However, investigation into the
effect of aggregate angularity on the deformation behaviour revealed that no
difference in aggregate angularity existed between stable and unstable cores. The sites
that were sampled as part of this research consisted of similar aggregate types, and
therefore it was not possible to fully investigate the effects of angularity of different
types of aggregates on chip seal structure stability. Aside from inherent characteristics
of an aggregate, the angularity of an aggregate is affected by the ability of that
aggregate to resist wear and breakage when subjected to loading. In the presented
research no information was available about the original angularity characteristics of
the aggregates that made up the chip seal cores, and therefore it was not possible to
accurately investigate how much wear had occurred to the aggregates prior to being
cored. Further investigations should aim to investigate how the angularity of different
types of aggregates, for example steel slag aggregates, affected the stability and
flushing potential of chip seals. It is also recommended that the breakage-resistance of

3-99
aggregates in the tested cores is investigated to determine how the breakage-resistance
correlates to the observed flushing and deformation behaviours of the cores. To fully
investigate the aggregate breakage effects, full scale pavement test sections need to be
set up in the field to monitor the aggregate breakage over time and samples from these
seals need to be wheel tracked to determine how the flushing and deformation patterns
differ between different types of aggregates.

3-100
CHAPTER 4 - USE OF IMAGE
ANALYSIS TO QUANTIFY
CHIP SEAL MATERIAL
INTERACTIONS

4.1 Introduction
This chapter investigates the micromechanical changes that occur within chip seals in
order to determine how these changes affect flushing. The chapter presents the use of
the imaging technique of Computed Tomography (CT) scanning to quantify the
changes that occur to air voids in chip seals. The main objectives of the chapter were
to:
 Quantify the changes that occur to the volume and distribution of air voids
within a chip seal due to loading and determine how these changes affect the
flushing potential of a chip seal;
 Develop three dimensional models of the air voids structure of a chip seal
layer; and
 Investigate the effectiveness of using image analysis techniques to assess the
volumetric changes that occur in a chip seal layer.

The analysis presented in this chapter was conducted on the laboratory tested chip seal
cores from Chapter Three.

4-101
4.2 Use of Image Analysis in Pavement Material Modelling
Image analysis is the extraction of quantitative information from two-dimensional
(2D) and three-dimensional (3D) digital images. The technique is widely used in
medical imaging to reproduce images of internal body organs and other objects
(Masad and Sivakumar, 2004; Taniguchi et al., 2008; Tashman et al., 2007). Image
analysis combines x-ray imaging methods, such as CT scanning, with image analysis
techniques, such as area and volume calculations, to produce 3D models of the objects
under observation. When considering pavements and their constituent materials,
conventional methods of analysing the internal structure of a pavement, such as
volumetric measurements, rely on indirect measures to evaluate the changes that
occur within the material structure. These conventional methods can lack accuracy,
especially when considering the changes that occur to the distribution of air voids or
the orientation of aggregates within a pavement sample. Computed tomography
scanning is especially useful when evaluating the internal structure of pavement
materials as it is a non-destructive method, and this technique has been successfully
used to evaluate the distribution of air voids and the structure, orientation and shape of
aggregates within asphalt pavement layers (Masad et al., 1999a; Shashidhar, 1999;
Brzezicki and Kasperkiewicz, 1999; Frost, 1999). In the study by Masad et al.
(1999a), a CT scanner was used to obtain 2D images of asphalt concrete pavement
samples. The 2D images were used to construct 3D models of the sample, which were
then used to evaluate the volume and distribution of air voids between different
pavement mixtures.

Image analysis is useful in pavement research because the technique has the ability to
produce computer generated 3D images which can then be used to quantitatively
evaluate the internal structure of a material in its original state. Additionally, as the
images are computer-generated, it is possible to isolate different components in the
sample or material under observation, for example, in an asphalt concrete sample it is
possible to isolate the air voids or aggregates in order to investigate their orientation
and distribution. While there are some studies that have looked at this method of
pavement material analysis for asphalt concrete layers, currently there are no studies
which have utilised image analysis and CT scanning to examine the internal structure
characterisation of chip seal layers. The combination of image analysis and CT

4-102
scanning has much potential to provide valuable information about chip seal
structures, especially when looking at the origin of complex surface defects such as
flushing. The ability to isolate the constituents within a chip seal material can show
how the structures of well-performing seals differ from those that are performing
inadequately, which can in turn allow pavement practitioners to design better
pavement mixtures which perform well.

4.3 Computed Tomography Scanning of Cores

The wheel tracked chip seal cores that were obtained from flushed sites in Auckland,
Waikato, Christchurch and Dunedin regions were analysed using a CT scanner to
observe the distribution of air voids within the cores. The information obtained by
scanning the cores was used to investigate the differences that had occurred in the air
voids distribution at different loading levels. The CT scanner that was used in this
study was the SkyScan 1172, which is a desktop micro-CT scanner (Skyscan, 2010).
This CT scanner was equipped with a 100 kV x-ray source to scan the material under
observation and a 10-Megapixel CCD camera to capture layer images, which allowed
an image resolution down to 0.005 mm. The images were taken along the xy, xz, and
yz axes of the sample as shown in Figure ‎4.1a, and the images were processed using
the imaging software, CTAn (Skyscan, 2010), which produced horizontal layer
images of the sample at 0.035 mm distances apart along the depth of the sample. The
number of images that were taken per sample differed between the samples depending
on their thicknesses. Due to restrictions on the size of the samples that could be used
in the CT scanner, small samples having a diameter of 52 mm were extracted from the
wheel tracked cores. The samples were taken from two areas of the core as illustrated
in Figure ‎4.1b. One sample was taken from the centre of the core on the wheel tracked
area and the other sample was taken from the untracked area of the core. The sample
from the untracked area was used to represent the pre-tracked structure of the core and
the sample from the tracked area was used to represent the post-tracked structure of
the core. Samples were taken from all cores that were wheel tracked at each of the
different wheel pass cycles and were scanned. Figure ‎4.2 shows a sample that was
used for scanning.

4-103
 z
Core Tracked
area of core
2 1
y
xz
Samples for
yz CT scanning
x
Top view of a tracked core
xy

a) Scanning directions on a sample b) Location of sample extraction on wheel tracked

cores

Figure ‎4.1 Details of CT scanning

Figure ‎4.2 A sample that was used for scanning

4.3.1 Calculation of Area and Volume of Air Voids

The resulting scan images were analysed using the image analysis software ImageJ
(Rasband, 1997-2011), which is an open-source software with capability for quick
editing and processing of images. The editing and analysing capabilities of ImageJ
have seen it used regularly for radiological image processing. The images obtained
from the CT scans were entered into ImageJ in a bmp file format. Due to the high

4-104
density of the bitumen and aggregate mixture that made up the samples, the scan
images‎contained‎some‎“noise”,‎which‎was‎shown‎as‎graininess‎or‎white‎speckles‎in
the dark regions of the images. The presence of noise would have over-estimated the
air voids area in an image, leading to inaccurate air void area calculations. The noise
was reduced by applying a smoothing filter to each image, which ensured accurate
estimation of air voids. Furthermore, as the aggregates, bitumen and air voids were of
different densities the CT scanner produced images that had a range of grey
intensities, as shown in Figure ‎4.3a. The dark grey areas on the image in Figure ‎4.3a
indicate the aggregates, the light grey areas indicate the binder and the white areas
indicate the air voids. In order to perform area calculations, the images were
converted to binary format, which assigned either black or white to each pixel in the
image depending on its original grey intensity. Figure ‎4.3b shows a scan image in
binary format with the air voids visible in white and the binder and aggregates
coloured in black. Due to the smaller thicknesses of some of the samples, for example
26.3 mm, it was necessary to limit the portion of the samples on which air voids
analyses were conducted. A portion with a thickness of 19 mm, from the bottom
portion of each sample as illustrated in Figure ‎4.4, was analysed for each sample.

a) Original image of a sample b) Edited image of a sample in binary format

Figure ‎4.3 CT scan images of a pavement sample showing the original image
and the binary converted image

4-105
 52 mm

Analysed portion
of sample
19 mm

Figure ‎4.4 Illustration of air void volume analysis of samples

When converting the original images to binary format it was required to select the
threshold values that differentiated the air voids from the other material. The specific
threshold value for a core was determined using the scan images of the untracked
sample of that core. The initial threshold value to use was selected by visual
examination where the black and white threshold in the images was adjusted until
only the air voids were visible. The air voids distribution at this selected threshold
value was calculated and plotted on a graph against the thickness of the sample, as
shown in Figure ‎4.5. The approximate average value of air voids distribution obtained
from scan images, as indicated by the dotted line in Figure ‎4.5, was then compared to
the air voids value that was measured using the standard ASTM D3203 – 05 test
method (broken line in Figure ‎4.5). This process was repeated until the average value
of the air voids distribution from scan images were similar to the air voids value
obtained from the ASTM test method and the corresponding threshold value was used
for analysing the tracked sample. In the untracked sample shown in Figure ‎4.5 the
threshold value that corresponded to the shown air voids distribution was 8%, and this
threshold value was used for converting the tracked sample images.

Once the images were converted to binary format, the area recognition tool in ImageJ
was used to select the area of air voids (white area) on an image. This area was
calculated as a percentage of the total surface area of the sample, and this calculation
was performed for each scan image. The air void areas that were calculated from the
scan images were used to calculate the volume of air voids by multiplying the air void
area by the distance between two consecutive scan images. The scan results were then

4-106
used to quantitatively analyse the air voids distribution of a sample by plotting the
volume of air voids as a percentage along the depth of the sample, and the air void
volumes of the tracked and the untracked samples from each core were compared. As
the distance between the scan images were extremely small (0.035 mm) the method
that was used to calculate the air void volume would have given an accurate measure
of the actual air void volume of the sample. In the results that were obtained, attention
was given to the patterns in the clustering of the air voids within the samples.

Figure ‎4.5 Method used for selecting threshold values for binary conversion
of scan images of samples of Core 1 from Waikato site 1

When comparing the air voids percentage in the untracked sample from Waikato site
1 Core 1 calculated from CT scan images to the total air voids percentage measured
using ASTM D3203 – 05 test method (ASTM International, 2005b), one can see that
there is a difference in the amount of detail that is available from the two methods of
air voids calculation. The ASTM test method gives an average value of air voids
compared to the CT scanning results and fails to show how the air voids are
distributed inside the sample.

4-107
4.3.2 Three Dimensional Modelling of Air Voids
The purpose of developing 3D models of the scanned samples was to determine how
3D models of the air void distributions can help to assess the future behaviour of a
chip seal. The air void volume analysis aimed to relate the wheel tracking behaviour
of chip seal cores that was presented in Chapter Three to their air void distributions,
especially investigating the state of the air void distribution when a core reached a
stable deformation pattern. By examining the air void distribution of a stable chip seal
core it is possible to determine the cause of the stable state of the core, for example
whether the core had reached its stable state due to all the air voids being expelled
from its structure or whether it was due to the core actually reaching a stone-on-stone
state. If a particular core had reached a stable state due to all the air voids being
expelled, then it is clear that any further deformation will result in binder exuding to
the surface of the core and flushing, and when this situation arises the best method of
treatment would be to rehabilitate the pavement surface. On the other hand, if the
stable state of the chip seal was due to it reaching a stone-on-stone state and there
were some air voids present within the chip seal structure, then the only problem with
the chip seal structure is the presence of an excessive amount of bitumen that can
cause flushing.

The development of the 3D air void models was performed using the image modelling
software AMIRA version 5.4 (Visage Imaging and Konrad-Zuse-Zentrum fur
Informationstechnik Berlin, 2012). AMIRA is a software that is widely used in
medical imaging and it uses image processing and visualising techniques such as
image filtering, surface and volume rendering, and area and volume measurement to
construct 3D models from radiological images. The original scan images of each
sample were converted to tiff format and entered into AMIRA. The images were
filtered using a smoothing filter to remove noise and scatter and the smoothed images
were converted to binary format of black and white to distinguish the air void areas
from the binder and aggregates, similar to the process described in the earlier section.
AMIRA’s‎ surface‎ rendering‎ tools,‎ Isosurface and Voltex, were used to create a 3D
visualisation of each sample and its air voids structure.

4-108
4.4 Air Void Volumes and Three Dimensional
Reconstructions of Air Void Distributions
4.4.1 Waikato Samples
The scan images of the samples that were taken from Core 1 from Waikato site 1 are
shown in Figure ‎4.6. This core was wheel tracked using 5,000 wp. Figure ‎4.6a shows
a scan image of the sample that was taken from the tracked area of the core and Figure
‎4.6b shows a scan image of the sample that was taken from the untracked area of the
core. These scan images were taken at a distance of 5 mm from the bottom of the
sample. As can be seen, the untracked sample has more air void area compared to the
tracked sample. On all the cores that were wheel tracked the area that was tracked
displayed flushing on the core surface at the conclusion of the tracking test, and this
observation coupled with the air void visualisations shown in Figure ‎4.6 confirmed
the supposition that flushing is caused by the tighter compaction of chip seal materials
that reduces the air voids.

a) Tracked area sample b) Untracked area sample

Figure ‎4.6 CT scan images of the samples taken from Core 1 from Waikato
site 1 (5,000 wp)

4-109
Volume of Air Voids Calculated from CT Scan Images
The distribution of air void volumes of the tracked and untracked samples of Core 1
(5,000 wp) and Core 2 (10,000 wp) from Waikato site 1 are presented in Figure ‎4.7a
and Figure ‎4.7b. On Core 1 samples the volume of air voids in the untracked sample
was significantly higher when compared to the volume of air voids in the tracked
sample. This pattern in air voids volume validated the observations that were made
from the CT scan images of the samples (Figure ‎4.6), where in the untracked sample
there was much variation in the air void volume along the depth of the sample,
whereas in the tracked sample the air voids were consistently small along the depth of
the sample. The pattern of air voids that was observed in the untracked sample was
due to the multiple chip layers that made up the samples, where Waikato site 1 had
four to five layers of chips and the scan results clearly showed the likely locations of
layer interfaces between the different chip layers, which create extra void spaces in
the chip seal structure. When a chip seal structure is subjected to loading the chip
layers compact together and the voids in the chip layer interfaces are reduced. This
occurrence was demonstrated by the small variation in the air voids in the tracked
sample.‎ The‎ deformation‎ trend‎ for‎ Waikato‎ site‎ 1’s‎ Core‎ 1,‎ as‎ illustrated‎ in‎ Figure‎
3.15a, revealed that deformation (rutting) stabilised after approximately 2,000 wp,
which was an indication that the core had reached a stone-on-stone state. When
examining the air void volume plot for the tracked sample it can be seen that
particularly in the middle section of the sample (5 mm – 15 mm) the air voids had not
reached zero. This observation indicated that there was still some air voids left in the
sample even though there was flushing present on 70% of the wheel tracked area,
which was significant. This observation also confirmed the possibility that the core
had reached the stone-on-stone state.

The air voids pattern that was observed on samples from Core 2 (10,000 wp) was
similar to the pattern that was seen on Core 1 samples, where there was a wide
difference in the volume of air voids between the tracked sample and the untracked
sample. Core 2 also had a significant amount of flushing on its wheel tracked area
(75% flushing), and had reached a stabilised rutting pattern (from Figure 3.15a in
Chapter Three) which was suspected to be due to Core 2 reaching a stone-on-stone
state. The presence of some air voids in the tracked sample of Core 2 (Figure ‎4.7b)
indicated that this core had indeed reached a stone-on-stone state.
4-110
In addition to the air void volume plots, calculations were made to determine the
percentage of decrease in air void volume that had occurred along the depth of the
tracked sample and the untracked sample of Core 1 and Core 2 using the values
calculated from the CT scan images. In Core 1, the air void volume in the untracked
sample had decreased by 80% after being wheel tracked, and in Core 2, the air voids
in the untracked sample had decreased by 65%. The air void volume reduction values
were compared to the amount of flushing that was measured on the cores after being
wheel tracked in order to establish the relationship between air void volume reduction
and flushing. In Core 1 the air voids decrease calculated from the scan images was
similar to the area of flushing that was measured on the core surface after being wheel
tracked, while in Core 2 the decrease in air voids calculated from scan images was
less than the amount of flushing measured on the core. A full investigation of the
relationship between flushing measured on cores and the air voids reduction
calculated from scan images is presented in Section ‎4.5 of this chapter.

a) Core 1 – 5,000 wp

4-111
 b) Core 2 – 10,000 wp

Figure ‎4.7 Variation in air void volume along the depth of the tracked
samples and the untracked samples from Waikato site 1 cores

Three-Dimensional Models of the Samples

The 3D models of the untracked and tracked samples from Waikato site 1 Core 1 are
shown in Figure ‎4.8. The images in Figure ‎4.8a and Figure ‎4.8c show the 3D
reconstructions of the entire untracked sample and the tracked sample, and the images
in Figure ‎4.8b and Figure ‎4.8d show only the air voids structures of the untracked and
the tracked samples.

When comparing the air voids of the untracked sample to the tracked sample one can
see that there are less air voids in the tracked sample, which confirmed the findings
shown in the air voids volume analysis that was presented earlier. The untracked
sample had large clusters of air voids that were distributed near the middle of the
sample. In contrast, the tracked sample had smaller bubbles of air voids that were
distributed throughout the whole sample. The reason for the air voids size and
clustering in the untracked sample was due to the manner in which chip seals are
constructed where, when aggregates are laid to construct a chip seal surface, the
manner in which the individual aggregates fall on the pavement surface can create
4-112
 areas of inter-connected void spaces, and these air voids would appear as large
clusters in the CT scans. When the chip seal cores were wheel tracked, most of the air
voids would have been expelled, leaving only small areas of air voids trapped within
the chip seal structure. Additionally,‎ the‎ tracked‎ sample’s‎ air‎ voids‎ graph‎ in‎ Figure
‎4.7a showed that there was a higher amount of air voids near the bottom of the tracked
sample than when compared to the top and the middle of the sample, and this
observation‎was‎confirmed‎by‎the‎3D‎model‎of‎the‎tracked‎sample’s‎air‎voids‎shown‎
in Figure ‎4.8d.

19 mm
19 mm

52 mm
52 mm

a) 3D reconstruction of untracked sample b) Air voids structure of untracked sample

19 mm
19 mm

52 mm

52 mm

c) 3D reconstruction of tracked sample d) Air voids structure of tracked sample

Figure ‎4.8 3D reconstruction of the untracked sample and the tracked sample
of Core 1 from Waikato site 1

4-113
 The air voids distributions within the untracked sample and the tracked sample from
Core 2 of Waikato site 1 are shown in Figure ‎4.9.‎In‎Core‎2’s‎tracked‎sample‎the‎air‎
voids were not simply placed at the bottom part of the sample, rather they were spread
throughout the depth of the sample. This distribution of air voids aligned with the
result shown in Figure ‎4.7b.

19 mm 19 mm

52 mm 52 mm

a) Untracked sample b) Tracked sample

Figure ‎4.9 3D reconstruction of the air voids distribution in the untracked

sample and the tracked sample of Core 2 from Waikato site 1

The air voids analysis for Core 3 (20,000 wp) from Waikato site 1 is shown in Figure
‎4.10. In this core the air voids in the tracked sample was nearly zero throughout the
sample, except in the region between 5 mm and 10 mm. Given that Core 3 had 80% of
the wheel tracked area flushed and the core had reached a stable rutting state close to
the end of the wheel tracking test, the nearly minimal air voids in the tracked sample
confirmed that the stable rutting state that was reached was due to the continuous
compaction of the seal and the subsequent flushing was due to a large proportion of
the air voids being expelled.

4-114
 Figure ‎4.10 Volume of air voids along the depth of the tracked sample and the
untracked sample of Core 3 from Waikato site 1

The 3D reconstructions of air voids in the untracked sample and the tracked sample of
Core 3 from Waikato site 1 are shown in Figure ‎4.11. As can be seen, the air voids in
these samples were low, especially in the untracked sample when compared to the
untracked samples from Core 1 and Core 2. The tracked sample of Core 3 had a small
collection of air voids towards the bottom of the sample, as was indicated by the
graph in Figure ‎4.10. It is likely that the position where this air voids collection was
located was an aggregate layer interface.

19 mm 19 mm

52 mm 52 mm

a) Untracked sample b) Tracked sample

Figure ‎4.11 3D reconstruction of the air voids distribution in the untracked

sample and the tracked sample of Core 3 from Waikato site 1
4-115
4.4.2 Auckland Samples
The air void volume analyses for Core 2 and Core 3 from Auckland site 1 are shown
in Figure ‎4.12. The air void volume in the samples from Core 2 varied along the depth
of the sample especially in the untracked sample. In contrast, the variation in the air
void volume along the depth is low in the samples from Core 3. After being wheel
tracked Core 2 and Core 3 displayed flushing on 90% of the wheel tracked area, and
as can be seen in Figure ‎4.12 the air void volumes in the tracked samples of both
cores were low, which explained the high quantity of flushing that was observed.

The 3D reconstructions of the air void distribution plots of the tracked sample and the
untracked sample of Core 2 from Auckland site 1 are shown in Figure ‎4.13. There
were large clusters of air voids present within the untracked sample. The air void
clusters in the tracked sample were smaller in size. The positions where the air voids
were clustered in the tracked sample and the untracked sample indicated the locations
of aggregate layer interfaces. The 3D reconstructions of the air voids distribution plots
of the tracked sample and the untracked sample of Core 3 from Auckland site 1 are
shown in Figure ‎4.14. In the tracked sample (Figure ‎4.14b) the air voids were
significantly low and they were clustered in the bottom part of the sample. In general,
the air void clusters in the samples from Core 3 were smaller in size when compared
to the air void clusters in Core 2. When examining the air voids distributions of Core
2 and Core 3 from Auckland site 1 it was evident that the air void structures of the
Auckland cores were similar to the air void structures of the Waikato cores where the
air voids had formed large clusters at aggregate layer interfaces.

The air void volume plots and 3D reconstructions of air void structures of cores from
Auckland site 2 are presented in Appendix D and F.

4-116
 a) Core 2 – 5,000 wp

b) Core 3 – 10,000 wp

Figure ‎4.12 Volume of air voids along the depth of the tracked samples and the
untracked samples of cores from Auckland site 1

4-117
19 mm 19 mm

52 mm 52 mm

a) Untracked sample b) Tracked sample

Figure ‎4.13 3D reconstruction of air voids distribution in the untracked sample

and the tracked sample of Core 2 from Auckland site 1

19 mm 19 mm

52 mm 52 mm

a) Untracked sample b) Tracked sample

Figure ‎4.14 3D reconstruction of air voids distribution in the untracked sample

and the tracked sample of Core 3 from Auckland site 1

4.4.3 Dunedin Samples

The air void volume analyses for Core 1, Core 2 and Core 3 from Dunedin site 1 are
shown in Figure ‎4.15. These three cores had flushing areas of 15%, 35% and 75%
after wheel tracking, and when comparing these flushed areas to the air voids
distributions that are plotted on Figure ‎4.15 it can be seen that the amount of air voids
left in the tracked samples decreased progressively, which correlated with the
amounts of flushing that were measured on the cores after wheel tracking.

4-118
A point of difference that was present on the air voids distributions of these Dunedin
cores when compared to the cores from Waikato site 1 was the variation in the air
voids along the depth of the cores. On Waikato site 1 cores, the air voids distributions
of the untracked and the tracked samples exhibited clusters of air voids along the
depth of the samples. This pattern of air void clustering allowed identification of the
locations of aggregate layer interfaces. In contrast, on Dunedin site 1 cores the air
voids distributions in the tracked samples were consistent throughout the depth of the
samples and it was difficult to identify the locations of aggregate layer interfaces.

a) Core 1 – 2,500 wp

4-119
 b) Core 2 – 5,000 wp

c) Core 3 – 10,000 wp

Figure ‎4.15 Volume of air voids along the depth of the tracked samples and the
untracked samples of cores from Dunedin site 1

The 3D reconstructions of the air voids distributions of samples from Core 1, Core 2
and Core 3 of Dunedin site 1 are shown in Figure ‎4.16, Figure ‎4.17 and Figure ‎4.18.
A pattern that was observed on these samples was that the air voids were smaller in
4-120
 size when compared to the samples from Waikato site 1, and the air voids were spread
throughout the samples. The air void area plots and air void volume plots of samples
from Dunedin site 2 are shown in Appendix C and Appendix D respectively. The 3D
reconstructions of the samples and the air voids distributions of Dunedin site 2
samples are shown in Appendix E and Appendix F.

19 mm 19 mm

52 mm 52 mm

a) Untracked sample b) Tracked sample

Figure ‎4.16 3D reconstruction of the air voids distribution in the untracked

sample and the tracked sample of Core 1 from Dunedin site 1

19 mm
19 mm

52 mm 52 mm

a) Untracked sample b) Tracked sample

Figure ‎4.17 3D reconstruction of the air voids distribution in the untracked

sample and the tracked sample of Core 2 from Dunedin site 1

4-121
19 mm 19 mm

52 mm 52 mm

a) Untracked sample b) Tracked sample

Figure ‎4.18 3D reconstruction of the air voids distribution in the untracked

sample and the tracked sample of Core 3 from Dunedin site 1

4.4.4 Christchurch Samples

The air voids analyses for Core 1, Core 2 and Core 3 from Christchurch site 2 are
shown in Figure ‎4.19. The amount of flushing that was displayed on these cores at the
end of the wheel tracking test were 10%, 20% and 60% respectively. The reduction in
air void volume calculated from scan images of Core 1, Core 2 and Core 3 correlated
well with the measured flushing, with the scan image analysis indicating air void
volume reductions of 20%, 20% and 55% respectively on the three cores. The air
voids distributions that were calculated in the tracked samples and the untracked
samples from the three cores did not have a significant variation along the thickness
of the samples and as a result, in these samples it was difficult to identify the locations
of the aggregate layer interfaces. In all the samples the air void volume was low at the
bottom of the samples, however, the air void volume increased towards the top of the
samples.

The deformation patterns that were observed on the three cores from Christchurch site
2 showed continuous deformation on all three cores. As the volumes of air voids left
in the samples from Christchurch site 2 cores (Figure ‎4.19) were limited even at
relatively low levels of flushing, it was predicted that if these cores were to reach a
stable deformation state then that would be due to all the air voids being expelled
from within the chip seal structures, and in which case Christchurch site 2 will need to
be rehabilitated to remedy the subsequent flushing.

4-122
The air void area plots for Christchurch site 1 and Christchurch site 2 samples are
shown in Appendix C. The air void volume plots for samples from Christchurch site 1
are shown in Appendix D. Due to significant scatter that was present in the scan
images of the Christchurch site 1 samples and Christchurch site 2 samples, it was not
possible to develop clear 3D reconstructions of the air voids distributions of these
samples.

a) Core 1 – 2,500 wp

4-123
 b) Core 2 – 5,000 wp

c) Core 3 – 10,000 wp

Figure ‎4.19 Volume of air voids along the depth of the tracked samples and the
untracked samples of cores from Christchurch site 2

4-124
4.5 Relationship of Air Voids Reduction to Flushing
Trends
The purpose of conducting analysis on the air void areas and the air void volumes of
samples was to determine how the changes to the air voids distribution of a chip seal
sample related to the quantity of flushing that would develop on its surface. The air
void volumes that were calculated by analysing the CT scan images of the samples
were used to establish the correlation between the scan-determined air void volumes
and the quantity of flushing that was measured on the chip seal cores at the end of the
wheel tracking tests. For each core, the air void volume that was calculated from the
scan images in the untracked sample was compared to the air void volume of the
tracked sample, and the reduction in air void volume that had occurred between the
untracked sample and the tracked sample was calculated. This air void volume
reduction was determined for each point along the depth of the untracked sample and
the tracked sample and the calculated values were averaged to determine the average
reduction in air void volume. The calculated air void volume reduction value for each
core was then compared to the quantity of flushing that was measured on that core.
Table ‎4.1 shows the air void volume reductions that were calculated from the scan
images. The correlation between the measured quantity of flushing and the calculated
reduction in air void volume is shown in Figure ‎4.20.

Understanding of chip seal behaviour and knowledge from previous studies that were
documented in Chapter Three indicate that the volume of air voids in a chip seal plays
a significant role in its volumetric balance, and is particularly important for the
occurrence of flushing. A lack of air void volume in a chip seal structure affects the
movement of bitumen when subjected to loading, which in turn leads to flushing. As
can be seen in Figure ‎4.20, there is a strong correlation (R2 = 0.820) between the
quantity of flushing that was measured on a core after being wheel tracked and the
reduction in air void volume that was calculated from the CT scan images of the
samples of the corresponding core. This correlation aligned with the supposition that
flushing was caused by a reduction in the volume of air voids. As the correlation
between the observed quantity of flushing and the calculated air void reduction was
strong, this correlation can be used for determining the amount of flushing that would
develop on a given chip seal core.
4-125
 Table ‎4.1 Reduction in air void volumes calculated from CT scan images

Quantity of flushing Reduction in air void

No. of wheel
Region Site Core ID measured after wheel volume calculated from
passes (wp)
tracking test (%) CT scan images (%)
Core 2 5,000 90 70
1
Auckland Core 3 10,000 90 55

2 Core 2 5,000 80 70

Core 1 5,000 70 80

Waikato 1 Core 2 10,000 75 65

Core 3 20,000 80 75

Core 1 2,500 15 15

1 Core 2 5,000 35 45

Core 3 10,000 75 90
Dunedin
Core 1 2,500 95 80

2 Core 2 5,000 100 80

Core 3 10,000 95 80

1 Core 1 2,500 25 35

Core 1 2,500 10 20
Christchurch
2 Core 2 5,000 20 20

Core 3 10,000 60 55

A point to note regarding the information presented in Figure ‎4.20 is that the data set
that was available to investigate the flushing/air void reduction correlation was
limited, having data from only 16 chip seal cores. The seven remaining cores did not
have any samples that were suitable for scanning as some difficulties were
encountered during the sample extraction process on these cores. These difficulties
are explained in the discussion section of this chapter. Despite the limited size of the
data set, the strength of the correlation between the quantity of flushing and the air
voids reduction cannot be overlooked. The presented method of utilising calculations
of air void volume reduction to predict the quantity of flushing of a chip seal surface
is recommended for use by pavement management authorities when conducting in-
depth assessment of flushed chip seal pavements. Additionally, it should also be noted
that the air void volume calculations that were conducted on the CT scanned samples
4-126
did not include the interstitial spaces between the aggregates on the surface of the
samples. In the presented research, the air void volume investigation of each sample
was concentrated on the portion of the sample that was below the top surface. The
practicalities of the wheel tracking test and the sizes of the scanned samples prevented
the inclusion of the voids of the interstitial spaces in the air void volume analysis,
where the smaller size of the samples that were taken for scanning (52 mm diameter)
meant it was difficult to accurately estimate the volume of the surface texture of the
samples. Hence, the volumes of the interstitial spaces of the samples were not
included in the air void volume analyses. It was recognised that the exclusion of the
interstitial space volumes in the air void volume analysis was a limitation of the CT
scanning method. It is expected that the availability of a scanner that has the
capability to scan larger samples would overcome the above recognised limitation, as
the surface texture of larger samples could be estimated more accurately using
methods such as sand circle testing (Gransberg, 2007), which would allow the
inclusion of the interstitial space volume in air void volume estimations.

Figure ‎4.20 Correlation between flushing measured after wheel tracking tests
and the air void volume reduction calculated from CT scan images

4-127
4.5.1 Inter-Relationship between Binder Volume, Air Void Volume
and Flushing
In order to establish the cause of air void volume reduction within the cores, analysis
was conducted to investigate the inter-relationship between flushing, air void volume
and binder volume. For each core, the ratio between the binder volume and the air
void volume that was available after being wheel tracked (bitumen-to-air void volume
ratio) was calculated, and this value was plotted against the amount of flushing that
was present on the core after being wheel tracked. The results are shown in Figure
‎4.21. As can be seen in Figure ‎4.21, there is a strong correlation between flushing and
the bitumen-to-air void volume ratio (R2 = 0.633). This result confirmed that the air
void volume reduction that causes flushing occurred due to the chip seal surface
having an excessive quantity of bitumen in comparison to its available volume of air
voids. The result shown in Figure ‎4.21 aligned with findings from past research that
had investigated the behaviour of asphalt concrete pavements. Past research had
indicated that the primary reason for the appearance of bitumen on a pavement surface
was the lack of availability of air void volume for bitumen movement (Krishnan and
Rajagopal, 2005; Wang et al., 2004; Buttlar et al., 1999). Therefore, it can be
concluded that flushing occurs as a direct result of a chip seal surface having a high
bitumen-to-air void volume ratio.

The understanding of the relationship between flushing and the ratio of bitumen-to-air
void volume of a chip seal has immense value when selecting maintenance treatments
for flushed chip seal pavements. Presently, the most common method of treating
flushing is by resurfacing using a new chip seal layer. Resurfacing is an inexpensive
treatment that restores the texture of a pavement surface immediately. The drawback
of resurfacing is that it adds more bitumen to the existing flushed surface. Well-
designed chip seal layers usually have a sufficient volume of air voids to allow
compaction without flushing. However, when a new chip seal layer is used for
resurfacing an existing flushed surface, the air void volume provided by the new chip
seal is not sufficient for satisfactory performance of the entire chip seal structure even
if the new chip seal layer was well-designed. Therefore, it is only a matter of time
before flushing re-develops on the pavement. The findings from the presented air void
volume analyses highlighted the importance of using maintenance treatments such as

4-128
UHP water cutting or rehabilitation on flushed surfaces in place of resurfacing. Water
cutting has been shown to effectively improve the macrotexture of flushed sites
(Gransberg and Pidwerbesky, 2007), and the advantage that water cutting has over
resurfacing is that it does not add excess bitumen or thickness to an existing flushed
surface, which in turn reduces the likelihood of flushing recurring as a result of
excessive amounts of bitumen or increased thickness. Rehabilitation also has the same
advantages where it prevents a build up of chip seal layers that results in a structure
that has a high bitumen-to-air void volume ratio. Water cutting and rehabilitation may
have high initial cost. Nevertheless, the long-term benefits gained by using these
treatments are far greater than resurfacing.

Figure ‎4.21 Correlation between flushing and the ratio of bitumen volume to
air voids volume

4-129
4.6 Discussion
The presented chapter investigated the micromechanical changes that occurred within
chip seal samples in order to quantify how those changes affected the development of
flushing. The micromechanical changes were investigated using a combination of CT
scanning and image analysis techniques, which was a novel method of examining the
internal structure of chip seal samples. The combination of CT scanning and image
analysis has not been used previously for chip seal analysis and it was a unique
component of the research presented in this thesis. The following sections describe
some of the positive features of CT scanning and image analysis as well as some of
the issues that were faced when conducting the presented analysis.

4.6.1 Effectiveness of Computed Tomography Scanning

Samples that were taken from wheel tracked cores were scanned using a CT scanner
in order to analyse their internal structure, particularly examining the state of air void
volume and distribution within the samples. The obtained scan images were then
analysed using image analysis techniques to determine the volume of air voids within
the samples and to develop 3D models of the air voids distributions. The images that
were obtained by CT scanning were effective at showing the arrangement and the
distribution of air voids within a sample. When the air void volume and distribution
within tracked-area samples were compared to their corresponding untracked-area
samples, there were notable reductions in the air void volumes in the tracked-area
samples. This reduction in air void volume confirmed the supposition that air voids
within a chip seal structure were remarkably reduced when loaded, thus making the
chip seal prone to flushing.

Computed tomography scanning provided a visual representation of the undisturbed

state of the tracked and the untracked samples, which was an extremely effective
method of analysing a chip seal structure. Typically, analysing the volumetric
properties of a pavement mixture requires the use of destructive methods, such as the
standard air voids measurement tests set out by the American Society for Testing and
Materials (ASTM), which fail to show the in-situ state of a pavement sample. In
contrast, CT scanning preserved the in-situ state of a chip seal surface allowing

4-130
determination of how the size and the placement of air voids affected flushing.
Additionally, the method of analysing pavement samples using CT scanning is useful
as this method has the capability to isolate certain areas of a sample in order to
perform in-depth analysis. The CT scanner that was used in the presented research
was able to produce images at 0.035 mm apart, which provided detailed information
about the position and placement of the air voids and aggregates, and the likely
locations of aggregate layer interfaces. By having detailed information about the
internal structure of a chip seal it was possible to examine the differences in air void
volumes between tracked and untracked states of a chip seal sample more accurately
than conventional test methods.

From the results obtained in this research it was evident that CT scanning is a valuable
analysis tool that can greatly aid with assessing flush-prone chip seal surfaces. The
combination of CT scanning and laboratory-based wheel tracking can provide more
detailed information about the internal structure of a chip seal surface than when
conventional empirical test methods are used. A detailed discussion of the practical
applications of CT scanning in detecting flushing of chip seal surfaces is presented in
Chapter 6.

4.6.2 Limitations of Equipment

While many positive aspects of the CT scanning method were found, this analysis
method had some limitations, particularly when considering the size of the samples
that were able to be scanned. The CT scanner that was available for the presented
research was a micro-CT scanner which was designed for scanning items that were
much smaller than the cores that were taken from pavements. This meant that the
scanner could only accommodate samples with a maximum diameter of 65 mm and as
a result, it was not possible to scan each chip seal core completely and smaller
samples had to be taken. This limited the amount of information that could be
gathered about a chip seal core as it was not possible to obtain a complete picture of
the air voids distribution of each core. The smaller size of the samples may be a
source of errors in the test results as the samples were only 52 mm in diameter, which
accounted for a small proportion of the full chip seal core. As the objective of the

4-131
presented research was to investigate the changes that were occurring to the air voids
volume and distribution during loading, it was thought that the small size of the
samples did not have a significant impact on the results. However, if a similar
scanning procedure was to be used in the future for assessing pavement mixes, it is
recommended that a large number of samples are used to confirm the applicability of
the results obtained from scanning. Additionally, using a CT scanner that could
accommodate larger objects would be ideal as it would allow scanning of entire cores
without the need for small samples.

Another challenge that was encountered while performing CT scanning was that due
to the density of the chip seal samples it was difficult to obtain clear images of the
samples. The CT scanner that was used in this research was designed for scanning
materials that were much lower in density than pavement samples, such as bones,
muscles, tissue and pieces of food. The mixture of binder and aggregates in the chip
seal samples, especially in the samples that were taken from the tracked area of cores,
was high in density and this meant the quality of the scan images was reduced, and a
lot‎ of‎ “noise”‎ (seen‎ as‎ graininess)‎ was‎ present‎ in‎ the‎ images.‎ When image analysis
was conducted on the scan images their quality was improved to some extent by
applying smoothing filters to the images. The application of smoothing filters needed
to be done carefully as over-smoothing had the potential to remove important
information from the images. If CT scanning methods similar to the ones presented in
this chapter are to be used in the future for pavement mix assessments, it is
recommended that a larger scanner with capability to scan dense material, especially
pavement cores, is used as it will make it easier to analyse the resulting images.

4.6.3 Difficulties with Sample Extraction

Some issues were encountered while extracting the samples from the wheel tracked
cores. As the samples needed to be less than 65 mm to be able to fit into the CT
scanner, a drill with an inside diameter of 52 mm was used to extract the samples. In
most of the cores the sample extraction process was performed without a problem.
However, in some cores, particularly the cores from Christchurch, it was difficult to
extract samples that were suitable for CT scanning. The biggest problem when drilling

4-132
samples from Christchurch cores was that the bitumen/aggregate mix that formed the
cores was not rigid enough to withstand the pressure caused by the drill, and the
samples crumbled when attempting to remove them from the cores. This meant that
some of the samples that were drilled were not in a state to be scanned, and therefore
these samples were discarded. As there was limited area on the cores to extract the
samples, especially on the tracked area, replacement samples could not be taken. As a
result, some of the cores did not have scan images for analysis. The issue with sample
disintegration was unexpected and unavoidable as it depended on the mix
characteristics of individual cores and it was not possible to predict which samples
would crumble. To overcome such issues in the future, the sample extraction process
could be changed, for example, using a diamond saw to trim square sections from the
tracked and untracked areas of a core would minimise sample disintegration.

Another issue that was encountered during the sample extraction process was that as
the drill was drilling through the chip seal cores, the heat that was generated due to
friction between the core material and the drill surface caused some of the bitumen on
the outside surface of the samples to soften. It was important to make sure the drilling
was done at a slow speed to minimise softening of the samples. Furthermore, some of
the chip seal cores, particularly Waikato and Auckland cores, were only
approximately 200 mm in diameter which meant extracting two samples that were
side-by-side was difficult. In some of the Waikato and Auckland cores when the
tracked area sample was extracted, the drilling process disturbed the untracked area of
the core which meant in these cores it was not possible to take a sample from the
untracked area, and as a result analysis of the air void volumes could not conducted.
This issue could also be overcome by using a diamond saw to trim sections of the chip
seal core, as described above. As the diamond saw is sharper than a drill, it would
have minimal disturbance on the core structure.

4.7 Conclusions
This chapter documented the image analysis that was carried out on wheel tracked
cores to examine the volumetric changes that occurred within chip seal structures and

4-133
to determine how those changes impacted on the development of flushing. The main
conclusions that were reached from the analysis are shown in Table ‎4.2.

Table ‎4.2 Main conclusions of the study

Objective Conclusions
Objective One:  A strong correlation existed between flushing and the
Quantify the changes percentage reduction in air void volume calculated from
that occur to the volume CT scan images of the cores. This correlation confirmed
and distribution of air the supposition that flushing was caused by a reduction in
voids within a chip seal air voids.
due to loading and  A strong correlation existed between flushing and the
determine how these bitumen-to-air void volume ratio of a core. This
changes affect the correlation confirmed that the air void volume reduction
flushing potential of a that causes flushing occurred due to the chip seal core
chip seal. having an excessive amount of bitumen in comparison to
the available volume of air voids.
 Air void volume reduction in a chip seal sample can be
used to determine the amount of flushing that would
develop on a site.
 The air void volume reduction in wheel tracked samples
followed a pattern similar to their pre-tracking
conditions, with more air voids left at the aggregate layer
interfaces.
 Air void volume calculations made from CT scan images
can be used to confirm the state of stability of a chip seal
and to determine cause of the stability, for example
whether a chip seal layer has reached a stable state due to
reaching the stone-on-stone state or whether it is due to
all the air voids being expelled.

4-134
Objective Two:  Three-dimensional models of air void structures
Develop three demonstrated the size of air voids and how they were
dimensional models of distributed within the samples. The 3D models also
the air voids structure of showed the differences in air voids characteristics of
a chip seal layer. samples from different regions.
 Plots of air void volume calculated from CT scan images
showed the likely locations of aggregate layer interfaces
between different chip seal layers in a sample.

Objective Three:  The combination of CT scanning and image analysis was

Investigate the an extremely effective method that provided information
effectiveness of using about a chip seal’s‎internal‎structure.
image analysis  The non-destructive nature of CT scanning provided
techniques to assess the valuable‎ information‎ about‎ a‎ chip‎ seal’s‎ undisturbed‎
volumetric changes that state, which allowed identification of pavements or
occur in a chip seal pavement mixtures that were likely to flush.
layer.  Computed tomography scanning provided detailed
information about the air voids structure within a chip
seal sample. Computed tomography scanning showed the
variation in air void volume within a sample whereas the
standard test method for air void measurement only
showed the average air void volume.

4.8 Recommendations and Further Research

The results presented in this chapter demonstrated the use of CT scanning and image
analysis techniques to examine the internal structure of chip seal cores and to utilise
the information that can be gained through analysis of scan images to predict the
flushing potential and stability of chip seals. The analysis presented in this chapter
was based on chip seal cores that were taken from four regions in New Zealand. The
cores were subjected to laboratory-based wheel tracking prior to examining the
changes that had occurred to the volume and distribution of air voids within the cores.

4-135
From the analysis a flushing prediction method was identified which allowed
determination of the quantity of flushing that was likely to develop on a chip seal
surface based on the reduction in its air void volume. This method of using air void
volume reduction for predicting flushing quantity would be extremely useful when
conducting in-depth assessment of chip seal pavements that were prone to flushing.
While the correlation that was found between flushing quantity and air void volume
reduction was strong, it is necessary to test this correlation by conducting analysis on
field-trafficked chip seal pavements. In the laboratory-tested cores the movement of
chip seal material in the lateral direction was controlled due to the mould that was
used in the wheel tracking test and this may have impacted on the way in which the
air voids were expelled. In the field lateral constrictions will not be present and
therefore, it is necessary to establish the comparability of air voids reduction in the
laboratory and in the field. It is recommended that in-service flushed chip seal
pavement sections are monitored periodically and analysed using CT scanning
methods to determine how the air void volume reduction correlates with flushing
development. Additionally, accelerated pavement testing could be conducted on chip
seal pavement sections to simulate flushing and samples taken from these pavements
could be used to confirm the findings presented in this thesis.

4-136
CHAPTER 5 - DEVELOPMENT
OF A FLUSHING PREDICTION
MODEL

5.1 Introduction
This chapter presents the outcomes from data analysis carried out to investigate the
relationships between flushing and variables known to cause flushing. The chapter
also presents the development of a multi-stage model to predict flushing. The primary
goal of the flushing modelling process that was undertaken in this research was to
develop a data-driven prediction model based on Pavement Management Systems
(PMS) data that have been‎ collected‎ through‎ New‎ Zealand’s‎ Long-Term Pavement
Performance (LTPP) programme. The data analysis and the model development were
performed by analysing pavement condition data from the state highway and the local
authority LTPP databases. The objectives of the presented data analysis and model
development were to:
 Investigate the relationship between flushing and variables known to cause
flushing;
 Identify the combination of factors that provide the best indication of flushing
potential of a chip seal surface;
 Determine a threshold to identify when a pavement surface is flushed; and
 Develop a data-driven model to predict the initiation and progression of
flushing on a chip seal pavement.

5-137
5.2 Data Set Description
5.2.1 The Long-Term Pavement Performance Data Set
The New Zealand LTPP programme records highly accurate pavement condition data
from selected state highway and local authority pavement sections in New Zealand.
Data from the state highway and local authority LTPP programme databases were
selected to be analysed to determine the effects of variables known to cause flushing.
The LTPP data set comprised of pavement condition data from 58 pavement sections
from the state highway LTPP programme and 82 pavement sections from the local
authority LTPP programme. The data on the state highway pavement sections have
been collected annually since year 2001 and the local authority pavement data have
been collected annually since year 2003. The analysis that is presented in this chapter
was performed on state highway LTPP data from nine survey years (from 2001 to
2011) and on local authority LTPP data from seven survey years (from 2003 to 2011).

The calibration sections that were included in the LTPP database were 300 metres in
length and each section was divided into six 50-metre subsections. The LTPP
database recorded flushing on a section by measuring the area percentage of the
pavement surface that was displaying flushing. Flushing was recorded in three
severity categories: low; moderate/medium; and high. The low flushing severity
category consisted of sections that had separate spots of flushing that were distributed
sparsely on the chip seal surface, as shown in Figure ‎5.1a. The moderate or medium
flushing severity category consisted of sections that had flushing spots joined together
and the aggregates visible on the chip seal surface, as shown in Figure ‎5.1b. The high
flushing severity category consisted of sections where the aggregates were completely
covered by bitumen, as shown in Figure ‎5.1c.

5-138
 a) Low flushing b) Medium flushing c) High flushing

Figure ‎5.1 Illustration of flushing ratings used in the LTPP database

Table ‎5.1 shows the pavement condition measures (variables) that were recorded in
the LTPP database and were investigated as part of the data analysis. These variables
had been identified to contribute to flushing from previous research (Alderson, 2008;
Ball et al., 1999; Ball and Patrick, 1998; Lawson and Senadheera, 2009; Park, 2007;
Weissmann and Martino, 2009). The variables were divided into three categories:
pavement composition characteristics; traffic-related factors; and the effects of other
pavement defects.

Table ‎5.1 Factors likely to affect flushing

Variable Description Type

Pavement composition characteristics:

Surface age (years) Age of chip seal layer. Discrete

Surface thickness (mm) Full thickness of the surfacing layer which has Continuous
resulted from multiple resealing.
Grade of aggregates Grade of aggregates that makes up the chip seal. Discrete

Number of chip seal layers The number of chip seal layers that are in the Discrete
pavement resulting from multiple resealing.
Material type Material type of the surfacing layer, eg. Two- Discrete
coat seal, void fill seal etc.
Polymer Modified Binder (PMB) The presence of PMB in the chip seal surfacing Binary
layer (0 - no, 1 - yes).

5-139
Traffic-related factors:
Number of Vehicles Per Day (vpd) Number of vehicles as measured by Average Discrete
Annual Daily Traffic (AADT).
Percentage of Heavy Commercial Percentage of vehicles (from total AADT) that Discrete
Vehicles (HCVs) (%) are classified as HCVs.

Climatic factors:
Temperature (°C) Average air temperature. Continuous

Rainfall (mm) Average yearly rainfall. Continuous

Humidity (Hpa) Mean vapour pressure. Continuous

Pavement defects:
Rutting (mm) Depth of rutting on the pavement surface. Continuous

Roughness (m/km) Surface roughness, measured by IRI. Continuous

5.2.2 Study Methodology

The method that was followed in the data analysis of LTPP data was as follows:
1. The first step of the data analysis process was to perform factor analysis to
determine the factors (independent variables) that had an effect on flushing.
Correlation matrices were created to identify factors (from Table ‎5.1) most
suitable for correlations analysis.
2. Independent variables were plotted against flushing (dependant variable) using
scatter plot graphs. This task was carried out using SPSS Statistics 19 (IBM
Corporation, 2010). The strengths of the relationship between flushing and
each of the identified factors were analysed using Pearson correlation and one-
way ANOVA. The correlation coefficient, R2 values and standard errors were
calculated.
3. Partial correlations were carried out on the data sets to investigate the
combined effects of independent variables on flushing. This procedure was
intended to reveal inter-relationships between factors that caused flushing.
4. Regression analysis was then performed on the factors identified from above
steps to determine their strength in predicting flushing. In the presented

5-140
 chapter, flushing was modelled in two phases: one model was used to predict
whether flushing had initiated on a pavement, and a second model was used to
predict the progression of flushing.

5.3 Data Analysis Results

The following section presents results from factor analysis of variables known to
cause flushing as identified from Table ‎5.1. Correlation matrices were used to identify
the variables that were most suited to be included in the correlation analysis. The
classification that was used to assess the strength of the correlations is shown in
Appendix G.

5.3.1 Factor Analysis of Variables Causing Flushing

Effects of Pavement Characteristics, Traffic-Related Factors and Other
Pavement Defects
Factor analysis was performed to identify the independent variables that were most
likely to impact on flushing and to identify multicollinearity between variables in
order to select the most suitable variables for detailed analysis. The pavement
characteristics that were included in factor analysis were age of the surfacing layer
(Surfage) and thickness of the surfacing layer (Surfthick). The traffic-related factors
that were included were traffic volume (AADT) and percentage of heavy commercial
vehicles (HCV). The other pavement surface defects that were identified as likely to
influence flushing were rutting and surface roughness, where measurements for left
wheel path and right wheel path rutting (Rutlwp and Rutrwp) and roughness (LwpIRI
and RwpIRI) were selected for analysis. Table ‎5.2 shows the resulting correlation
matrix from factor analysis.

5-141
 Table ‎5.2 Correlation matrix for pavement composition and traffic-related
factors – LTPP data set

TotalFlushing Surfthick Surfage AADT HCV Rutlwp Rutrwp LwpIRI RwpIRI

TotalFlushing 1.000 0.770 0.474 -0.172 0.307 0.210 -0.170 0.109 -0.214
Surfthick 0.770 1.000 0.309 -0.136 0.210 0.117 -0.241 0.133 -0.157
Surfage 0.474 0.309 1.000 -0.111 0.324 0.078 -0.109 0.129 0.085
AADT -0.172 -0.136 -0.111 1.000 -0.318 0.124 0.190 -0.026 -0.156
HCV 0.307 0.210 0.324 -0.318 1.000 0.080 -0.165 -0.120 -0.100
Rutlwp 0.210 0.117 0.078 0.124 0.080 1.000 0.436 -0.021 -0.094
Rutrwp -0.170 -0.241 -0.109 0.190 -0.165 0.436 1.000 -0.203 0.088
LwpIRI 0.109 0.133 0.129 -0.026 -0.120 -0.021 -0.203 1.000 0.568
RwpIRI -0.214 -0.157 0.085 -0.156 -0.100 -0.094 0.088 0.568 1.000

As can be seen in Table ‎5.2, both pavement characteristics (surface thickness and
surface age) had strong correlations to flushing (TotalFlushing), thus both of these
factors were further evaluated in detail. Of the traffic-related factors that were
included in the factor analysis, only HCV showed any significant relationship to
flushing. Rutting and roughness also had a correlation to flushing. Thus, all of these
factors (shown in bold font) were included in further analyses.

Effects of Climatic Factors

The correlation matrix for climatic effects, namely humidity (MeanHumidity), rainfall
(MeanRain) and temperature (MeanTemp) is shown in Table ‎5.3. Temperature and
humidity showed correlations to flushing and these two factors were investigated
further. Rainfall showed a weak correlation to flushing, with a correlation coefficient
of -0.029, thus rainfall was not investigated further.

5-142
 Table ‎5.3 Correlation matrix for climatic factors

TotalFlush MeanHumidity MeanRain MeanTemp

TotalFlush 1.000 -0.257 -0.029 -0.483

MeanHumidity -0.257 1.000 0.251 0.827

MeanRain -0.029 0.251 1.000 0.078

MeanTemp -0.483 0.827 0.078 1.000

5.3.2 Correlation Analysis of Variables Causing Flushing

Pavement Composition Characteristics Analysis
Surface Age and Surface Thickness
Pavement surface age and surface thickness were identified as being two pavement
composition characteristics that influenced flushing greatly. Figure ‎5.2 presents the
correlations between (a) flushing and surface age and (b) flushing and surface
thickness. The correlation between flushing and surface age was positive and
moderately strong (R2 = 0.225) (Cohen, 1988). The correlation between flushing and
surface thickness was also positive and strong (R2 = 0.593).

The strong correlation between flushing and surface thickness was consistent with
findings from the laboratory testing component of this research presented in Chapter
Three, where thicker chip seal surfaces were found to have a higher amount of
flushing. The correlation between flushing and surface age also aligned with past
findings, where older chip seal surfaces were found to exhibit more flushing
(Alderson, 2008).

5-143
 Line of best fit
95% Confidence
Interval (CI)

a) Flushing vs Surface age

Line of best fit

95% Confidence
Interval (CI)

b) Flushing vs Surface thickness

Figure ‎5.2 Correlation between flushing and pavement surface characteristics

5-144
Pavement Seal Generation and Flushing
The pavement sections that were included in the data analysis were classified into the
generation of the seal, which was determined based on the maintenance that had been
performed on the pavement sections. The seal generation was assigned as 1, 2, 3 or 4,
where 1 indicated first generation seals which were pavements that consisted the
original seal from the time when the pavement was first constructed, 2 indicated
second generation seals which were pavements that had been resurfaced once, 3
indicated third generation seals which were pavements that had been resurfaced twice,
and 4 indicated fourth generation seals which were pavements that had been
resurfaced three times. Descriptive analysis of the data set showed that surfaces that
were classed as second generation or higher had increased levels of flushing
compared to first or new generation surfaces. Figure ‎5.3 shows the flushing
measurements in each seal generation group where there was a clear difference in the
mean flushing values between first generation seals and higher generation seals (p
value < 0.005). Of the data set that was analysed for seal generation effects, only
0.50% of data points were identified as outliers.

Key

25–75 percentiles

° Outliers

Note: The whiskers extend to 1.5 x Inter Quartile Range (IQR) from the box

1 – first generation; 2 – second generation; 3 – third generation; 4 – fourth generation

Figure ‎5.3 The amount of flushing with respect to seal generation

5-145
Flushing and Aggregate Grade
One of the maintenance methods that is commonly used for remedying flushing is the
use of specific types of seals that are designed to reduce flushing, such as sandwich
seals and texturising seals. These seals add texture to an existing flushed surface
without adding extra bitumen. The LTPP database provided information about the
type of seal that was used on some of the sites, such as a single coat seal, two-coat
seal, void fill seal, sandwich seal, texturising seal or racked-in seal. There were also
some sites in the data set that were identified‎as‎having‎“reseals”‎as‎the‎surfacing‎type,‎
which did not clearly indicate which type of seal had been used. Due to this
inconsistency in the seal definition, it was not possible to directly use the seal type for
assessing flushing. As an alternative, the aggregate grade of the seal was used for
flushing assessment. The aggregate grades that made up the LTPP data set were grade
2 (19 mm), grade 3 (16 mm), grade 4 (14 mm), grade 5 (9.5 – 5.0 mm) and grade 6
(6.7 – 3.0 mm). The relationship between flushing and aggregate grade is shown in
Figure ‎5.4, and as can be seen, the amount of flushing is lower on surfaces that have
smaller sized aggregates. This trend indicated that aggregate size was a factor that
needs to be included in the flushing prediction model development. Of the data set
that was analysed for aggregate grade effects, only 0.05% of data points were
identified as outliers.

5-146
 Key

25–75 percentiles

° Outliers

Note: The whiskers extend to 1.5 x Inter Quartile Range (IQR) from the box

Figure ‎5.4 The relationship between flushing and the grade of aggregates

Traffic Loading and Flushing

The correlation between flushing and traffic volume was assessed as shown in Figure
‎5.5. The results show the flushing/traffic volume relationship to be negative, and the
strength of the correlation was low (R2 = 0.030) and was not statistically significant.
The weak correlation between flushing and traffic volume was unexpected, as traffic
volume typically contributes to the deterioration of pavement texture. Partial
correlations between flushing and traffic volume, and surface thickness and surface
age were tested to determine if pavement composition characteristics had an impact
on the flushing/traffic volume relationship. However, neither surface thickness nor
surface age were found to impact on the relationship between flushing and traffic
volume. In addition to the traffic volume, the effect of HCVs on flushing was also
explored as HCV volume was identified from factor analysis to impact on flushing.
Correlation analysis revealed a positive correlation between flushing and HCVs, but
the R2 value of this correlation was only 0.094, which indicated that HCV only
accounted for a small percentage of the variance in flushing.

5-147
 Figure ‎5.5 Relationship between flushing and traffic volume

The lack of a strong correlation between flushing and traffic volume was attributed to
the design of the LTPP sites. The traffic volumes that were measured on the LTPP
sites were between 42 vpd and 24,360 vpd, and these traffic volumes would have
covered the range of traffic volumes that were expected on these LTPP sites. The
LTPP pavements would have been constructed to satisfactorily withstand their
expected traffic volumes and as a result flushing development due to overloading
would have been minimised. If the LTPP pavements were exposed to significantly
higher traffic volumes than those observed in the data set, traffic volume could
become a significant variable that causes flushing. A detailed discussion about the
contribution of chip seal design specifications for minimising traffic-related flushing
is presented in Section ‎5.6.1.

Climatic Factor Analysis

The climatic factors that were investigated with respect to flushing were air
temperature (°C) and humidity (measured by mean vapour pressure, Hpa). Correlation
analysis of flushing and air temperature showed a negative correlation which was

5-148
moderate in strength between the two variables (R2 = 0.234). This correlation is
shown in Figure ‎5.6.

Line of best fit

95% Confidence
Interval (CI)

Figure ‎5.6 Correlation between flushing and mean air temperature

The negative correlation between flushing and temperature was contradictory to what
was expected. A positive correlation was expected between flushing and temperature
as binder behaviour with respect to temperature indicated that softer binder was
associated with high temperatures, which can lead to more flushing. Furthermore, the
lack of a stronger correlation between flushing and temperature was surprising, as
available literature (Lawson and Senadheera, 2009) indicated that temperature was
likely to have a significant effect on the occurrence of flushing. It was determined that
the effects of temperature were being altered to an extent by factors such as the
amount of bituminous binder in the seal layers which determines the extent to which
binder migration can occur, or the presence of harder binder which can minimise
temperature-related flushing. These factors would have been a part of the seal design
process which would have assigned binder quantities and grades according to the
expected temperatures of a site. As a result of this seal design process, temperature-
related flushing would have been minimised. Moreover, it was concluded that the
5-149
temperature data that was available in the LTPP database which presented the average
air temperature of the pavement sites was inadequate at accurately representing the
correlation between flushing and temperature, and it was identified that the
temperature of the pavement surface would better represent the effects of temperature
on flushing. However, pavement surface temperature data was not available in the
LTPP database, hence the effects of pavement surface temperature were not explored
in the presented data analysis. Additionally, the relationship between flushing and
humidity was found to have a weak R2 value of 0.066. The effects of humidity on chip
seal pavements would have also been taken into account in the design process of the
seals. Due to the above noted reasons, temperature and humidity were not included in
the regression model.

Another climate-related site factor that has the potential to influence flushing is the
quantity of moisture below the surfacing layer of the pavement, where previous
studies have identified the mechanism of bitumen movement to the surface of chip
seal pavements caused by water vapour movement as being a cause for flushing (Ball
et al., 1999). The availability of pavement subsurface moisture data in the LTPP
database would have allowed further analysis into the effects of water vapour on
flushing development, however, the lack of availability of such moisture data in the
LTPP database prevented pavement moisture from being explored further or being
included in the developed flushing prediction model.

Effects of Other Pavement Defects

Rutting
Pavement defects that were identified to have an impact on the occurrence of flushing
were rutting and surface roughness. Correlation analysis for the flushing/rutting
relationship revealed a weak positive correlation (R2 = 0.044) between flushing and
rutting measurements. Knowledge of pavement behaviour indicates that rutting affects
flushing where the presence of large ruts can contribute to binder accumulation,
particularly during construction, which then creates an area rich in binder that can
lead to flushing occurring as a secondary defect. To investigate the flushing/rutting
relationship further the rutting values were categorised into different groups. The
resulting box plot is presented in Figure ‎5.7 which shows an increase in the median
5-150
flushing values with an increase in rut depths, although the trend only becomes
significant when large rut depths are present (greater than 10 mm). The result in
Figure ‎5.7 confirmed that rutting was a factor contributing to flushing.

Key

25–75 percentiles

° Outliers

Note: The whiskers extend to 1.5 x Inter Quartile Range (IQR) from the box

Figure ‎5.7 Relationship between flushing and rutting

It should be noted that rutting on a pavement can occur due to two mechanisms. One
mechanism is where rutting occurs due to failure in the seal layers caused by uneven
densification of aggregate layers. The other mechanism is where rutting occurs due to
failure of the pavement layers underneath the seal layer. When considering the effect
on flushing from these two rutting mechanisms, it is the seal-failure mechanism that
needs to be investigated as this type of failure is an attribute of the seal composition.
In contrast, rutting due to pavement failure is more of a defect in the base layers of the
pavement and is unrelated to the composition and the performance of the seal layer.
Accurate identification of the cause of rutting on a given pavement site can determine
whether flushing is affected by rutting at that site. However, in the data that was
analysed there was no information available regarding the cause of rutting, and it was
not possible to identify the sites that were rutting solely due to seal failure. Therefore,

5-151
correlation analysis of the flushing and rutting data did not distinguish between the
two rutting mechanisms and this was identified as the reason why the expected
flushing/rutting relationship was not seen from the correlation analysis.

Roughness
Correlation analysis of the flushing/roughness relationship of LTPP data revealed a
weak correlation between the two variables, where the R2 value of the relationship
was 0.046. A stronger correlation was expected between flushing and roughness as a
pavement that had not flushed would retain its roughness. It is likely that the obtained
result of the flushing/roughness relationship was the product of a lack of severe
flushing on all the sites in the data set. Severe flushing would have a more significant
impact on roughness than low or moderate flushing, however, only a few sites in the
LTPP data set had severe flushing, and most of the sites that were analysed exhibited
lower severity flushing. Due to this pattern of flushing in the data set any negative
correlation between flushing and roughness was unlikely to be translated in the
roughness results.

5.3.3 Summary of Findings from Correlation Analyses

The descriptive data analysis that was performed on the LTPP data identified the
factors that had the greatest effect on the development of flushing, and confirmed and
revealed the nature of the correlations between these factors and flushing. The factors
affecting the development of flushing were separated into four categories, which were
pavement composition characteristics, traffic volume, climatic factors and other
pavement defects. The pavement composition characteristics that were found to
notably affect flushing were surface age, surface thickness, generation of the seal and
aggregate grade. Surface age and surface thickness had a positive correlation to
flushing and a clear distinction was observed in the amount of flushing occurring on
new seals compared to older seals. A clear trend was also present in the amount of
flushing that was measured on seals with different grades of aggregates, where
flushing was lower on the seals that had smaller sized aggregates. The traffic-related
factors that were explored were traffic volume and the percentage of HCVs. Both of

5-152
these factors had weak correlations to flushing which was contrary to expectations.
The climatic factors that were investigated were mean air temperature and humidity.
A moderate strength negative correlation was found between flushing and temperature
while a weak correlation was found between flushing and humidity. The negative
correlation between flushing and temperature was counter-intuitive because a positive
correlation was expected between these two variables. However, taking into account
the seal design processes that are used for chip seal pavements, it was understandable
that temperature was not a significant contributor to flushing. Due to the lack of
correlations between flushing and temperature and flushing and humidity, temperature
and humidity were not explored further. It was concluded that the air temperature data
that was available in the LTPP database was not accurately representing the
correlation between temperature and flushing. It was identified that pavement surface
temperature was a better factor to use rather than air temperature, however, as such
pavement surface temperature data was not available in the LTPP database, pavement
surface temperature was not explored further. The other pavement defects that were
explored in relation to flushing were rutting and surface roughness. The contribution
of rutting on flushing was evident at high rut depths (greater than 10 mm) while
roughness had a weak correlation to flushing.

The results of the correlation analyses revealed aspects of surface design practices that
were resulting in well-performing seals. For example, in chip seal design, adjustments
are made to binder grades, binder application rates and aggregate sizes to account for
differences in traffic volumes and environmental conditions, particularly temperature.
It was identified that these adjustments within the chip seal design process were the
reason why traffic volume and temperature did not have the expected effect on
flushing. Conversely, some surfacing practises were ineffective and added to the
flushing problem. Specifically, the practice of multiple resealing, which is a
commonly used and low-cost method of treating flushing of a chip seal, often leads to
a build-up of chip seal layers that results in an unstable structure. This phenomenon
was demonstrated by the positive correlations that were present between flushing and
surface age and surface thickness and was confirmed by the difference in flushing
values measured on new seals and older seals. Accordingly, the factors that were used
in the flushing model development were surface age, surface thickness, generation of
the seal, grade of aggregates, rutting and the number of HCVs. Additionally, another
5-153
element that affects the flushing of a chip seal is the stability of the seal design. Seals
that have shortcomings in its designs, such as construction defects, are difficult to
model as it is unclear whether flushing of those seals was due to construction faults or
due to characteristics of the seal. It was assumed that the pavement sections that were
included in the LTPP database were designed as per current practice at the time of
construction and as such there were no known issues with the design of the surfaces at
the time of selection. Therefore, it was assumed that the pavement sites in the LTPP
database were well-designed and as a result, flushing modelling was only performed
for chip seals that had been designed well.

5.4 Predicting the Initiation of Flushing

In pavement distress modelling it is important to determine when a distress gets to a
certain point such that intervention is needed to minimise the progression of that
distress. Identifying the point at which intervention is needed can allow for
appropriate monitoring and planning of suitable maintenance methods to address the
distress. The point of intervention can vary depending on the type of distress that is
being modelled, such as the instance when the distress first appears on the pavement
compared to when the distress reaches an unacceptable level. In the case of flushing, a
pavement can have a small amount of flushing and still be structurally sound and safe
for‎ use‎ by‎ traffic,‎ as‎ in‎ the‎ “pre-intervention”‎ phase‎ in‎ Figure ‎5.8. Once flushing
becomes more widespread on the pavement surface such that the safety performance
and structural integrity of the seal layers are compromised, the pavement must be
monitored for maintenance. Thus, when modelling flushing the point at which
intervention is needed can be considered as the point at which flushing initiates (as
shown in Figure ‎5.8), and is referred to as the point of flushing initiation in the
following model development.

5-154
 Figure ‎5.8 Deterioration phases of flushing

5.4.1 Model Development

The point of flushing initiation was modelled using logistic regression, which is an
effective method that is commonly used to model the likelihood of a certain event
occurring (Hosmer and Lemeshow, 1989). The advantage that a logistic model format
has over a simple linear regression model format is that the former provides the
probability of failure occurring, which allows for flexibility when applying the model
to a wide range of data sets. In the logistic model, the dependant variable takes a
binary form, typically 1 to denote the occurrence of a given event and 0 to denote the
non-occurrence of the event, and the output of a logistic model is a probability plot of
the given event occurring with respect to an independent variable. The logistic model
takes the form shown in Equation ‎5.1 (Kleinbaum and Klein, 2002).

5-155
 ( ) ( )

Equation ‎5.1

Where, p is the probability of a specific event occurring;

x is the independent variable;
a is the coefficient of the regression constant; and
b is the coefficient of the independent variable.

The percentage of flushing recorded on each pavement site in the LTPP database was
converted to a binary variable, FlushingInitiated, where 1 was used to indicate that
flushing had initiated on a pavement and 0 was used to indicate that flushing had not
initiated. The threshold for the initiation of flushing was determined by examining the
relationship between flushing and surface texture depth as given by Mean Profile
Depth (MPD) on pavement surface data collected from the state highway network in
the Napier‎ and‎ Hawke’s‎ Bay‎ regions‎ of‎ New‎ Zealand. The Napier/Hawke’s‎ Bay‎
pavement network consists of 1617 km of pavement, with approximately 75% being
sealed. Reduction in surface texture has been prominent early on in chip seal lives of
pavements in the Napier/Hawke’s‎Bay‎network‎and‎flushing, or low texture caused by
flushing, was identified as one of the main reasons for resurfacing on chip seal
pavements in this network (Ball and Patrick, 2005). The wide presence of flushing on
the‎ Napier/Hawke’s‎ Bay‎ network‎ made this pavement network ideal for use in the
data analysis. Historical pavement texture data from six years of surveying (from
2004 to 2009) were collected from the RAMM database and analysed to determine the
relationship between flushing and surface texture. This relationship is shown in Figure
‎5.9. The greatest change in surface texture occurred when flushing was between 0%
and 20%. When flushing reached higher than 20% the change in surface texture was
slower. The point of deterioration of surface texture is an important maintenance
trigger which also indicates the presence of noticeable levels of flushing. Hence, a
flushing threshold value of 20% was chosen as the point at which flushing initiates on
a given pavement site.

5-156
 Key

25–75 percentiles

° Outliers

Note: The whiskers extend to 1.5 x Inter Quartile Range (IQR) from the box

Figure ‎5.9 Relationship between flushing and surface texture

In order to make the flushing initiation prediction model more applicable to a wide
range of pavements, the LTPP data from all survey years from the state highway and
local authority LTPP databases were combined. The combined data set was then
separated into two subsets by allocating the data entries randomly to the subsets and
making sure that the subsets were similar in size. One subset was used for model
development and the other subset was used for model validation. The subset that was
used for the logistic model development contained 827 data entries and the subset that
was used for model testing contained 812 data entries.

The variables that were included in the logistic regression were determined by
investigating the effect that each potential independent variable had on flushing
initiation. Figure ‎5.10 shows the effect of surface age (in years), surface depth (in
millimetres), rutting (in millimetres) and number of HCVs on flushing initiation. Seal
generation as shown in Figure ‎5.10c indicates the generation of the seal, where 1
indicates first generation seals, 2 indicates second generation seals, 3 indicates third
generation seals and 4 indicates fourth generation seals. It was clear from Figure ‎5.10
that surface age, surface thickness and rutting had an effect on flushing initiation,
where for all three variables the difference in the mean between pavements that had
5-157
flushing initiating and pavements that did not have flushing initiating was statistically
significant (p value < 0.001). The number of HCVs was seen to not have a notable
effect on flushing initiation.

a) Surface age b) Surface thickness

c) Rutting d) Number of HCVs

Figure ‎5.10 Effect of independent variables on flushing initiation

5-158
Logistic regression was performed using a forward stepwise method. The independent
variables that were used in the regression were surface age (surfage), surface
thickness (SurfDepth), rutting (RutLANE), and number of HCVs (no_of_heavies).
The model coefficients from the logistic regression are shown in Table ‎5.4, and as can
be seen, surface age and surface thickness were statistically significant variables in the
model (p value < 0.001). Rutting and the number of HCVs were not included in the
final model as the significance of the contribution of rutting (p value = 0.654) and the
number of HCVs (p value = 0.347) to the model was low. The contribution of each of
the independent variables was as expected, as shown by the positive constants in
column‎ “B”‎ in‎ Table ‎5.4, where flushing was expected to increase with increasing
surface age and surface thickness.

Table ‎5.4 Model coefficients from logistic regression to model initiation of

flushing

Variable B S.E. Wald df p value Exp(B)

SurfDepth 0.046 0.005 77.134 1 < 0.001 1.047

surfage 0.293 0.025 132.594 1 < 0.001 1.340
Constant -2.913 0.265 120.994 1 < 0.001 0.054

Based on the coefficients in Table ‎5.4 the logistic model for initiation of flushing is as
shown in Equation ‎5.2:

( ) ( )

Equation ‎5.2

Where, p(FlushingInitiated) is the probability that flushing has initiated;

surfage is the age of the pavement surface in years; and
SurfDepth is the thickness of the pavement surfacing layer in millimetres.

5-159
The graphical representation of the logistic model format to indicate the initiation of
flushing is shown in Figure ‎5.11, which presents the probability of flushing initiating
on a given pavement surface with respect to surface age and surface thickness. The
output in Figure ‎5.11 shows that the time at which flushing initiates varies
significantly for different surface thicknesses. If the probability of flushing initiating
was taken as 0.5, then for a surface with a thickness of 10 mm, flushing will initiate at
8 years, while when the surface thickness is 30 mm flushing will initiate at 5 years.
When the surface thickness of a pavement is approximately 60 mm, flushing will
initiate immediately from the beginning of the pavement life, which is attributed to
the build-up of chip seal layers from multiple reseals.

1
Probability of flushing initiating on a pavement

0.8

0.6

0.4

Surface thickness 10mm

0.2
Surface thickness 30mm

Surface thickness 60mm

0
0 5 10 15 20 25 30
Surface Age (years)

Figure ‎5.11 Model outcome to predict the probability of flushing initiation

5.4.2 Test of Logistic Model Accuracy

The prediction accuracy of the logistic model is shown in Table ‎5.5. The predicted
values‎for‎“Selected‎Cases”‎show‎the‎model‎predictions‎for‎the‎data‎set that was used
to‎ develop‎ the‎ model,‎ and‎ the‎ predicted‎ values‎ for‎ “Unselected‎ Cases”‎ show‎ the

5-160
model predictions for the data set that was used to validate the logistic model. The
overall prediction accuracy of the logistic model was established by determining the
percentage of the model-validation data set for which flushing initiation was predicted
correctly (calculation detailed in Appendix G). The developed logistic model is
predicting at an accuracy of 76% (75.5% from Table ‎5.5). The model is much more
accurate when predicting flushing initiation for a pavement where flushing had
initiated (82% from Table ‎5.5) than on a pavement where flushing had not initiated
(64% from Table ‎5.5). A graphical representation of the actual and the predicted
results of flushing initiation for the data that were used to validate the logistic model
is shown in Figure ‎5.12. As can be seen in Figure ‎5.12, the majority of the data entries
were predicted accurately by the model, where on the plot of pavements that did not
have flushing initiating (Figure ‎5.12a) the bars were clustered to the left side of the
plot as expected and on the plot of pavements that had flushing initiating (Figure
‎5.12b) the bars were clustered to the right side of the plot.

Table ‎5.5 Prediction accuracy of the logistic model for flushing initiation

Predicted

Selected Cases Unselected Cases

Observed
FlushingInitiated Percentage FlushingInitiated Percentage
0 1 Correct 0 1 Correct

0 194 126 60.6 190 106 64.2

FlushingInitiated
1 109 398 78.5 93 423 82.0

Overall Percentage 71.6 75.5

5-161
 a) Observed state of flushing initiation: 0 - no b) Observed state of flushing initiation: 1 - yes

Figure ‎5.12 Predicted probabilities for flushing initiating on LTPP pavement

sections

5.5 Predicting the Progression of Flushing

The following section presents the results obtained during the development of the
flushing progression model. The independent variables that were used in the
development of the flushing progression model were pavement surface thickness,
surface age, percentage of HCVs and rutting. The progression of flushing on the
LTPP sites was modelled using a forward stepwise linear regression model (Kendall
and Stuart, 1969; Salama et al., 2008; Hocking, 2003). The linear regression model
for flushing progression modelling was selected by conducting curve estimations on
the correlations between each independent variable and the dependant variable
(Pallant, 2007), where different regression curves such as linear, exponential,
quadratic and polynomial curves were tested to observe which curve best represented
the correlations. From the curve estimation exercise it was found that the linear curve
fitted the data the best, hence linear regression was selected for developing the
flushing progression prediction model. Analysis of the distribution of flushing
measurements in the data set revealed that the flushing data distribution was right
skewed and violated normality assumptions of linear regression (Figure ‎5.13), thus the
dependant variable TotalFlushing was transformed by obtaining the square root of
5-162
TotalFlushing. The transformed data distribution was more normally distributed than
the original data distribution and was suitable for linear regression.

a) Original data b) Transformed data

Figure ‎5.13 Distribution of original dependant variable TotalFlushing data

and the transformed data

Analysis of flushing measured on seals of different generations revealed that there

were clear differences in flushing measurements between first generation or new seals
compared to second and higher generation seals. Flushing on first generation seals
were notably lower compared to higher generation seals and this variation was taken
into account in the flushing progression model by developing two variations of the
linear model. The data set for the analysis was separated based on the surface
generation, where set 1 consisted of data from first generation seals and set 2
consisted of data from second and higher generation seals, and two linear models
were developed for the two seal generation types. The regression model coefficients
for first generation seals are shown in Table ‎5.6. The variables that were found to be
significant contributors to the linear model (p value < 0.001) were surface age
(surfage), surface thickness (SurfDepth) and percentage of HCVs (pc_heavy). The R2
value of the linear model was 0.351, which indicated that the model had moderate
statistical robustness. Analysis plots of the residuals of the linear model for first
5-163
generation seals are shown in Figure ‎5.14. The normally distributed residuals in the
histogram and the normal probability plot showed that normality and linearity
assumptions of linear regression were satisfied for the data set.

Table ‎5.6 Linear regression model coefficients for first generation seals

Unstandardised Standardised 95% Confidence

Coefficients Coefficients Interval for B
Model t p value
Lower Upper
B Std. Error Beta
bound bound
(Constant) -0.509 0.584 -0.871 0.384 -1.66 .641

surfage 0.280 0.025 0.546 11.080 < 0.001 .230 .329

pc_heavy 0.141 0.020 0.349 6.985 < 0.001 .101 .181

SurfDepth 0.160 0.051 0.155 3.158 0.002 .060 .260

a. Dependent Variable: sqrt_flushing

a) Histogram of residuals b) Normal probability plot

Figure ‎5.14 Evaluation of residuals from linear regression for first generation
seals

5-164
For pavements with seals that were in the second or higher generation category the
linear regression coefficients are shown in Table ‎5.7. The independent variables that
were found to be significant contributors to the linear model (p value < 0.001) were
surface age (surfage), surface thickness (SurfDepth), rutting (RutLANE) and grade of
aggregates (Chipsize). The R2 value of the linear model for second and higher
generation seals was 0.643, which was statistically robust. Results from the analysis
of residuals for the linear model for second and higher generation seals are shown in
Figure ‎5.15. The normality and linearity assumptions of linear regression were also
satisfied for the second and higher generation seals model.

Table ‎5.7 Linear regression model coefficients for second and higher
generation seals

Unstandardised Standardised 95% Confidence

Coefficients Coefficients Interval for B
Model t p value
Lower Upper
B Std. Error Beta
bound bound
(Constant) 0.108 0.474 0.227 0.821 -.824 1.04

surfage 0.416 0.017 0.742 24.193 < 0.001 .382 .450

SurfDepth 0.040 0.005 0.288 8.453 < 0.001 .031 .049

RutLANE 0.110 0.029 0.111 3.838 < 0.001 .054 .167

Chipsize 0.170 0.065 0.094 2.615 0.009 .042 .299

a. Dependent Variable: sqrt_flushing

5-165
 a) Histogram of residuals b) Normal probability plot

Figure ‎5.15 Evaluation of residuals from linear regression for second and
higher generation seals

The model that was developed for predicting flushing progression on LTPP
pavements is shown in Equation ‎5.3 and Equation ‎5.4:

First generation seals:

Flushing = [0.280surfage + 0.160SurfDepth + 0.141pc_heavy]2

Equation ‎5.3

Second and higher generation seals:

Flushing = [0.416surfage + 0.040SurfDepth + 0.110RutLANE + 0.170Chipsize]2

Equation ‎5.4

Where, Flushing is the percentage of flushing on the pavement;

surfage is the age of the pavement surface;
SurfDepth is the thickness of the pavement surfacing layer (in mm);
RutLANE is the average rutting on the pavement (in mm);
pc_heavy is the percentage of heavy commercial vehicles; and
Chipsize is the grade of aggregates on the pavement surface.
5-166
5.5.1 Testing of Flushing Progression Model
The model developed to predict the progression of flushing was tested using the
second subset of LTPP pavement data. As the model for first generation seals only
showed moderate robustness, this model was not tested further and testing was only
performed for the second and higher generation seals model. To assess the fit of the
developed model, the observed flushing values on second and higher generation seals
from the second subset of data were plotted against the predicted flushing values from
the developed model and the R2 value of the correlation was found. Figure ‎5.16 shows
the predicted flushing values against the observed flushing values for pavements with
second and higher generation seals. The R2 value of the relationship of observed
flushing and model predicted flushing was 0.561. The linear model had statistically
significant strength in its predictions of flushing for second and higher generation
seals, as shown in Figure ‎5.16. The R2 value of the observed versus predicted flushing
plot was lower than the R2 value for the model itself, but given that the model was
applied to a completely separate data set, the obtained prediction still showed
statistically significant strength.

Figure ‎5.16 Model predictions for flushing progression on second and higher
generation seals
5-167
The predictive strength of the model for first generation seals was expected to be low
because the mechanisms that cause flushing on first generation seals can vary much
more than the mechanisms that cause flushing on second and higher generation seals.
On second and higher generation seals, much of the flushing was attributed to the
build-up of surfacing layers over time, leading to pavements that were unstable and
susceptible to shear and thus flushed quickly. The contribution of surface thickness to
flushing was confirmed by the data analysis that was presented earlier in this chapter
as well as laboratory testing results presented in Chapter Three. Conversely, on first
generation seals the surface thickness would not be as great as for second and higher
generation seals and therefore flushing due to instability in the chip seal structure was
less likely. On first generation seals flushing was attributed more to the increase in
surface age, or due to construction defects which were difficult to identify from the
available data, therefore flushing progression on first generation seals was less
predictable than for second and higher generation seals.

The results from the analyses presented in this chapter demonstrated the use of
statistical techniques to predict the occurrence and progression of flushing on chip
seal pavements based on pavement and site characteristics. The outcomes of the data
analysis provided a valuable understanding into the nature of flushing as currently
there is no known model which considers the combination of different variables and
their effect on flushing. The developed model takes into account a number of site
characteristics when determining the likely amount of flushing at a site, and this
method of flushing prediction gives a more accurate indication of flushing than when
using a single variable or pavement characteristic was used to identify flushing. As
the presented method of flushing prediction considered the whole site when
determining the flushed condition, it is an efficient method to use for flushing
identification on large pavement networks.

5-168
5.6 Limitations of the Data Set Affecting Flushing
Prediction
5.6.1 Traffic Volume Data
The presented flushing model was developed primarily using historical pavement
condition data, and while this approach gave good model formats with statistically
significant accuracy to predict flushing, there were variables that would intuitively be
expected to contribute to flushing but were not seen to be significant from the data
analysis. One such variable was traffic volume, which did not have a significant effect
on flushing occurrence. The reason for this observation was that the range of traffic
volumes that were included in the data set, which was between 42 vpd and 24,360
vpd, covered the traffic volumes that the chip seal pavements in the data set would
have been designed to withstand. These pavements would have been constructed
according to the standard design specifications used for chip seal surfacings in New
Zealand (Austroads, 2004), which adjust aggregate sizes and binder application rates
to allow sufficient texture on a particular pavement surface to suit its expected traffic
volume. As a result, flushing development on these seals was not due to overloading
of the seals. When considering pavements that are subjected to significantly higher
loadings than the traffic volumes observed in the analysed data set or loading
conditions that these chip seal surfaces would have been designed for, traffic volume
could become a significant variable that contributes to flushing.

5.6.2 Unavailability of Air Voids Data

The image analysis results documented in Chapter Four of the thesis revealed that the
reduction in air void volume over time was an extremely useful measure for
predicting flushing. However, it was not possible to incorporate air void volume in the
developed flushing prediction model because the LTPP database does not provide
data relating to the air void volumes of the sites that are surveyed. It is expected that
the availability of air void volume data would be of significant value for pavement
deterioration modelling, and especially for flushing prediction as flushing is affected
by the micromechanical changes that occur within a chip seal structure. It is
recognised that it may not be possible to provide air void volume data for every LTPP
site and for every survey year, however, when it has been identified from the
5-169
prediction model that flushing had initiated on a particular site, air void volume
analysis should be carried out to confirm the model predictions. Using two methods of
assessment can ensure that sites were correctly identified as flushed and also that the
most effective treatment method was selected to remedy the flushing.

5.7 Conclusions
The purpose of this chapter was to present the results of developing a two-stage model
to predict the initiation and progression of flushing on chip seal pavements. The
model presented in this chapter was developed using pavement condition data from
the LTPP database. The complete set of LTPP data was separated into two subsets and
one subset was used for model development while the other subset was used for
model testing. Based on the data analysis performed on LTPP data, the variables that
were found to have the most significant effect on flushing were surface age, surface
thickness, rutting and the size of aggregates.

The first stage of modelling flushing was to develop a model to predict the probability
of flushing initiating on a given pavement. The point at which flushing initiates was
determined based on analysis of the relationship between flushing and surface texture,
and the threshold for flushing initiation was determined as being the point in time
when a given pavement was showing flushing on 20% or more of the pavement
surface area. A logistic model was employed to predict the probability of flushing
initiation and the developed model was tested on a separate set of LTPP data.
Outcomes from testing the logistic model revealed that the model had statistically
significant accuracy (76%) in predicting the likelihood of flushing initiating on a
pavement. The second stage of flushing that was modelled was the progression of
flushing, and a nonlinear model was employed for this task. Initial data analysis
revealed that the flushing trend was different on pavements that had only one seal
(first generation seals) compared to pavements that had multiple seals (second and
higher generation seals), and based on this observation flushing progression was
modelled separately for these two types of pavements. The model for second and
higher generation seals was statistically strong (R2 value of 0.643) while the model for
new or first generation seals was only moderately strong (R2 value of 0.351) (Cohen,

5-170
1992). The developed model for second and higher generation seals was tested using a
separate set of LTPP data and the model predictions revealed that the model was
robust at predicting the progression of flushing. Thus, the flushing progression model
for second and higher generation seals was recommended for use as part of the
management process for pavement networks containing chip seal pavements.

5.7.1 Recommended Model Formats for Predicting Flushing

The model developed from the data analysis for modelling flushing using pavement
condition data is presented in Table ‎5.8.

Table ‎5.8 Data-driven models to predict flushing

Stage of Flushing Recommended Model Format

Initiation of Flushing
( ) ( )

Where,
p(FlushingInitiated) is the probability flushing has initiated;
surfage is the age of the pavement surfacing layer, in years; and
surfDepth is the thickness of the surfacing layer, in millimetres.

Progression of flushing Second and higher generation seals:

Flushing = [0.416surfage + 0.040SurfDepth + 0.110RutLANE +
0.170Chipsize]2

Where,
Flushing is the amount of flushing on the pavement;
surfage is the age of the pavement surface;
SurfDepth is the thickness of the pavement surfacing layer;
RutLANE is the average rutting on the pavement; and
Chipsize is the grade of aggregates in the surfacing layer.

5-171
5.8 Recommendations
5.8.1 The Need for Testing of Flushing Prediction Model
It is recommended that the model developed in this study to predict the initiation and
progression of flushing be tested on data from different pavement networks to test its
accuracy and to detect any adjustments needed for regional differences, for example
to account for differences in soil types and climatic conditions. At present, the lack of
availability of data relating to the soil characteristics of a site in the LTPP database
limits the use of soil characteristics in distress prediction models.

The effects of the presence of moisture below the pavement surface on the
development of flushing were briefly investigated in this research by collecting
density and water content measurements of the soil directly underneath the pavement
surface from 11 state highway and local authority LTPP sites. The LTPP sites that
were‎surveyed‎were‎all‎from‎the‎Napier/Hawke’s‎Bay region of New Zealand. Each
LTPP site (300 m) was divided into six 50-metre sections and the moisture
measurements, referred‎ to‎ as‎ “soil‎ moisture”‎ from‎ here‎ onwards,‎ were measured at
each of these sections (shown in Figure ‎5.17). A nuclear densometer (Smith and
Diefenderfer, 2008) was used to perform the moisture measurements where the
measurements were taken from both wheel paths of the pavement. Along with soil
density measurements, characteristics of surface texture (eg. flushed, coarse textured)
were also recorded.

5-172
 50 m
Pavement edge

x x x x x x x Left wheel path

Direction of traffic

x x x x x x x Right wheel path

Centreline

300 m
x - Location of testing

Figure ‎5.17 Locations of moisture measurements

The correlations between flushing and soil water content and soil dry density were
found (Figure ‎5.18), and the results revealed a moderately strong correlation between
flushing and water content. The availability of a wide ranging soil moisture
characteristics data set would be of significant use especially for modelling pavement
defects such as flushing or shoving which are likely to be affected by subsurface
moisture, as shown by the data analysis results obtained in this study. Thus, the
pavement condition measures that are collected as part of the LTPP programme need
to be widened to include soil moisture characteristics of LTPP sites in future surveys
to allow soil characteristics to be included in the development of future deterioration
models.

5-173
 Line of best fit
95% Confidence
Interval (CI)

a) Flushing vs Water content of soil

Line of best fit

95% Confidence
Interval (CI)

b) Flushing vs Dry density of soil

Figure ‎5.18 Correlation between flushing and a) soil water content and b) soil
dry density

5-174
5.8.2 Use of Model Outcomes for Better Prediction of Flushing
The data analysis and model development that were presented in this chapter
demonstrated how the current design, construction and maintenance practices that
were used on chip seal pavements can be improved to minimise flushing. The data
analysis revealed that some of the design practices that are currently used for chip
seals, particularly specifications that adjusted binder grades, binder application rates
and aggregate sizes depending on expected traffic volumes and local climatic
conditions, were effective at minimising flushing. This was evident from the traffic
volume data analysis which showed that traffic volume did not make a significant
contribution to flushing occurrence. Similarly, adjustments that were made to chip
seals, particularly to binder grades to account for temperature differences, were also
effective as the data analysis revealed that temperature did not have the expected
positive correlation to flushing.

The data analysis also indicated that some of the maintenance practices that were used
on chip seal pavements were accelerating the occurrence of flushing. Specifically,
common maintenance practices that were used for treating flushed surfaces such as
resealing promoted flushing development by creating layer upon layer of chip seals
which result in an unstable surface that flushes quickly. These unstable flushed
surfaces can only be remedied using methods such as water cutting or complete
rehabilitation of pavement surface. While water cutting and rehabilitation have higher
initial costs than resealing, these methods can be much more effective than resealing
at treating flushing and preventing future flushing development. When determining
maintenance treatments for a flushed site, multiple investigative methods should be
used in order to identify the best maintenance treatment that suits each individual site.
It was demonstrated in Chapter Three and Chapter Four of this thesis how laboratory
testing in the form of wheel tracking can be used to determine the manner in which
chip seal layers behaved when subjected to loading. Combining the outcomes from
the flushing prediction model as presented in this chapter with laboratory operated
wheel tracking to identify at-risk seals and to determine how they behave under
loading is a robust and efficient method of selecting the best maintenance treatment
for a potential flushed site.

5-175
CHAPTER 6 - SUMMARY AND
RECOMMENDATIONS OF
RESEARCH FINDINGS

6.1 Introduction
This thesis documents research that was conducted to investigate the development of
the pavement surface defect of flushing. Flushing is one of the most widespread
defects on chip seal pavements in New Zealand and it has a detrimental effect on the
functional performance of a chip seal pavement. Prior to conducting the documented
research the understanding of the processes that led to the development of flushing
was limited, and there were no methods available for early detection of flushing. The
documented research developed new test methods for assessing the structural
characteristics of chip seal surfaces that lead to flushing and through this process, a
fundamental understanding was gained about the development of flushing. The
research outcomes confirmed some important aspects of chip seal material behaviour
that lead to flushing. In particular, the contribution of chip seal thickness to flushing
development was confirmed by the laboratory-based testing as well as data analysis.
The research outcomes also contested some theories about the processes leading to
flushing, namely the effect of fine solid material on flushing development. Past
research had theorised that the presence of fine solid material in a chip seal had a
major contribution to flushing. However, the outcomes from this research
demonstrated that fine solid material did not directly affect the development of
flushing. Through this research a new understanding was developed into the
contribution of internal structural characteristics of chip seal pavements to the
development of flushing. The research demonstrated that air void volume of a chip
seal surface had a direct effect on flushing, and the reduction in air void volume could
be used to predict the amount of flushing likely to occur on a chip seal surface.
6-176
Furthermore, the flushing forecasting model that was developed through this research
provides an efficient method for identifying and predicting network-wide flushing.
Overall, this research has significantly advanced the knowledge of the flushing defect
and has developed techniques for effectively assessing flush-prone pavements. The
current chapter brings together the understandings gained in this research to
demonstrate how the research findings can be utilised to improve current pavement
designs and management practises to minimise pavement failures due to flushing.

6.2 Implications of Research Findings

One of the primary aims of this research was to develop a fundamental understanding
of the mechanistic behaviour of chip seal surfaces in order to determine how that
behaviour related to the eventuation of flushing. Prior to conducting this research
there were some studies that had identified some of the likely causes of flushing
(Lawson and Senadheera, 2009; Alderson, 2008; Ball and Patrick, 1998), however,
none of these studies had investigated how the interactions that take place between
chip seal materials contributed to flushing. The presented research therefore took a
different approach of investigation for flushing by firstly understanding how chip seal
materials interact as this material interaction is often the underlying cause for flushing
development.

There were three main methodological components to this research, which were
laboratory-based mechanical testing of chip seal samples, CT scanning and image
analysis of chip seal samples, and analysis of pavement condition data. These three
methodological components investigated the development of flushing at different
scales. Specifically, the laboratory testing and image analysis components of the
research investigated flushing from a fundamental perspective concentrating on the
effects of volumetric changes and micromechanical changes within chip seal layers.
The data analysis component of the research concentrated on developing a flushing
prediction model based on available network-level pavement performance data. The
laboratory testing and image analysis components were largely separate from the data
analysis component of the research, although the findings from all three
methodological components contributed to developing a thorough understanding of

6-177
the development of flushing. The following sections summarise the main findings
from the research and the implications of those findings for the management of
flushed chip seal pavements.

6.2.1 Mechanical Testing of Pavement Samples

The main constituents that form a chip seal layer are aggregates, bitumen and air
voids, and when considering flushing, the changes in the volumetric balance between
these constituents have a major effect on the occurrence of flushing. In this research,
laboratory-based mechanical testing was conducted on chip seal pavement samples to
examine how the constituents in a chip seal layer affected its structural performance
and to understand the interactions that occurred between these constituents that
contributed to flushing. In particular, the laboratory testing aimed to determine the
effects of loading on the flushing behaviour and the deformation behaviour of chip
seal samples. The testing also aimed to establish how the deformation behaviour of a
chip sample and its aggregate and bitumen characteristics could be used to identify
chip seals that were prone to flushing. The mechanical tests that were conducted on
the samples were adopted from standard test methods that are used for asphalt
concrete pavement samples. The main tests that were conducted were the
measurement of air void volumes in the samples, controlled loading in the form of a
wheel tracking test, measurement of aggregate grading profiles and measurement of
bitumen contents within the samples. The task of examining the volumetric changes
within a chip seal was based on studies of asphalt concrete pavements as there was
limited research which had examined flushing from this perspective. The samples that
were tested were obtained from flushed, in-service chip seal pavements from
Auckland, Waikato, Christchurch and Dunedin regions of New Zealand.

The laboratory testing demonstrated how a chip seal structure behaves when subjected
to loading. By using controlled loading conditions of different magnitudes and lengths
of time, it was possible to examine how the deformation behaviour of a chip seal
related to flushing. The laboratory test results confirmed some of the findings from
previous research and from field observations regarding chip seal behaviour
(Austroads, 2008; Gray and Hart, 2003), specifically, that chip seal thickness had a

6-178
direct relationship to the quantity of flushing that eventuated on a surface and that
thicker surfaces were more unstable than thinner surfaces and therefore, they
deformed much more easily. Additionally, the pattern of deformation of a chip seal
indicated the state of stability of that chip seal structure, whether the structure would
reach a stable state when loaded or whether the structure would continuously deform
and become unstable with continued loading. The test results showed that while
flushing eventually develops with continued loading, it is the stability of the chip seal
that determines whether flushing will continue to develop or reach a stable state that
can be treated easily. This understanding of the connection between flushing and chip
seal material behaviour was not reported in previous literature.

The stability of a chip seal was also dependant on the quantity of fine solid material
(fines) that was present within the chip seal, where stable surfaces had less fines
compared to unstable surfaces. Previous research had theorised that fines had a direct
effect on the occurrence of flushing, where it was supposed that fines occupied the air
void volume within a chip seal layer that was available for bitumen movement thus
causing the bitumen to exude to the surface of the chip seal layer resulting in flushing.
While these studies were able to conclude that fines were generated in a chip seal,
they were not able to explain the exact effect of the fines on flushing occurrence or
exactly how the fines were generated. These gaps in the knowledge of chip seal
behaviour were answered in the documented research where it was demonstrated that
fines affected the deformation stability of a chip seal structure and this is the reason
why flushing continues to develop in some surfaces even after maintenance. The
reason why some surfaces exhibit a stable deformation pattern and others exhibit an
unstable deformation pattern was because when large aggregates disintegrate
overtime, due to either traffic impacts or environmental impacts, the disintegration
prevents the aggregates from creating a strong matrix that resists deformation. This
means that despite resurfacing, the underlying chip seal layers will continue to behave
unstably and deform overtime, and continuously exude bitumen to the pavement
surface. Furthermore, the laboratory results demonstrated that it is possible to simulate
flushing even under conditions where fines generation is not possible, such as from
the wheel tracking test. As mentioned earlier in the thesis, the impact of the rubber
wheel that was used to apply loading to the cores would not have caused any
disintegration of aggregates, and if it is assumed that it is the generation of fines that
6-179
causes flushing, then it would not have been possible to simulate flushing on the
wheel tracked cores. Therefore, from the laboratory test results it was concluded that
flushing was not necessarily caused due to the fines occupying the air voids spaces,
rather that the presence of fines contributed to instability in the chip seal structure and
it was this instability that caused the flushing.

Observation of the deformation behaviour of a chip seal surface can be used within
pavement asset management programmes to prevent the development of flushing.
Conducting mechanical loading tests on unflushed chip seal surfaces as part of routine
monitoring can aid with identification of the likely pattern of deformation of a chip
seal surface. This process in turn can indicate how the surface needs to be managed.
For example, if a chip seal surface that has been identified as having the potential to
flush was displaying the unstable deformation pattern, then that surface could be
treated with a reseal consisting of steel slag aggregates. The interlocking properties of
steel slag aggregates are known to resist deformation (Byers et al., 2004; Li et al.,
2012), which will be very beneficial when trying to reduce bitumen movement
through excessive densification. The deformation resistance of steel slag aggregates
will also aid with preventing future unstable deformation. Furthermore, steel slag
aggregates are much more durable than commonly used aggregates, such as
greywacke and alluvial aggregates, which means fines development within a chip seal
due to disintegration of aggregates will be minimised. The use of aggregates such as
steel slag may have a high initial cost, however, these aggregates have the potential to
reduce the long term cost of a surface, and especially maintenance costs. Similarly, if
a chip seal surface was displaying the stable deformation pattern, then that is an
indication that any flushing that is likely to develop on that surface will be mainly due
to the surface containing an excessive amount of bitumen. In this situation it is
important to avoid introducing more bitumen into the chip seal surface and any
routine maintenance treatments that are conducted on the surface needs to take this
into account.

6-180
6.2.2 Image Analysis of Pavement Samples
Computed tomography scanning and image analysis techniques were utilised to
examine the internal structure of mechanically-tested chip seal samples. The purpose
of the image analysis task was to quantify the changes that occurred to the air void
volume and distribution within chip seals in order to determine the effect of those
changes on the flushing potential of a chip seal. This task developed 3D images of
chip seal samples in order to visualise the placement of air voids within a chip seal.

The role of air voids in the behaviour of asphalt concrete surfaces has been well
documented in past literature (Krishnan and Rajagopal, 2005; Buttlar et al., 1999).
However, the effect of air voids within chip seal structures has not been explored
previously. The analyses conducted in this research demonstrated that air voids in
chip seal surfaces were equally important as in asphalt concrete surfaces when
considering the volumetric behaviour of a surface layer. Air void volume calculations
made using CT scan images showed that the reduction in air void volume due to
loading had a strong, positive correlation to the amount of flushing that was present
on the sample surface after being subjected to loading. The reason that was identified
for the reduction in air void volume was the presence of excessive amounts of
bitumen within the chip seal surface. The amount of flushing that was present on a
sample was also strongly correlated to the ratio between bitumen volume and air voids
volume, where the samples that had a high bitumen-to-air void volume ratio had
higher amounts of flushing when compared to samples that had a low bitumen-to-air
void volume ratio. When a chip seal layer that has a high ratio of bitumen-to-air void
volume is subjected to compaction, then it is only a matter of time before flushing
develops on that surface. These findings confirmed that flushing developed as a direct
result of air void volume reduction. The reduction in air void volume in a chip seal
sample can therefore be used to determine the amount of flushing that would occur on
the corresponding chip seal surface.

The 3D models of the air void distributions within chip seal samples that were
developed from CT scan images were extremely effective at indicating the size and
placement of air voids within a sample. The 3D models also allowed differentiation of
the structural characteristics of samples that were obtained from different

6-181
geographical regions. For example, samples from the Waikato region exhibited large
areas of interconnected air voids particularly at the aggregate layer interfaces, while
samples from the Dunedin region exhibited smaller-sized, unconnected air voids that
were distributed throughout the sample. It was supposed that the different
combinations of seals, such as single coat seals, two-coat seals and texturising seals,
that would have been used at these sites was the cause of the difference in air voids
distribution characteristics. Moreover, 3D models of the air voids could be used to
assess the performance of different types of seals when they are used to introduce air
voids into a chip seal structure that was lacking in air voids in order to prevent
flushing development.

In addition to the technical findings, the CT scanning and image analysis processes
provided a unique insight into the internal structure of chip seals containing multiple
chip layers. The use of CT scanning for pavement structure analysis had been limited
to asphalt pavements, and there had not been any research conducted which had
investigated the use of CT scanning for chip seal structure analysis. The documented
research was the first of its kind which utilised CT scanning for chip seal structure
analysis and the research results demonstrated the value of CT scanning and image
analysis techniques when analysing the internal structure of chip seal surfaces.
Moreover, previously the assessment of chip seals surfaces had been conducted using
indirect test methods and calculations. However, these indirect methods failed to
incorporate the undisturbed state of the internal components of a chip seal surface,
namely the placement of air voids and aggregates, into chip seal behaviour. In
contrast, the CT scanning method that was used in this research was able to provide
much more detailed and accurate information about the undisturbed state of the
internal structure of chip seal layers, allowing for a more comprehensive assessment
of chip seal behaviour.

As outline above, the combination of CT scanning and image analysis has many
practical applications, especially when conducting in-depth analysis of chip seal
behaviour. The use of CT scanning to gather data about the internal structure of
pavements as part of routine pavement assessments is extremely beneficial in the case
of flushed chip seal pavements as it will allow for confirmation of the likely causes of

6-182
flushing. The following sections present a detailed discussion of the role of CT
scanning in the flushing prediction and assessment processes.

6.2.3 Data Analysis and Prediction Model Development

The primary aim of the data analysis component of this research was to develop a
statistical model to predict the initiation and progression of flushing on chip seal
pavements based on network-level pavement management data. The data analysis also
aimed to identify variables known to cause flushing based on pavement condition data
gathered from chip seal pavements in the LTPP database. The analysis methods used
in the flushing prediction model development included factor analyses, correlation
analyses and linear and logistic regression.

The data analysis identified that the factors that had the most significant effect on
flushing initiation were pavement surface age, surface thickness, rut depth and the
size of surface aggregates. These factors were used in the prediction model
development, which resulted in developing a two-phase model. The first phase of the
model utilised a logistic model format to predict the probability of flushing initiating
on a given pavement based on the age and thickness of the pavement surface. The
second phase of the model utilised a non-linear model to predict the quantity of
flushing on a pavement based on the age and thickness of the pavement surface, rut
depth on the pavement surface and the grade of aggregates of the surface. The
accuracy of the prediction model was tested using a separate set of LTPP data and it
was found that the logistic model for flushing initiation had statistically significant
accuracy (76%) in correctly predicting the likelihood of flushing initiating on a
pavement, while the non-linear model for flushing progression was also found to be
statistically strong (R2 value of 0.643) and was robust at predicting the progression of
flushing. The developed two-phase model was recommended for use in condition
assessment tasks for pavement networks containing chip seal pavements.

The developed flushing prediction model is the first method of assessment that can be
applied to a pavement network where the model can be used to identify specific
pavement sections that are failing or likely to fail due to flushing. Once a problem site

6-183
has been identified, then laboratory-based assessment methods can be used to conduct
an in-depth inspection of the identified site and to confirm the model findings.
Samples obtained from the identified site can be wheel tracked to determine how its
chip seal structure behaves, whether it is an unstable structure or a stable structure.
The number of wheel passes to use on the samples should be selected to match the
traffic volume (AADT) of the site under consideration, as this will show most
accurately the likely in-field behaviour of the chip seal. If deformation continues to
develop on the samples throughout the wheel tracking tests then the corresponding
surface is an unstable surface, where as if deformation on the samples reaches
maximum level and stabilises, then the corresponding surface is a stable surface.
Further investigations such as CT scanning should be conducted on the samples to
determine the underlying causes of flushing. As was shown in Chapter Four, the
reduction in air void volume calculated from CT scan images had an extremely strong
correlation to flushing, thus, CT scanning will aid with confirming the flushing values
predicted using the data model.

6.2.4 Pavement Condition Assessment Guideline for Flushing

As outlined in the preceding sections, when determining maintenance treatments for a
flushed site, multiple investigative methods must be used for selecting the most
effective maintenance treatment for the site. Combining the outcomes from the
flushing prediction model with laboratory-based testing is a robust and efficient
method of selecting the best maintenance treatment for a potential flushed site. Figure
‎6.1 presents a condition assessment guideline that is recommended for use in
assessing and selecting appropriate maintenance treatments for flushed pavements.

The network-wide application of the assessment guideline presented in Figure ‎6.1 will
aid with establishing a balance of different types of maintenance treatments for the
whole pavement network. Presently, the lack of availability of a flushing prediction
model in pavement management systems means most chip seal surfaces are resealed
on a regular basis to prevent future flushing. While this process delays flushing
development in the short term, maintenance by resealing is ineffective in the long
term. The use of the presented flushing assessment guideline therefore will aid with

6-184
early identification of potential flushing within a pavement network, allowing
selection of the most effective methods of maintenance for specific areas of a
pavement network. Furthermore, identification of the specific maintenance needs of a
pavement network allows efficient allocation of funds within asset management
programmes.

Flushing prediction

Flushing initiated?

Yes - Predict quantity No - Continue

of flushing monitoring

Conduct in-depth
analysis - wheel
tracking of samples

Deformation pattern

Stable deformation - Unstable deformation

flushing present? - flushing present?

No - No No - No
Yes - Conduct Yes - Need for maintenance
maintenance
internal visualisation rehabilitation needed. Continue to
needed. Continue to
assessment monitor
monitor

Air voids present in

wheel tracked area?

No - Stability due
Yes - Stablility
to complete
due to stone-on-
reduction of air
stone state
voids

Use water Need for

blasting rehabilitation

Figure ‎6.1 Assessment of flushing of chip seal pavements

6-185
6.3 Limitations of the Research
6.3.1 Industrial Applications of Research Findings
The presented research demonstrated the value of CT scanning as an in-depth
monitoring method for flushed chip seal surfaces. However, some limitations were
encountered during the research process which highlighted the difficulties that may be
present in adopting the CT scanning technology for widespread use in the pavement
industry at present. In particular, the availability of CT scanners for industrial uses
such as pavement sample scanning is currently limited and this is an issue that
restricts widespread use of the scanning technique. Moreover, the scanner that was
available for this research was only able to scan small samples out of the cores that
were obtained from field pavements. As this research primarily focused on
investigating the variability of air voids in a sample due to loading, the size of the
samples that was used was sufficient to ascertain the potential mechanisms by which
the air voids decreased and caused flushing. If a similar scanning process was to be
used for assessing pavement mix designs, it is recommended that a large number of
samples are used to confirm the validity of the results obtained from scanning.
Nonetheless, even though the CT scanning technology is currently only available as a
research application, it has the potential to be a very strong analysis tool for use in the
pavement industry.

6.3.2 Applicability of the Flushing Prediction Model

The flushing deterioration model that was developed in this research was only
developed to predict flushing on pavements that contained second and higher
generation seals. The data analysis revealed that predicting flushing on first
generation seals, specifically a surface that has not had any maintenance, was difficult
due to the randomness that is present in flushing occurrence on those seals. In most
first generation seals the only reason for flushing is construction faults, which are
difficult to predict, where as in second and higher generation seals flushing occurs
mainly due to the seal characteristics. Having a model only for second and higher
generation seals may reduce the proportion of pavements on which flushing can be
predicted, however in order to develop an accurate model, it was necessary to restrict
the types of pavements that were modelled.
6-186
6.4 Recommendations
6.4.1 Model Calibration and Data Requirements
The developed flushing prediction model was based on data that was available from
the LTPP database. The LTPP database provided pavement condition data from a
wide range of surfacing types, ages, and geographical regions, however, because the
database only included a fixed number of sites from each geographical region of New
Zealand, it was not possible to test the model for climatic conditions of each
individual region. As a result, presently it is not possible to know how well the model
will predict flushing in regional pavement networks. It is recommended that the
flushing model be tested on data from different pavement networks to test its accuracy
and to detect any calibration requirements to account for differences in soil types and
climatic conditions of various geographical regions.

As part of this research, a survey was conducted on a small number of LTPP sites to
collect soil moisture data, specifically measurements of water content and dry density
of soil underneath the pavements. The purpose of this task was to examine the effects
of soil moisture on flushing occurrence, as previous studies had indicated the
possibility of in-pavement moisture causing flushing (Ball et al., 1999). A brief
analysis was conducted on the soil moisture data and this analysis indicated that soil
water content and density did have an effect on flushing. Soil moisture data is not
routinely collected as part of LTPP surveys and therefore, it was not possible to
incorporate soil water content and density in the flushing model. If a larger set of soil
moisture data is available then these two factors could be incorporated into the
flushing model. Thus, it is recommended that the data items that are collected as part
of the LTPP programme be extended to include measurements of soil moisture
characteristics. In addition to LTPP surveys, routine pavement surveys that are
undertaken as part of RAMM could also be extended to incorporate soil moisture
measurements as soil moisture data can be used for regional calibrations of the
flushing model.

6-187
6.4.2 Characterisation of Chip Seal Structures for Accurate
Selection of Maintenance Treatments
The research findings demonstrated that the combination of mechanical testing and
CT scanning can be used to determine the state of stability of a chip seal and to select
the best maintenance treatment for flushing. These two methods are useful especially
when assessing thick chip seal pavements where it is possible to obtain only the chip
seal section for mechanical loading. However, for flushed pavements that are
relatively thin, usually below 30 mm, the use of mechanical loading tests may not be
accurate, and as a result it may not be possible to confirm the best method of
treatment for flushing. Consequently, there is a need to develop a technique that can
characterise the stability of the internal structure of flushed chip seals in order to
easily identify pavements that are likely to show the stable deformation state and
pavements that are likely to show the unstable deformation state. Currently there is
published research that has developed indices for characterising pavement structural
stability in relation to surface defects such as cracking and rutting (Sefidmazgi et al.,
2012; Taniguchi et al., 2008; Tashman et al., 2007), and it is recommended that an
index be developed to characterise the structural stability of a surface layer in relation
to flushing. The development of a stability index can be undertaken using samples
from existing flushed pavements as these will be the best representation of what
happens in the field. This work can utilise digital images of samples to visualise the
aggregate contact areas and angles, which can then be used to determine the structural
stability of a chip seal. A stability index will be especially useful as a routine
assessment measure when it has been identified that flushing had initiated on a
particular chip seal surface. Furthermore, in situations where access to CT scanning is
limited, a structural stability index can be used as an alternative method for internal
structure analysis.

6.4.3 Assessing the Viability of CT Scanning in Practice

A major outcome of the documented research was the usefulness of CT scanning in
the assessment of chip seal surfaces, particularly for assessing the micromechanical
changes that take place within flushed chip seal pavements. One of the areas that were
identified in the process of this research that needed further investigation was into the

6-188
effect of sample size on the accuracy of air voids data that was gathered through the
CT scanning process. The conducted analysis concentrated on the changes to the air
void volumes in the chip seal cores and any sample size-effects on the air void volume
results were not considered. This was due to two reasons; firstly, because of features
of the available scanner, it was only possible to scan samples that were less than
65 mm in diameter, and secondly, as the analysis was mainly concerned with the
changes that had occurred to the air voids, the sample size-effects were thought to be
not significant. There is a need to establish exactly how the sample size impacts on
the data that can be gathered from CT scanning, particularly when trying to predict the
behaviour of flushed chip seal structures. It is recommended that scanning be
conducted on a large number of samples with differing sizes to establish the
agreement between air voids results of different sample sizes as well as to determine
the applicability of the scan image-based air voids to the designed air voids value.

6-189
CHAPTER 7 - CONCLUSIONS

7.1 Research Conclusions

There were three main objectives to the documented research. These objectives were
to:
1. Develop a mechanistic understanding of the processes involved with the
flushing defect;
2. Understand the micromechanical behaviour and interactions that occur
between the materials that make up a chip seal surface; and
3. Develop a pavement deterioration model that can determine the progression of
flushing on chip seal pavements.

Three main methodological components were utilised in achieving the above

objectives. These methodological components were mechanical testing of chip seal
samples, CT scanning and image analysis of chip seal samples, and analysis of
pavement condition data and flushing deterioration model development. The
mechanical testing component and image analysis component together contributed to
understanding the fundamental behaviour of chip seal layers that results in flushing.
The analysis of pavement condition data was separate from the mechanical testing and
image analysis components, where the data analysis resulted in the development of a
flushing forecasting model based on network-level pavement data. Figure ‎7.1
demonstrates how these methodological components added to the understanding of
flushing.

7-190
 Understanding flushing of
chip seals

Develop a mechanistic Understand the Develop a pavement

understanding of the micromechanical deterioration model to
processes involved with interactions that occur predict the progression of
flushing between chip seal materials flushing

Laboratory testing Image analysis Data analysis and modelling

Overall understanding of
the flushing defect

Better identification and

Improvements to chip seal
prediction methods for
surface design
flushing

Figure ‎7.1 Main methodological components of the research

Objective One: Develop a mechanistic understanding of the processes

involved with the flushing defect.
This objective aimed to identify the main causes of flushing on chip seal pavements as
well as to understand the effects of other defects such as rutting on the flushing
mechanism. The research outcomes indicated that the internal characteristics of a chip
seal had the most significant effect on flushing. In particular, an increase in surface
thickness and surface age were two main causes of flushing, where thick surfaces
often behaved in an unstable manner and led to flushing. The reduction in air voids
was the other major cause of flushing. Investigation into the effect of rutting

7-191
(deformation) revealed that observing the pattern of deformation of a chip seal surface
was a useful method for determining the behaviour of a flushed surface and for
selecting the best method of treatment for flushing. For example, when deformation
reaches a stable state on a chip seal surface, any subsequent flushing can be treated
using methods such as UHP water cutting, while when deformation continues to
develop on a chip seal surface, the resulting flushing needs to be treated using
pavement rehabilitation. This relationship between flushing and deformation was not
examined in previous literature and the understanding that was gained by observing
the flushing/deformation relationship can be used as a way to conduct in-depth
assessment of pavements identified as flushed.

Objective Two: Understand the micromechanical behaviour and

interactions that occur between the materials that make up a chip seal surface.
This objective was focused on determining how loading affected the volume and
distribution of air voids within a chip seal layer and how that affected flushing. This
objective also aimed to determine the effects that the aggregate grading profile of a
chip seal surface had on its flushing potential. Furthermore, this objective focused on
investigating the effectiveness of using image analysis techniques for quantifying the
micromechanical changes that occurred in a chip seal layer due to loading.

The analysis that was conducted to answer this objective revealed that the reduction in
air void volume due to loading had a strong correlation with flushing and that air void
volume reduction was an effective indicator of the amount of flushing likely to
develop on a particular chip seal surface. Moreover, the air void volume reduction
analysis revealed that the main cause of the reduction in air void volume was the
presence of excessive amounts of bitumen within the chip seal surface. It was found
that samples that had a high bitumen-to-air void volume ratio displayed higher
amounts of flushing when compared to samples that had low bitumen-to-air void
volume ratios. This finding confirmed that the volumetric balance between the
bitumen and air voids was vital to the development of flushing.

The‎reduction‎of‎air‎voids‎during‎loading‎followed‎a‎similar‎pattern‎to‎ a‎chip‎ seal’s‎

pre-loading conditions, where air voids were left at the aggregate layer interfaces. The
7-192
air void volume of a loaded chip seal can be used to confirm the state of stability of a
chip seal and to determine the cause of the stability, for example whether a chip seal
layer had reached a stable state due to reaching the stone-on-stone state or whether it
was due to all air voids being expelled. The imaging method of CT scanning which
was used in the assessment of air void volume reduction was found to be an extremely
useful analysis method for assessing chip seals that were prone to flushing.
Additionally, investigation into the relationship between flushing and aggregate
grading properties revealed that the proportion of aggregates that was classed as
“fines”‎ was‎ higher on cores that displayed an unstable deformation pattern when
compared to cores that displayed a stable deformation pattern. However, the fines
content did not have a direct effect on flushing. It was concluded that the fines content
in a chip seal surface does affect flushing, but this effect is indirect.

Objective Three: Develop a pavement deterioration model that can determine

the progression of flushing on chip seal pavements.
This objective was focused on developing a prediction model to identify flushed chip
seal pavements. Flushing is one of the most prevalent defects on chip seal pavements
in New Zealand, however, it is often not modelled in pavement management systems
as‎currently‎there‎is‎no‎model‎incorporated‎in‎World‎Bank’s‎HDM-4 and NZdTIMS
system to forecast flushing. The development of the flushing prediction model was
undertaken‎ using‎ pavement‎ condition‎ data‎ collected‎ through‎ New‎ Zealand’s‎ LTPP‎
database. From the data analysis it was found that the combination of variables that
provided the best indication of flushing were surface thickness, surface age, grade of
aggregates of the surface and the rut depth on the pavement surface. The data analysis
was also used to establish a threshold to indicate when flushing had initiated on a
pavement. This threshold for flushing initiation was established as when flushing was
displayed on 20% of the pavement surface area. Once this flushing initiation threshold
was established, a logistic model was developed to predict whether flushing had
initiated on a pavement. Further to this, a nonlinear model was also developed to
predict the quantity of flushing on a pavement. Both components of the prediction
model were tested using a set of LTPP data that was not used in the model
development and it was found that the model had statistically significant strength in

7-193
its predictions. In order to make the developed model applicable to non-state highway
pavements, the model will need to be calibrated to suit regional conditions.

7.2 Overall Understanding of Chip Seal Behaviour

Developed Through the Research Findings
The research results showed that flushing develops due to a combination of internal
characteristics of a chip seal surface and external factors, such as traffic volume and
environmental conditions. Of these two components, the internal factors are the most
important to the behaviour of a chip seal surface. The manner in which a chip seal
surface was designed and the volumetric balance between the constituents that form a
chip seal surface determine whether or not a surface will flush. The observations
made from both the laboratory testing and the data analysis revealed that the thickness
of a chip seal had a significant impact on flushing development. Additionally, the
availability of air voids within a chip seal surface was vital to its functional
performance.

It is important to note that external factors, especially traffic volume, also play a key
role in flushing occurrence. Compaction due to traffic loading initiates the processes
that result in air void volume reduction which in turn leads to flushing. However, the
extent to which traffic loading contributes to flushing development depends on how
well a chip seal surface is designed. A well designed surface that is lower in thickness
and has an adequate volume of air voids to allow binder movement will perform
satisfactorily when subjected to traffic volumes for which it was designed. In contrast,
a poorly designed chip seal surface that is high in thickness will flush even when
subjected to traffic volumes well below its design traffic volume. This occurrence was
evident in the data analysis results which demonstrated that some sites exhibited high
amounts of flushing at relatively low traffic volumes while other sites that had high
traffic volumes had significantly low amounts of flushing. This observation indicated
that traffic volume on its own does not cause major flushing. The explanation for this
observation is the use of effective chip seal design practices. The chip seal design
practises that are currently used in New Zealand adjust the quantity, grade and spray
rate of binder to cater for varying environmental and traffic conditions. It was
7-194
identified that these adjustments are the reason why much of the premature flushing
failures caused by traffic loading has been minimised. Thus, the current chip seal
design practice of adjusting binder quantity, grade and spray rate to suit local
conditions of a site should be continued to achieve well-performing seals.

While some aspects of current chip seal design practices are working well, there are
some sealing practices that are contributing to flushing, particularly the practice of
resealing to treat flushing. A flushed chip seal pavement often has excessive amounts
of binder on its surface and the use of reseals, such as single coat seals, for treating
flushing simply adds more bitumen on to the surface. The thickness that results from
numerous reseals over time on a pavement contributes to instability and affects the
volumetric balance between the chip seal constituents. This occurrence was clearly
demonstrated by the research results which showed that flushing developed as a direct
result of having an excessive volume of bitumen in comparison to the available
volume of air voids within a chip seal surface. Therefore, the practice of resealing of
flushed surfaces needs to be minimised to prevent the development of thick chip seal
surfaces that are prone to flushing. It is important to note that given the thin, granular
nature of many of the pavements that are used in low traffic volume areas, it is vital
that the pavement surfaces are made impermeable to prevent sub-surface deterioration
and to prolong the pavement life. Resealing remains the most cost-effective method of
achieving a prolonged pavement life despite some of its drawbacks. From the results
that were obtained in this research it was evident that the decision making process for
selecting maintenance treatments for flushed pavement surfaces had failed in some
situations, such as the scenario mentioned earlier of the build-up of chip seal layers as
a result of continuous resealing. These failures reconfirm the importance and need for
the documented research outcomes in order to improve current pavement maintenance
practices and to aid the decision making process for selecting appropriate maintenance
treatments for flushed pavement surfaces. The use of maintenance methods such as
pavement rehabilitation or water cutting were identified as being effective for treating
flushing. While these treatments may have a high initial cost, the long-term
improvements in performance gained by using these treatments would far outweigh
the short-term improvements made through resealing. Moreover, the use of abrasion-
resistant aggregates with strong interlocking properties, such as steel slag aggregates,
for constructing chip seal surfaces will contribute to creating stable surfaces that can
7-195
resist excessive compaction and minimise flushing. The use of steel slag aggregates
may be expensive in the short-term, however the long-term benefits that can be
achieved through the use of these aggregates would compensate for the initial cost.

In addition to understanding the behaviour of chip seal surfaces, the laboratory test
component of the research highlighted some important pavement measures that could
aid with better prediction of flushing on chip seal pavements. One of the most
significant findings of the CT scanning process was the strength in the relationship
between flushing and the volume of air voids, where it was found that the reduction in
the volume of air voids was strongly correlated to flushing. This finding highlighted
the value of the availability of air void volume data for inclusion in the flushing
prediction model. It is expected that the inclusion of information about the state of air
void volume of a chip seal site in pavement management databases would be of
immense value when conducting in-depth analysis of flush-prone sites.

Overall, the presented research has greatly increased the understanding of the
mechanisms that cause flushing. Prior to starting this research there were limited
studies that had investigated how the micromechanical interactions and the volumetric
changes that occur between materials within a chip seal structure contributed to
flushing. The documented research utilised laboratory-based mechanical testing to
simulate flushing on chip seal samples in order to observe the processes that led to
flushing. The investigation into the micromechanical interactions between chip seal
materials was extended by combining the laboratory testing results with image
analyses techniques to observe the volumetric changes that took place during loading
and to determine how those volumetric changes contributed to flushing. The
development of the flushing prediction model added to the understanding of flushing.
Through this research, more efficient and effective methods for identifying and
assessing flushing on chip seal pavement networks have been identified.

7-196
REFERENCES

ABDUL-MALAK, M. U., FOWLER, D. W. & CONSTANTINO, C. A. (1996)

Aggregate Characteristics Governing Performance of Seal Coat Highway
Overlays. Transportation Research Record, 1547, 15-22.
ALDERSON, A. (2008) Flushing Distress Mechanism in Seals. 1st Sprayed Sealing
Conference - Cost Effective High Performance Surfacings. Adelaide,
Australia, July 2008, ARRB Group Ltd, Vermont South, Victoria, Australia.
ASTM INTERNATIONAL (1995) ASTM D2041 - 95 Standard Test Method for
Theoratical Maximum Specific Gravity and Density of Bituminous Paving
Mixtures.
ASTM INTERNATIONAL (2005a) ASTM D2726 - 05a Standard Test Method for
Bulk Specific Gravity and Density of Non-Absorptive Compacted Bituminous
Mixtures. DOI: 10.1520/D2726-10.
ASTM INTERNATIONAL (2005b) ASTM D3203 - 05 Standard Test Method for
Percent Air Voids in Compacted Dense and Open Bituminous Paving
Mixtures. DOI: 10.1520/D3203-05.
ASTM INTERNATIONAL (2006) ASTM C136 - 06 Standard Test Method for Sieve
Analysis of Fine and Coarse Aggregates. DOI: 10.1520/C0136-06.
ASTM INTERNATIONAL (2011) ASTM D2172M - 11 Standard Test Methods for
Quantitative Extraction of Bitumen From Bituminous Paving Mixtures. DOI:
10.1520/D2172-11.
AUSTROADS (2004) Sprayed Sealing Guide. Austroads Publication No. AP-G76/04.
Sydney, Australia, Austroads Incorporated.
AUSTROADS (2006a) AGPT/T231 Deformation Resistance of Asphalt Mixtures by
the Wheel Tracking Test.
http://www.austroads.com.au/pavement/testmethods.html.
AUSTROADS (2006b) Maintenance of Asphalt Surfacings. Austroads Publication
No. AP-T67/06. Sydney, Australia, Austroads Incorporated.
AUSTROADS (2008) Seal Distress Mechanisms: An Initial Study into Flushing.
Austroads Publication No. AP-T108/08. Sydney, Australia, Austroads
Incorporated.
AUSTROADS (2009) Guide to Asset Management - Part 5H: Performance
Modelling. Austroads Project No: AP1177. Sydney, Australia, Austroads
Incorporated.
BALL, G. F. A. (1997) End of Life Chip Seal Conditions - Stage 2, Progress Report:
Chipseal Lifetimes in New Zealand. Transfund New Zealand Research Project
PR3-0126.
BALL, G. F. A., LOGAN, T. C. & PATRICK, J. E. (1999) Flushing Processes in
Chipseals: Effects of Water. Transfund New Zealand Research Report 156.
BALL, G. F. A., PATRICK, J. & HERRINGTON, P. (2004) Solutions for Improving
Chipseal Life. Central Laboratories Report: 04-521071.00. Opus International
Consultants Limited

197
BALL, G. F. A. & PATRICK, J. E. (1998) Flushing Processes in Chipseals: Effects of
Trafficking. Transfund New Zealand Research Report 122.
BALL, G. F. A. & PATRICK, J. E. (2005) Failure Modes and Lifetimes of Chipseals
on New Zealand State Highways. Central Laboratories Report: 05-521071.01.
Opus International Consultants.
BALL, G. F. A. & PATRICK, J. E. (2008) Resealing Strategies to Increase Seal Life
and Prevent Seal Layer Instability. NZ Transport Agency Research Report
372.
BALL, G. F. A., PATRICK, J. E. & HERRINGTON, P. R. (2005) Factors Affecting
Multiple Chipseal Layer Instability. Land Transport New Zealand Research
Report 278.
BARBU, V. S. & LIMNIOS, N. (2008) Semi-Markov Chains and Hidden Semi-
Markov Models Toward Applications: Their Use in Reliability and DNA
Analysis, Springer, New York. United States of America. ISBN 978-0-387-
73171-1.
BEAR, J. (1972) Dynamics of Fluids in Porous Media, Dover, New York, United
States of America.
BENNETT, C. R. (1996) The HDM-4 Model. Combined ARRB Transport Research
Conference / Transit NZ Land Transport Symposium. Christchurch, New
Zealand. 2-6 September.
BERGER, J. O. (1985) Statistical Decision Theory and Bayesian Analysis, Second
ed., Springer Series in Statistics. Springer-Verlag. ISBN 0-387-96098-8.
BERNARDO, J. M. & SMITH, A. F. M. (1994) Bayesian Theory, Wiley. ISBN 0-
471-49464-X.
BLACK, P. M. (2009) Geological Inventory of North Island Aggregate Resources:
Influences on Engineering Material Properties.
BOWEN, R. M. (1976) Theory of Mixtures, Part 1, Eringen, A.C. ed., Continuum
Physics III, Academic Press, New York, United States of America.
BRZEZICKI, J. M. & KASPERKIEWICZ, J. (1999) Automatic Image Analysis in
Evaluation of Aggregate Shape. Journal of Computing in Civil Engineering,
13, 123-128.
BUTTLAR, W. G., BOZKURT, D., AL-KHATEEB, G. G. & WALDHOFF, A. S.
(1999) Understanding Asphalt Mastic Behaviour through Micromechanics.
Transportation Research Record, 1681.
BYERS, C. C., SALEH, M. F. & PIDWERBESKY, B. D. (2004) Evaluation of
Melter Slag as a Base Course Material. The International Journal of Pavement
Engineering, 5, 193-199.
BYRNE, M., SANJAYAN, J., ALBERCHT, D., KODIKARA, J. & MARTIN, T.
(2005) Application of Data Mining in Pavement Performance Modelling - A
Case Study. Road & Transport Research, 14 (4), 27-44.
COHEN, J. (1988) Statistical Power Analysis for the Behavioural Sciences. 2nd Ed.,
Hillsdale, New Jersey, USA, Lawrence Erlbaum Associates, Inc.
COHEN, J. (1992) A Power Primer. Psychological Bulletin, 112, 155-159.
COREE, B. J. (1999) Hot-Mix Asphalt Volumetrics Revisited - A New Paradigm.
Transportation Research Record, 1681, 50-58.
COSTELLO, S. B., PREMATHILAKA, A. S. & DUNN, R. C. M. 2012. Stochastic
Rail Wear Model for Railroad Tracks. Transportation Research Record:
Journal of the Transportation Research Board, 2289, 103–110.
CHRISTIE, A. B. & BARKER, R. G. (2007) Mineral Resource Assessment of the
Northland Region, New Zealand GNS Science Report 2007/06, 179.
198
DAHIR, S. H. (1979) A Review of Aggregate Selection Criteria for Improved Wear
Resistance and Skid Resistance of Bituminous Surfaces. Journal of Testing
and Evaluation, 7, 245-253.
DE FARIAS, M. M. & DE SOUZA, R. O. (2009) Correlations and Analyses of
Longitudinal Roughness Indices. Road Materials and Pavement Design, 10,
399-415.
FLINTSCH, G. W., DE LEON, E., MCGHEE, K. K. & AL-QADI, I. L. (2003)
Pavement Surface Macrotexture Measurement and Applications.
Transportation Research Record: Journal of the Transportation Research
Board, 1860, 168-177.
FROST, J. D. (1999) Editorial. Journal of Computing in Civil Engineering, 13, 59-60.
FUENTES, L., GUNARATNE, M. & HESS, D. (2010) Evaluation on the Effects of
Pavement Roughness on Skid Resistance. Journal of Transportation
Engineering, 136.
FURLONG, W., HENNING, T. & HOWELL, A. (2005) dTIMS CT Conceptual
Development for Low Volume Roads. Land Transport New Zealand.
GRANSBERG, D. & JAMES, D. M. B. (2005) Chip Seal Best Practices. National
Cooperative Highway Research Program, Synthesis 342. Transportation
Research Board.
GRANSBERG, D. D. 2007. Using a New Zealand Performance Specification to
Evaluate U.S. Chip Seal Performance. Journal of Transportation Engineering,
133.
GRANSBERG, D. D. & PIDWERBESKY, B. (2007) Strip Sealing and Ultra-High-
Pressure Watercutting Technique for Restoring Skid Resistance on Low-
Volume Roads Life-Cycle Cost comparison. Transportation Research Record:
Journal of the Transportation Research Board, 1989, 234-239.
GRAY, W. & HART, G. (2003) Recycling of Chip Sealed Pavements - New Zealand
Experience in Combating Top Surface Layer Instability Issues. The XXIInd
PIARC World Road Congress. Durban, South Africa.
HAJEK, J. J. & HAAS, R. C. G. (1987) Factor Analysis of Pavement Distresses for
Surface Condition Predictions. Transportation Research Record, 1117, 125-
133.
HENDERSON, R., CENEK, P., JAMIESON, N. & WILSON, D. J. (2011) The
Influence of Binder Rise in Reducing Tire-Road Friction. NZ Transport
Agency Research Report 447. 46pp.
HENDERSON, R., HERRINGTON, P., PATRICK, J., KATHIRGAMANATHAN, P.
& COOK, S. (2011) Analysis of Particle Orientation in Compacted Unbound
Aggregate. Road Materials and Pavement Design, 12 (1).
HENNING, T., DUNN, R., COSTELLO, S. & PARKMAN, C. (2009) A New
Approach for Modelling Rutting on the New Zealand State Highways. Road &
Transport Research, 18 (1), 3-18.
HENNING, T. F. P. (2008) The Development of Pavement Deterioration Models on
the State Highway Network of New Zealand. Doctoral Dissertation -
Department of Civil Engineering - Faculty of Engineering. The University of
Auckland. http://researchspace.auckland.ac.nz/handle/2292/4236.
HENNING, T. F. P., COSTELLO, S. B., DUNN, R. C. M., PARKMAN, C. C. &
HART, G. (2004) The Establishment of a Long-Term Pavement Performance
Study on the New Zealand State Highway Network. Road & Transport
Research, 13 (2), 17-32.

199
HENNING, T. F. P., COSTELLO, S. B. & WATSON, T. G. (2006) A Review of the
HDM/dTIMS Pavement Models based on Calibration Site Data. Land
Transport New Zealand Research Report 303.
HENNING, T. F. P. & ROUX, D. C. (2008) Pavement Deterioration Models for
Asphalt-Surfaced Pavements in New Zealand NZ Transport Agency Research
Report 367. NZ Transport Agency.
HOCKING, R. R. (2003) Methods and Applications of Linear Models: Regression
and the Analysis of Variance. 2nd ed., Hoboken, N.J., USA, Wiley-
Interscience.
HOSMER, D. W. & LEMESHOW, S. (1989) Applied Logistic Regression, Wiley,
New York.
IBM CORPORATION (2010) SPSS for Windows, release 19.0.0. Somers, New York,
United States of America.
KENDALL, M. G. & STUART, A. (1969) The Advanced Theory of Statistics.,
London, Griffin, v2 - 3rd ed.
KERALI, H. R. (2000) Overview of HDM-4. The Highway Development and
Management Series, Volume 1.
KLEINBAUM, D. G. & KLEIN, M. (2002) Logistic Regression: A Self-learning Text
(2nd ed.), Springer-Verlag, New York.
KODIPPILY, S., HENNING, T. F. P. & INGHAM, J. M. (2012) Detecting Flushing
of Thin Sprayed Seal Pavements using Pavement Management Data. Journal
of Transportation Engineering 138, 665-673.
KRISHNAN, J. M. & RAJAGOPAL, K. R. (2005) On the Mechanical Behavior of
Asphalt. Mechanics of Materials, 37, 1085-1100.
KRISHNAN, J. M. & RAO, C. L. (2000) Mechanics of Air Voids Reduction of
Asphalt Concrete using Mixture Theory. International Journal of Engineering
Science, 38, 1331-1354.
KRISHNAN, J. M. & RAO, C. L. (2001) Permeability and Bleeding of Asphalt
Concrete using Mixture Theory. International Journal of Engineering Science,
39, 611-627.
LAWSON, W. D. & SENADHEERA, S. (2009) Chip Seal Maintenance - Solutions
for Bleeding and Flushed Pavement Surfaces. Transportation Research
Record: Journal of the Transportation Research Board, 2108, 61-68.
LI, W., SUN, P. & ZHANG, C. (2012) Laboratory Test Study on Asphalt Concrete
with Steel Slag Aggregates. Applied Mechanics and Materials, 152-154, 117-
120.
LYTTON, R. L. (1987) Concepts of Pavement Performance Prediction and
Modelling. Second North American Conference on Managing Pavements.
Toronto, Canada. November.
MAHMOUD, E., GATES, L., MASAD, E., ERDOGAN, S. & GARBOCZI, E.
(2010) Comprehensive Evaluation of AIMS Texture, Angularity, and
Dimension Measurements. Journal of Materials in Civil Engineering, 22, 369-
379.
MARTINO, M. M. (2007) Evaluation of a System for Measuring Seal Coat Quality.
Master of Science dissertation. The University of Texas at San Antonio.
MASAD, E., MUHUNTHAN, B., SHASHIDHAR, N. & HARMAN, T. (1999a)
Internal Structure Characterization of Asphalt Concrete using Image Analysis.
Journal of Computing in Civil Engineering, 13, 88-95.

200
MASAD, E., MUHUNTHAN, B., SHASHIDHAR, N. & HARMAN, T. (1999b)
Quantifying Laboratory Compaction Effects on the Internal Structure of
Asphalt Concrete. Transportation Research Record, 1681, 179-185.
MASAD, E. (2003) The Development of a Computer Controlled Image Analysis
System for Measuring Aggregate Shape Properties. NCHRPIDEA Final Rep.
No. 77. National Cooperative Highway Research Program, Washington, D.C,
United States of America.
MASAD, E. & SIVAKUMAR, K. (2004) Advances in the Characterization and
Modeling of Civil Engineering Materials Using Imaging Techniques Journal
of Computing in Civil Engineering, 18.
MOROSIUK, G. & KERALI, H. R. (2001) The Highway Development and
Management Tool - HDM-4. IKRAM's Seminar on Asphalt Pavement
Technology (ISAPT 2001). Kuala Lumpur.
NDLI (1995) Modelling Road Deterioration and Maintenance Effects in HDM-4.
Final Report ADB BETA 5549. Vancouver.
O'FLAHERTY, C. A. (2002) Highways: The Location, Design, Construction and
Maintenance of Road Pavements. 4th Ed, Oxford, Butterworth-Heinemann.
ISBN 9780750650908.
OLIVER, J. (1999) The Performance of Sprayed Seals. Research Report ARR 326.
ARRB Transport Research Ltd.
ORTIZ-GARCÍA, J. J., COSTELLO, S. B. & SNAITH, M. S. 2006. Derivation of
Transition Probability Matrices for Pavement Deterioration Modeling. Journal
of Transportation Engineering, 132, 141-161.
OWUSU-ABABIO, S. (1998) Application of Neural Networks to Modeling Thick
Asphalt Pavement Performance. Proceedings of the International Symposium
on Artificial Intelligence and Mathematical Methods in Pavement and
Geomechanical Systems. Miami, Florida, USA. 5-6 November. p 23-30.
OZBAY, K. & LAUB, R. (2001) Models for Pavement Deterioration using LTPP.
Washington, Federal Highway Administration, US Department of
Transportation.
PALLANT, J. (2007) SPSS Survival Manual: A Step by Step Guide to Data Analysis
using SPSS for Windows (Version 15). 3rd ed., Crows Nest, NSW, Australia,
Allen & Unwin.
PAN, T., TUTUMULER, E. & CARPENTER, S. H. (2006) Effect of Coarse
Aggregate Morphology on Permanent Deformation Behaviour of Hot Mix
Asphalt. Journal of Transportation Engineering, 132.
PARK, T. (2007) Causes of Bleeding in a Hot-In-Place Asphalt Pavement.
Construction and Building Materials, 21, 2023-2030.
PARKMAN, C. & RAINSFORD, S. (2005) Predicting Texture Deficiency in
Pavement Management. Prepared for Transit New Zealand.
PATRICK, J. E. & DONBAVAND, J. (1996) Development of a Performance Based
Chipseal Specification. Combined 18th ARRB Transport Research Conference
and Transit New Zealand Land Transport Symposium, Roads 96.
Christchurch, New Zealand. 2-8 September.
RAGHURAJ RAO, K. & LAKSHMINARAYANAN, S. (2007) Partial Correlation
Based Variable Selection Approach for Multivariate Data Classification
Methods. Chemometrics and Intelligent Laboratory Systems, 86, 68-81.
RASBAND, W. S. (1997-2011) ImageJ. http://rsbweb.nih.gov/ij/. U. S. National
Institutes of Health, Bethesda, Maryland, USA.

201
READ, J. & WHITEOAK, D. (2003) The Shell Bitumen Handbook. Fifth Edition,
Thomas Telford Publishing.
ROSS, S. M. (1999) Stochastic processes. 2nd ed., Routledge, New York, United
States of America. ISBN 978-0-471-12062-9.
SALAMA, H. K., CHATTI, K. & HAIDER, S. W. (2008) Regression Models for
Permanent Deformation Parameters using In-Service Pavement Data from the
SPS-1 Experiment. International Journal of Pavement Engineering, 9, 317-
328.
SCHWEYER, H. E. (1973) Asphalt Rheology in the Near-Transition Temperature
Range. Highway Research Record, 468, 1-15.
SEFIDMAZGI, N. R., TASHMAN, L. & BAHIA, H. (2012) Internal Structure
Characterization of Asphalt Mixtures for Rutting Performance using Imaging
Analysis. Road Materials and Pavement Design, 13, 433-454.
SHASHIDHAR, N. (1999) X-Ray Tomography of Asphalt Concrete. Transportation
Research Record, 1681, 186-192.
SKYSCAN (2010) Manual for SkyScan CT-Analyser v.1.10.
www.skyscan.be/next/CTan_UserManual.pdf. Aartselaar, Belgium.
SMITH, B. C. & DIEFENDERFER, B. K. (2008) Comparison of Nuclear and
Nonnuclear Pavement Density Testing Devices. Transportation Research
Record: Journal of the Transportation Research Board, 2081, 121-129.
STORM, D. A., BARRESI, R. J. & SHEU, E. Y. (1996) Development of Solid
Properties and Thermochemistry of Asphalt Binders in the 25-65°C
Temperature Range. Energy and Fuels, 10, 855-864.
SYMONS, P. N. (1985) A Method of Road Pavement Condition Projection. Bureau
of Transport Economics Occasional Paper 64. Canberra, Australia, Australian
Government Publishing Service.
TAHERKHANI, H. & COLLOP, A. (2008) Volumetric Behaviour of Asphalt
Mixtures. Proceedings of the 1st International Conference on Transport
Infrastructure. Beijing, China.
TAN, P., STEINBACH, M. & KUMAR, V. (2005) Introduction to Data Mining,
Pearson Education, Inc.
TANIGUCHI, S., NISHIZAKI, I. & MORIYOSHI, A. (2008) A Study of
Longitudinal Cracking in Asphalt Pavement using CT Scanner. Road
Materials and Pavement Design, 9, 549-558.
TASHMAN, L., WAN, L. & THYAGARAJAN, S. (2007) Microstructure
Characterization for Modelling HMA Behaviour using Imaging Technology.
Road Materials and Pavement Design, 8, 207-238.
TRANSFUND NEW ZEALAND (1997) RAMM - Road Condition Rating and
Roughness Manual. Manual Number PFM6.
TRANSIT NEW ZEALAND (2003) Safety Related Surface Issues, Final Report.
TRANSIT NEW ZEALAND (2005) Specifications for Asphaltic Concrete. TNZ
M/10:2005.
TRANSIT NEW ZEALAND (2007) New Zealand's Long Term Pavement
Performance (LTPP) Programme.
TRANSIT NEW ZEALAND, ROAD CONTROLLING AUTHORITIES &
ROADING NEW ZEALAND (2005) Chipsealing in New Zealand,
Wellington, New Zealand.
UL-ISLAM, R. (2006) Flushing Model - Low Volume Roads. prepared for Land
Transport NZ.

202
VISAGE IMAGING & KONRAD-ZUSE-ZENTRUM FUR
INFORMATIONSTECHNIK BERLIN (2012) Amira User's Guide. Version
5.4. http://www.amira.com/images/stories/pdf/542/AmiraUsersGuide.pdf.
Berlin, Germany.
WANG, L., WANG, X., MOHAMMAD, L. & WANG, Y. (2004) Application of
Mixture Theory in the Evaluation of Mechanical Properties of Asphalt
Concrete. Journal of Materials in Civil Engineering, 16, 167-174.
WATERS, J. C. (2008) Developing and Trialling a Climate-Based Selection
Guideline for Chipseal Binders. NZ Transport Agency Research Report 358.
WEISSMANN, J. & MARTINO, M. M. (2009) Evaluation of Seal Coat Performance
by using Macrotexture Measurements. Transportation Research Board 88th
Annual Meeting. Washington DC, United States of America. 11-15 January,
Transportation Research Board.
WENINGER-VYCUDIL, A., SAMEK, G. & ROHRINGER, T. (2008) Section Based
Probabilistic Performance Prediction. 3rd European Pavement and Asset
Management Conference. Coimbra, Portugal. 7-9 July, University of Coimbra,
Coimbra, Portugal.
WHITEHEAD, D., DONBAVAND, J. & MITCHELL, J. (2011) Developing an
Objective Measure for Flushing. 3rd International Surface Friction
Conference, Safer Road Surfaces - Saving Lives. Gold Coast, Australia.
YANG, J., LU, J. J., GUNARATNE, M. & XIANG, Q. (2003) Forecasting Overall
Pavement Condition with Neural Networks, Application on Florida Highway
Network. Transportation Research Record, 1853, 3-12.

203
APPENDIX A – INVENTORY OF CORES

Table A.1 Inventory of cores used for wheel tracking

Thickness Diameter Mass (Air) Mass (SSD) Mass (Water) Wheel Final rut
Core ID Core No. Region Site
(mm) (mm) (g) (g) (g) passes depth (mm)
A.S1.4 1 Auckland 1 48.8 205.6 3389.6 3399.8 1895.4 2,000 14.2

A.S1.3 2 Auckland 1 49.3 204.5 3416.0 3425.5 1927.6 5,000 13.7

A.S1.1 3 Auckland 1 45.0 203.4 3092.9 3100.8 1731.7 10,000 9.9

A.S2.2 1 Auckland 2 36.9 204.7 2422.1 2437.5 1381.0 2,500 4.9

A.S2.1 2 Auckland 2 39.5 205.4 2717.9 2732.0 1539.1 5,000 9.9

A.S2.4 3 Auckland 2 48.7 204.1 3493.8 3514.7 2034.5 10,000 8.5

H.S1.5 1 Waikato 1 36.5 194.5 2065.6 2075.1 1118.0 5,000 6.4

H.S1.8 2 Waikato 1 40.0 196.5 2333.6 2343.2 1255.7 10,000 7.4

H.S1.1 3 Waikato 1 41.2 193.4 2456.4 2468.2 1348.4 20,000 8.5

H.S1.2 1 Waikato 2 35.6 193.8 2014.8 2024.5 1095.2 1,000 5.7

H.S1.9 2 Waikato 2 36.3 193.4 2098.3 2103.7 1141.9 2,500 8.9

H.S1.6 3 Waikato 2 44.5 194.5 2596.7 2609.5 1419.0 5,000 7.7

204
C.S1.8 1 Christchurch 1 30.5 243.0 2953.7 2969.9 1653.1 2,500 9.3

C.S1.12 2 Christchurch 1 31.0 243.5 2631.9 2652.8 1437.8 5,000 13.5

C.S1.9 3 Christchurch 1 30.0 243.5 2882.5 2896.7 1604.7 10,000 14.7

C.S2.7 1 Christchurch 2 26.3 243.8 2479.3 2484.0 1387.8 2,500 3.4

C.S2.3 2 Christchurch 2 29.5 243.8 2556.4 2561.0 1442.9 5,000 4.7

C.S2.10 3 Christchurch 2 30.0 242.8 2529.5 2536.7 1421.4 10,000 5.0

D.S1.8 1 Dunedin 1 43.0 220.0 3359.8 3366.5 1911.1 2,500 4.0

D.S1.11 2 Dunedin 1 37.3 220.3 3083.6 3089.4 1734.4 5,000 6.7

D.S1.12 3 Dunedin 1 35.3 220.0 3006.7 3013.7 1686.3 10,000 9.1

D.S2.2 1 Dunedin 2 48.0 220.0 3932.4 3952.5 2226.7 2,500 10.4

D.S2.3 2 Dunedin 2 44.0 219.8 3952.2 3962.6 2260.3 5,000 12.8

D.S2.4 3 Dunedin 2 49.5 219.3 4000.0 4009.2 2265.9 10,000 8.9

205
APPENDIX B – BITUMEN CONTENT, AGGREGATE GRADING, AND AGGREGATE
ANGULARITY MEASUREMENTS

Table B.1 Bitumen content measurements for chip seal cores

Binder content (% by
Region Site Core ID Mass of sample (g) Mass of aggregates (g) Mass of binder (g) Air voids in mix (%)
mass)
Auckland 1 A.S1.4 2857.9 2606.4 251.5 8.8 4.4

Auckland 1 A.S1.3 2775.4 2547.8 227.6 8.2 4.4

Auckland 1 A.S1.1 2955.0 2706.8 248.2 8.4 4.4

Auckland 2 A.S2.2 2248.4 2095.5 152.9 6.8 4.3

Auckland 2 A.S2.1 2258.8 2105.2 153.6 6.8 4.3

Auckland 2 A.S2.4 1731.8 1614.0 117.8 6.8 4.3

Waikato 1 H.S1.5 1666.5 1493.2 173.3 10.4 6.1

Waikato 1 H.S1.8 1894.9 1705.4 189.5 10.0 6.1

Waikato 1 H.S1.1 1945.3 1764.4 180.9 9.3 6.1

Waikato 2 H.S1.2 1602.4 1458.2 144.2 9.0 5.4

Waikato 2 H.S1.9 1706.3 1551.0 155.3 9.1 5.4

206
 Waikato 2 H.S1.6 2087.8 1856.1 231.7 11.1 5.4

Christchurch 1 C.S1.8 2289.0 2085.3 203.7 8.9 3.0

Christchurch 1 C.S1.12 2626.8 2403.5 223.3 8.5 3.0

Christchurch 1 C.S1.9 2855.5 2578.5 277.0 9.7 3.0

Christchurch 2 C.S2.7 2189.0 2007.3 181.7 8.3 3.3

Christchurch 2 C.S2.3 2043.0 1873.4 169.6 8.3 3.3

Christchurch 2 C.S2.10 2178.0 2005.9 172.1 7.9 3.3

Dunedin 1 D.S1.8 2839.3 2694.5 144.8 5.1 4.7

Dunedin 1 D.S1.11 2896.0 2678.8 217.2 7.5 4.7

Dunedin 1 D.S1.12 2545.3 2356.9 188.4 7.4 4.7

Dunedin 2 D.S2.2 3294.7 3047.6 247.1 7.5 4.4

Dunedin 2 D.S2.3 3090.1 2855.3 234.8 7.6 4.4

Dunedin 2 D.S2.4 3381.5 3124.5 257.0 7.6 4.4

207
 Table B.2 Aggregate grading measurements

Sieve size (mm)

Core ID 37.5 26.5 19.0 16.0 13.2 12.5 9.50 6.70 4.75 2.36 1.18 0.60 0.30 0.15 0.08

Proportion of aggregates passing sieve (%)

A.S1.4 100 100 100 100 84 74 48 32 24 18 14 11 9 7 5

A.S1.3 100 100 100 100 84 74 49 32 26 20 16 13 11 9 7

A.S1.1 100 100 100 100 84 76 52 35 26 19 14 11 9 7 6

A.S2.1 100 100 100 100 89 81 53 33 23 13 10 8 6 5 4

A.S2.4 100 100 100 100 91 84 55 35 25 13 9 7 6 5 4

H.S1.5 100 100 100 100 91 84 57 32 22 19 15 13 11 10 9

H.S1.8 100 100 100 100 90 86 62 36 25 19 15 12 10 9 7

H.S1.1 100 100 100 100 88 82 59 35 24 17 13 10 8 6 5

H.S1.2 100 100 100 100 86 81 54 33 23 14 11 9 8 6 5

H.S1.9 100 100 100 100 94 89 64 35 24 15 11 9 7 6 5

H.S1.6 100 100 100 100 92 88 62 38 28 19 14 11 9 7 6

C.S1.8 100 100 100 100 98 94 89 53 40 29 26 25 23 21 19

C.S1.12 100 100 100 100 100 95 88 49 38 26 19 15 13 9 6

C.S2.7 100 100 98 95 91 87 77 40 24 15 12 10 9 7 5

C.S2.3 100 100 99 94 92 88 80 39 24 15 12 10 9 6 5

208
C.S2.10 100 100 98 97 92 88 78 40 25 15 12 11 9 7 5

D.S1.8 100 100 100 100 100 100 97 84 74 55 38 27 19 15 12

D.S1.11 100 100 100 100 100 100 97 83 73 50 35 25 17 13 11

D.S1.12 100 100 100 100 100 100 96 83 72 47 31 21 15 11 8

D.S2.2 100 100 100 100 100 99 90 69 60 44 35 29 25 21 19

D.S2.3 100 100 100 100 99 98 83 57 47 34 24 19 15 12 10

D.S2.4 100 100 100 100 100 100 94 79 69 52 36 24 17 12 9

209
 Table B.3 Aggregate angularity measurements of chip seal cores

Angularity measurement of aggregate Average

Core ID
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 angularity

A.S1.4 1.20 1.24 1.28 1.19 1.26 1.20 1.28 1.25 1.33 1.27 1.21 1.25 1.22 1.25 1.21 1.24

A.S1.3 1.22 1.28 1.17 1.19 1.31 1.30 1.34 1.29 1.40 1.34 1.29 1.26 1.25 1.22 1.23 1.27

A.S1.1 1.26 1.22 1.20 1.22 1.30 1.29 1.24 1.32 1.24 1.21 1.24 1.28 1.21 1.24 1.31 1.25

A.S2.1 1.24 1.19 1.16 1.19 1.26 1.22 1.23 1.23 1.31 1.25 1.24 1.21 1.20 1.29 1.25 1.23

A.S2.4 1.20 1.23 1.19 1.25 1.24 1.21 1.18 1.38 1.33 1.18 1.35 1.26 1.27 1.30 1.22 1.25

H.S1.5 1.26 1.23 1.22 1.15 1.15 1.20 1.33 1.26 1.27 1.24 1.23 1.32 1.20 1.25 1.20 1.23

H.S1.8 1.18 1.18 1.46 1.18 1.27 1.17 1.26 1.19 1.29 1.22 1.21 1.21 1.36 1.22 1.26 1.24

H.S1.1 1.20 1.22 1.22 1.23 1.25 1.20 1.22 1.21 1.21 1.17 1.18 1.18 1.20 1.19 1.20 1.20

H.S2.2 1.22 1.15 1.22 1.22 1.19 1.22 1.22 1.26 1.21 1.19 1.21 1.26 1.21 1.20 1.22 1.21

H.S2.9 1.21 1.26 1.17 1.23 1.17 1.19 1.24 1.39 1.29 1.30 1.25 1.25 1.25 1.19 1.22 1.24

H.S2.6 1.27 1.17 1.26 1.30 1.20 1.21 1.42 1.20 1.22 1.28 1.25 1.19 1.26 1.23 1.18 1.24

C.S1.8 1.18 1.32 1.21 1.23 1.16 1.26 1.20 1.19 1.21 1.18 1.22 1.28 1.28 1.18 1.26 1.22

C.S1.12 1.14 1.21 1.21 1.22 1.26 1.28 1.19 1.27 1.29 1.27 1.22 1.22 1.22 1.20 1.19 1.23

C.S1.9 1.16 1.17 1.21 1.17 1.21 1.16 1.20 1.30 1.23 1.15 1.19 1.26 1.27 1.22 1.20 1.21

C.S2.7 1.18 1.18 1.19 1.19 1.33 1.26 1.19 1.23 1.24 1.25 1.27 1.20 1.21 1.23 1.23 1.22

C.S2.3 1.22 1.32 1.25 1.16 1.18 1.17 1.23 1.44 1.21 1.21 1.30 1.16 1.18 1.17 1.16 1.22

210
C.S2.10 1.21 1.14 1.15 1.27 1.31 1.28 1.21 1.29 1.27 1.22 1.30 1.09 1.32 1.16 1.17 1.23

D.S1.8 1.19 1.16 1.23 1.18 1.20 1.26 1.22 1.21 1.18 1.29 1.17 1.20 1.17 1.18 1.26 1.21

D.S1.12 1.29 1.20 1.13 1.20 1.35 1.19 1.25 1.28 1.23 1.24 1.20 1.21 1.29 1.20 1.26 1.23

D.S2.2 1.23 1.17 1.19 1.27 1.24 1.36 1.27 1.22 1.22 1.36 1.33 1.19 1.18 1.23 1.27 1.25

D.S2.3 1.23 1.18 1.27 1.23 1.19 1.18 1.23 1.27 1.25 1.20 1.31 1.20 1.25 1.23 1.24 1.23

D.S2.4 1.19 1.24 1.20 1.24 1.22 1.20 1.23 1.21 1.26 1.20 1.20 1.24 1.28 1.22 1.27 1.23

211
APPENDIX C – AIR VOID AREA PLOTS OF SAMPLES

Auckland Site 1

a) Core 2 – 5,000 wp

b) Core 3 – 10,000 wp

Figure C.1 Area of air voids in the tracked samples and the untracked
samples of cores from Auckland site 1
212
Auckland Site 2

Figure C.2 Area of air voids in the tracked sample and the untracked sample
of Core 2 from Auckland site 2

213
Waikato Site 1

a) Core 1 – 5,000 wp

b) Core 2 – 10,000 wp

214
 c) Core 3 – 20,000 wp

Figure C.3 Area of air voids in the tracked samples and the untracked
samples of cores from Waikato site 1

215
Christchurch Site 1

Figure C.4 Area of air voids in the tracked sample and the untracked sample
of Core 1 from Christchurch site 1

216
Christchurch Site 2

a) Core 1 – 2,500 wp

b) Core 2 – 5,000 wp

217
 c) Core 3 – 10,000 wp

Figure C.5 Area of air voids in the tracked samples and the untracked
samples of cores from Christchurch site 2

218
Dunedin Site 1

a) Core 1 – 2,500 wp

b) Core 2 – 5,000 wp

219
 c) Core 3 – 10,000 wp

Figure C.6 Area of air voids in the tracked samples and the untracked
samples of cores from Dunedin site 1

220
Dunedin Site 2

a) Core 1 – 2,500 wp

b) Core 2 – 5,000 wp

221
 c) Core 3 – 10,000 wp

Figure C.7 Area of air voids in the tracked samples and the untracked
samples of cores from Dunedin site 2

222
APPENDIX D – AIR VOID VOLUME PLOTS OF
SAMPLES

Auckland Site 2

Figure D.1 Volume of air voids along the depth of the tracked sample and the
untracked sample of Core 2 from Auckland site 2

223
Christchurch Site 1

Figure D.2 Volume of air voids along the depth of the tracked sample and the
untracked sample of Core 1 from Christchurch site 1

224
Dunedin Site 2

a) Core 1 – 2,500 wp

b) Core 2 – 5,000 wp

225
 c) Core 3 – 10,000 wp

Figure D.3 Volume of air voids along the depth of the tracked samples and the
untracked samples of cores from Dunedin site 2

226
 APPENDIX E – 3D RECONSTRUCTIONS OF CHIP SEAL
SAMPLES

Auckland Site 1

19 mm 19 mm

52 mm 52 mm

a) 3D reconstruction of untracked sample b) 3D reconstruction of tracked sample

Figure E.1 3D reconstruction of the untracked sample and the tracked sample
of Core 2 from Auckland site 1

227
19 mm 19 mm

52 mm 52 mm

a) 3D reconstruction of untracked sample b) 3D reconstruction of tracked sample

Figure E.2 3D reconstruction of the untracked sample and the tracked sample
of Core 3 from Auckland site 1

228
Auckland Site 2

19 mm 19 mm

52 mm 52 mm

a) 3D reconstruction of untracked sample b) 3D reconstruction of tracked sample

Figure E.3 3D reconstruction of the untracked sample and the tracked sample
of Core 2 from Auckland site 2

229
 Waikato Site 1

19 mm 19 mm

52 mm 52 mm

a) 3D reconstruction of untracked sample b) 3D reconstruction of tracked sample

Figure E.4 3D reconstruction of the untracked sample and the tracked sample
of Core 2 from Waikato site 1

19 mm
19 mm

52 mm 52 mm

a) 3D reconstruction of untracked sample b) 3D reconstruction of tracked sample

Figure E.5 3D reconstruction of the untracked sample and the tracked sample
of Core 3 from Waikato site 1

230
 Dunedin Site 1

19 mm 19 mm

52 mm 52 mm

a) 3D reconstruction of untracked sample b) 3D reconstruction of tracked sample

Figure E.6 3D reconstruction of the untracked sample and the tracked sample
of Core 1 from Dunedin site 1

19 mm 19 mm

52 mm 52 mm

a) 3D reconstruction of untracked sample b) 3D reconstruction of tracked sample

Figure E.7 3D reconstruction of the untracked sample and the tracked sample
of Core 2 from Dunedin site 1

231
19 mm 19 mm

52 mm 52 mm

a) 3D reconstruction of untracked sample b) 3D reconstruction of tracked sample

Figure E.8 3D reconstruction of the untracked sample and the tracked sample
of Core 3 from Dunedin site 1

232
 Dunedin Site 2

19 mm 19 mm

52 mm 52 mm

a) 3D reconstruction of untracked sample b) 3D reconstruction of tracked sample

Figure E.9 3D reconstruction of the untracked sample and the tracked sample
of Core 1 from Dunedin site 2

19 mm 19 mm

52 mm 52 mm

a) 3D reconstruction of untracked sample b) 3D reconstruction of tracked sample

Figure E.10 3D reconstruction of the untracked sample and the tracked

sample of Core 2 from Dunedin site 2

233
19 mm 19 mm

52 mm 52 mm

a) 3D reconstruction of untracked sample b) 3D reconstruction of tracked sample

Figure E.11 3D reconstruction of the untracked sample and the tracked sample
of Core 3 from Dunedin site 2

234
 APPENDIX F – 3D RECONSTRUCTIONS OF AIR VOIDS
DISTRIBUTIONS

Auckland Site 2

19 mm 19 mm

52 mm 52 mm

a) Untracked sample b) Tracked sample

Figure F.1 3D reconstruction of air voids distribution in the untracked sample

and the tracked sample of Core 2 from Auckland site 2

235
 Dunedin Site 2

19 mm 19 mm

52 mm 52 mm

a) Untracked sample b) Tracked sample

Figure F.2 3D reconstruction of air voids distribution in the untracked sample

and the tracked sample of Core 1 from Dunedin site 2

19 mm 19 mm

52 mm 52 mm

a) Untracked sample b) Tracked sample

Figure F.3 3D reconstruction of air voids distribution in the untracked sample

and the tracked sample of Core 2 from Dunedin site 2

19 mm 19 mm

52 52 mm
Dunedin
a) Untracked sample b) Tracked sample

Figure F.4 3D reconstruction of air voids distribution in the untracked sample

and the tracked sample of Core 3 from Dunedin site 2
236
APPENDIX G – DATA ANALYSIS REFERENCES

Correlation Strength Assessment Guideline

Classification used for the assessment of the strength of correlation coefficient and R2
values (Cohen, 1992; Cohen, 1988).

Correlation coefficient, r:
Weak correlation: r = 0.10 to 0.29 or r = -0.10 to -0.29
Moderately strong correlation: r = 0.30 to 0.49 or r = -0.30 to -0.49
Strong correlation: r = 0.50 to 1.00 or r = -0.50 to -1.00

R2 value:
Weak correlation: 0.01‎≤‎R2 < 0.09
Moderately strong correlation: 0.09‎≤‎R2 < 0.25
Strong correlation: 0.25‎≤‎R2

237
Method of Logistic Model Development
Detailed below are the steps that are used for determining the coefficients for the
independent variables in a logistic regression model. The logistic model takes the
following form:

( ) ( )

Where, p(x) is the probability of a specific event occurring;

x is the independent variable;
a is the coefficient of the regression constant; and
b is the coefficient of the independant variables.

In the presented research, logistic regression was conducted using a forward stepwise
method, where the independent variables SurfDepth, surfage, RutLANE and
no_of_heavies were used in the regression. The forward stepwise method started with
a model that included only a constant and no independent variables (Step 0 in Table
G.1) and at each step, the independent variables that had the largest score statistic
(from‎ column‎ “Score”)‎ with‎ a‎ significance‎ value‎ below‎ 0.05‎ at‎ each‎ step‎ (from‎
column‎ “Sig.”),‎ namely‎ SurfDepth and surfage, was added to the model. The
variables that had a significance value above 0.05, namely RutLANE and
no_of_heavies, were not added to the model. The Variable Significance table (Table
G.2) was used to determine whether an independent variable was useful to the logistic
model, where all the independent variables that were included in the model (boxed
rows in Table G.2)‎had‎a‎significant‎change‎(<0.05‎in‎column‎“Sig.‎of‎the‎Change”)‎in‎
the -2‎ log‎ likelihood‎ value‎ (from‎ column‎ “Change‎ in‎ -2‎ Log‎ Likelihood”). The
Parameter Estimates table summarised the effect of each independent variable to the
logistic model (shown Table G.3).

238
 Table G.1 Variable Selection table

Score df Sig.
Step 0 Variables SurfDepth 67.801 1 .000
surfage 142.319 1 .000
RutLANE 7.455 1 .006
no_of_heavies 3.054 1 .081
Overall Statistics 214.819 4 .000
Step 1 Variables SurfDepth 89.148 1 .000
RutLANE 7.989 1 .005
no_of_heavies 4.218 1 .040
Overall Statistics 89.482 3 .000
Step 2 Variables RutLANE .201 1 .654
no_of_heavies .883 1 .347
Overall Statistics .915 2 .633

Table G.2 Variable Significance table

Model Log Change in -2 Sig. of the

Variable df
Likelihood Log Likelihood Change
Step 1 surfage -551.907 156.168 1 .000
Step 2 SurfDepth -473.822 98.650 1 .000
surfage -514.745 180.495 1 .000

Table G.3 Parameter Estimates table for logistic regression model

Variable B S.E. Wald df p value Exp(B)

SurfDepth 0.046 0.005 77.134 1 < 0.001 1.047

surfage 0.293 0.025 132.594 1 < 0.001 1.340

Constant -2.913 0.265 120.994 1 < 0.001 0.054

The columns in Table G.3 denoted the following:

 B – The‎ values‎ in‎ column‎ “B”‎ were‎ the‎ values‎ for‎ the‎ logistic‎ regression‎
equation for predicting the dependent variable from the independent variables.
239
 From Table G.3, the coefficient of the independent variable SurfDepth was
0.046, coefficient of the independent variable surfage was 0.293 and the
constant was -2.913. Using the values in Table G.3, the logistic model for the
dependant variable p(x) was as follows:

( ) ( )

 S.E. – The‎values‎in‎ column‎“S.E.”‎were‎the‎standard‎errors‎associated‎ with‎

the coefficients. The standard error was used for testing whether the parameter
was significantly different from 0.
 Wald – The‎ values‎ in‎ column‎ “Wald”‎ denoted‎ the‎ Wald‎ statistics‎ of‎ the‎
independent variables. The Wald statistic was obtained by determining the
square‎ of‎ the‎ ratio‎ of‎ the‎ coefficient‎ (column‎ “B”)‎ to‎ its‎ standard‎ error‎
(column‎“S.E.”).‎‎
 df – This column listed the degrees of freedom for each of the tests of the
coefficients.
 p value – These values indicated the significance level of the Wald statistic. A
small p value (less than 0.05) denoted a factor that was statistically significant
to the logistic model.
 Exp(B) – These values were the exponentiation of the coefficients. These
values represented the ratio-change in the odds of the event of interest for a
one-unit change in the predictor.

In addition to the Parameter Estimates table, another output of the logistic regression
was the Classifications table, which showed the practical results of using the logistic
regression model (Table G.4). The‎ “Observed”‎ column in Table G.4 indicated the
number‎of‎0's‎and‎1's‎that‎were‎observed‎in‎the‎dependent‎variable.‎The‎“Predicted”‎
column in Table G.4 indicated the predicted values of the dependent variable based on
the full logistic regression model. For each case, the predicted response was 1 if that
case's model-predicted probability was greater than the cut-off value specified in the

240
dialogs‎(in‎this‎case,‎the‎default‎of‎0.5).‎The‎“Overall‎Percentage”‎column‎indicated‎
the overall percent of cases that were correctly predicted by the model.

Table G.4 Classifications table for logistic regression model

Predicted

Selected Cases Unselected Cases

Observed
FlushingInitiated Percentage FlushingInitiated Percentage
0 1 Correct 0 1 Correct

0 194 126 60.6 190 106 64.2

FlushingInitiated
1 109 398 78.5 93 423 82.0

Overall Percentage 71.6 75.5

When‎observing‎the‎predicted‎values‎under‎“Selected‎Cases”‎in‎Table G.4 which were

the data values that were used to create the model, 398 of the total 507 data values
where FlushingInitiated = 1 were classified correctly. 194 of the total 320 data values
where FlushingInitiated = 0 were classified correctly. The Overall Percentage of
correctly classified data values of 71.6% was calculated by determining the
percentage of the total number of data values that were classified correctly, as shown
below:

( )
( )

Validation of the logistic model is conducted using the data values under the
“Unselected‎Cases”‎in‎Table G.4. In the model validation dataset, 423 of the total 516
data values where FlushingInitiated = 1 were classified correctly, while 190 of the
total 296 data values where FlushingInitiated = 0 were classified correctly. Overall,
75.5% of the model validation dataset was classified correctly, as shown below:

( )
( )

241
Method of Linear Regression
Detailed below are the methods that were used to determine the coefficients of the
independent variables in a linear regression model. A linear regression model was in
the following form:

Where, y was the dependant variable;

x was the independent variable, i = 0.....n;
b was the coefficient of the independent variable, i = 0....n; and
a was the constant.

In the reported research, linear regression was conducted using a stepwise method. In
a stepwise linear regression the variables that were selected for regression were added
to the model one-by-one and the variables that did not have a significance to the
model were removed. The coefficients of the independent variables in the model at
each step in the regression were shown in a Coefficients table (Table G.5). Presented
below is an example of the linear regression method that was conducted to develop
the flushing progression model for second and higher generation seals. The
independent variables that were selected for use in the presented linear regression
were surfage, SurfDepth, RutLANE, Chipsize, pc_heavy and no_of_heavies. The
significance of each independent variable to the model was determined using the
values‎in‎ column‎labelled‎“p‎value”,‎ where‎a‎p-value less than 0.05 was considered
significant. In the presented example, the independent variables that were significant
to the linear regression model were surfage, SurfDepth, RutLANE and Chipsize
(boxed rows in the Table G.5). The coefficients of the independent variables were
determined‎ from‎ the‎ column‎ labelled‎ “B”.‎ The‎ coefficient‎ of‎ surfage was 0.416,
SurfDepth was 0.040, RutLANE was 0.110 and Chipsize was 0.170. Based on these
coefficients, the linear regression model for flushing progression was:

[ ]

242
The Excluded Variables table (Table G.6) showed the variables that were not included
in the model in each step, with pc_heavy and no_of_heavies (boxed in Table G.6) not
being included in the linear regression model. The predictive strength of the linear
model at each step of the regression process was shown in Table G.7. The final linear
model had an R2 value of 0.634 (highlighted in Table G.7).

Table G.5 Coefficients table from linear regression

Model Unstandardized Standardized t Sig. 95.0% Collinearity

Coefficients Coefficients Confidence Statistics
Interval for B
B Std. Beta Lower Upper Tolerance VIF
Error Bound Bound
(Constant) 3.05 .151 20.153 .000 2.752 3.347
surfage .424 .018 .757 24.221 .000 .390 .458 1.000 1.000
(Constant) 1.81 .211 8.576 .000 1.396 2.226
surfage .404 .017 .720 24.320 .000 .371 .436 .976 1.025
SurfDepth .033 .004 .234 7.908 .000 .025 .041 .976 1.025
(Constant) 1.10 .285 3.865 .000 .541 1.662
surfage .401 .016 .716 24.503 .000 .369 .433 .974 1.026
SurfDepth .034 .004 .242 8.249 .000 .026 .042 .971 1.030
RutLANE .105 .029 .105 3.646 .000 .049 .162 .995 1.005
(Constant) .108 .474 .227 .821 -.824 1.039
surfage .416 .017 .742 24.193 .000 .382 .450 .873 1.145
SurfDepth .040 .005 .288 8.453 .000 .031 .049 .709 1.411
RutLANE .110 .029 .111 3.838 .000 .054 .167 .990 1.010
Chipsize .170 .065 .094 2.615 .009 .042 .299 .638 1.567
a. Dependent Variable: sqrt_flushing

243
 Table G.6 Excluded variables table from linear regression

Model Beta t Sig. Partial Collinearity Statistics

In Correlation Tolerance VIF Minimum
Tolerance
b
SurfDepth .234 7.908 .000 .354 .976 1.025 .976
b
RutLANE .089 2.878 .004 .137 .999 1.001 .999
b
Chipsize -.066 -1.980 .048 -.094 .875 1.143 .875
b
no_of_heavies .019 .602 .547 .029 .999 1.001 .999
b
pc_heavy -.061 -1.905 .057 -.091 .961 1.041 .961
c
RutLANE .105 3.646 .000 .172 .995 1.005 .971
c
Chipsize .085 2.325 .021 .111 .641 1.560 .641
c
no_of_heavies .014 .465 .642 .022 .999 1.001 .975
c
pc_heavy -.040 -1.334 .183 -.064 .953 1.049 .933
d
Chipsize .094 2.615 .009 .125 .638 1.567 .638
d
no_of_heavies -.011 -.364 .716 -.017 .948 1.055 .943
d
pc_heavy -.053 -1.781 .076 -.085 .941 1.063 .933
e
no_of_heavies .022 .687 .493 .033 .809 1.237 .544
e
pc_heavy -.033 -1.082 .280 -.052 .863 1.159 .585
a. Dependent Variable: sqrt_flushing
b. Predictors in the Model: (Constant), new_surfage
c. Predictors in the Model: (Constant), new_surfage, SurfDepth
d. Predictors in the Model: (Constant), new_surfage, SurfDepth, RutLANE
e. Predictors in the Model: (Constant), new_surfage, SurfDepth, RutLANE, Chipsize

Table G.7 Model summary table of linear regression

R
Adjusted R Std. Error of
Model validate= 1.00 validate~= 1.00 R Square
Square the Estimate
(Selected) (Unselected)
1 .757a .573 .572 1.49156
2 .792 b
.627 .625 1.39647
3 .799 c
.638 .635 1.37719

4 .802d .588 .643 .640 1.36804

a. Predictors: (Constant), surfage
b. Predictors: (Constant), surfage, SurfDepth
c. Predictors: (Constant), surfage, SurfDepth, RutLANE
d. Predictors: (Constant), surfage, SurfDepth, RutLANE, Chipsize

244