Rowsell, Philip J. (1991) The Automatic Optimisation of Drilling Performance. PHD Thesis, University of

Download as pdf or txt
Download as pdf or txt
You are on page 1of 204

Rowsell, Philip J.

(1991) The automatic optimisation of


drilling performance. PhD thesis, University of
Nottingham.

Access from the University of Nottingham repository:


http://eprints.nottingham.ac.uk/11909/1/292168.pdf

Copyright and reuse:

The Nottingham ePrints service makes this work by researchers of the University of
Nottingham available open access under the following conditions.

This article is made available under the University of Nottingham End User licence and may
be reused according to the conditions of the licence. For more details see:
http://eprints.nottingham.ac.uk/end_user_agreement.pdf

For more information, please contact [email protected]


University of Nottingham
Department of Mining Engineering

The Automatic Optimisation of Drilling


Performance

by

Philip J. Rowsell
B.Eng

Thesis Submitted to the University of Nottingham


for the Degree of Doctor of Philosophy

January 1991
Affirmation

The work submitted in this thesis is my own and has not been previously submitted for
any other degree. The following publications have been based on this research:-

P.J.Rowsell and M.D. Waller


'Automatic Optimisation of Rotary Drilling Parameters'
Drillex 90

P.J.Rowsell and M.D. Waller


'An Intelligent Drill'
4th Canadian Symposium of Mining Automation, 16-18th September 90.
Saskatoon, Canada.

P.J.Rowsell and M.D. Waller


'Intelligent Control of Drilling Systems'
SPE/IADC 1991 Drilling Conference,11-13th March. RAI Congrescentrum,
Amsterdam.

1
Acknowledgements

The author would like to express his sincere thanks to all those people who have been
of assistance during the course of this study.

The author would particularly like to thank:-

From the University of Nottingham,

Mr Martin Waller my supervisor, for his consistent guidance, assistance, generation of


ideas and financial assistance. In addition for being a good friend, tolerating the
continual pestering with which he was subjected to.

Dr Bryan Denby for his help with various computer problems which have confronted
the author, as well as other useful advice.

Gratitude is also extended to the staff of the Mining Department Laboratories for their
help during the research project.

From Industry,

John Beswick, Kenting Drilling Services Ltd, for his time, help and advice given over
the course of this research project.

Stuart Wilson, DeBeers Industrial Diamond Division, Charters, Ascot for his help .

..
11
Abstract

The drilling industry, along with many others, is becoming increasingly competitive,
demanding greater efforts to improve safety and reduce costs. For this reason,
companies are progressively looking towards computerised automation to enhance
performance. Unlike most industries however, the drilling industry has been slow to
take advantage of the advances in computer and automation technology. Only recently
have automatic operations such as tubular handling been placed under computer
control. These activities relate to peripheral mechanical handling problems which are
relatively easy to solve. The concept of an automatic intelligent drill, capable of making
its own or assisted decisions about drilling parameters such as weight on bit or
rotational speed, may seem remote and far into the future.

Research in drilling automation, at the University of Nottingham, has the ultimate


objective of achieving computerized drill control through the the application of an
intelligent knowledge induction system. At the University, a laboratory rig has been
developed with such a system installed. Decisions for optimal performance are based
on either maximum penetration or minimum cost drilling. The system has a self-
learning capability, allowing a progressive improvement in performance. The
prototype system is currently undergoing trials, using real data collected while the
laboratory rig is drilling and artificial data. The results are very encouraging and
demonstrate the feasibility and advantages of optimised drill performance.

This thesis describes the design and development of this drill optimisation scheme
produced by the author. Both the theory behind the optimisation system, and the results
of the initial phase of Laboratory testing are included.

111
Contents

Affirmation .................................................................................. i
Acknowledgements ..................................................................... ii
Abstract ...................................................................................... iii
List of Figures .......................................................................... vii
List of Tables ........................................................................... xiv
Thesis Outline ............................................................................ xv

Chapter 1 - Introduction

"/ 1.1 Drill Optimisation ........................................................................ 1


Jl.2 The Research Project ..................................................................... 5

JChapter 2 - Modifications to the Laboratory Drill Rig and


Wear Measurement Jig

2.1 Introouction ............................................................................... 9


2.2 The Laboratory Drill Rig .............................................................. 10
2.2.1 Hardware modifications ........................................................ 10
2.2.1.1 Monitoring Hardware ..................................................... 14
2.2.1.2 Control Hardware .......................................................... 14
2.2.1.2.1 Stop / Start Control. .................................................. 16
2.2.1.2.2 Feedback Loops ...................................................... 16
2.2.2 Software Modifications ......................................................... 18
2.3 Wear Measurement Jig ................................................................ 20
2.3.1 Improvements to the Wear Jig .................................................. 21
2.3.2 MOOifications to the Software ................................................. 24
2.4 BBC - mM Data Link ................... , ............................................. 24
2.5 Conclusion .............................................................................. 27

J Chapter 3 - Preliminary Drilling Tests and Developments

3.1 Introouction ............................................................................. 28


3.2 Drill Response Tests ................................................................... 28
3.3 The Development and Testing of a Simple Drill Optimisation Scheme .......... 32

IV
3.3.1 Problems Arising from the Initial Tests ........................................ 32
3.4 The Drill Simulator ..................................................................... 36
3.5 Conclusion .............................................................................. 41

'/Chapter 4 - Optimisation Theory

4.1 Introduction ............................................................................. 42


4.2 The Governing Parameter and Its Relationship ..................................... 43
4.2.1 Maximised Penetration Rates ................................................... 43
4.2.2 Time ................................................................................ 43
4.2.2.1 Minimum Time for Each Bit Run ......................................... 43
4.2.2.2 Total Time to Completion .................................................. 45
4.3 Cost ...................................................................................... 47
4.4 Cost Equation Sensitivity Analysis ................................................... 52
4.4.1 The Effect of Bit and Rig Cost with Respect to
Tripping Time ............................................................................ 53
4.4.1.1 Bit Costs Greater than Rig Costs ......................................... 53
4.4.1.2 Bit Cost Equal to Rig Costs ............................................... 53
4.4.1.3 Bit Cost Less than Rig Costs .............................................. 56
4.4.2 The Effect of Bit Costs with Constant Rig Costs ....~ ........................ 56
4.4.2.1 Low Rig Costs .............................................................. 56
4.4.2.2 High Rig Costs ............................................................. 59
4.4.3 Conclusions of the Sensitivity Analysis ....................................... 59
4.5 Conclusion .............................................................................. 59

Chapter 5 - Bit Wear and Wear Prediction

5.1 Introduction ............................................................................. 62


5.2 The Wear Predictor ..................................................................... 63
5.2.1 Data Validity ...................................................................... 65
5.2.2 Data Enhancement by Interpolation ............................................ 66
5.2.2.1 Two Dimensional Interpolation ........................................... 67
5.2.2.2 Three Dimensional Ripple ................................................. 69
5.2.2.3 Multi-Dimensional Interpolation .......................................... 71
5.3 Testing of the Predictor. ............................................................... 73
5.4 Results ................................................................................... 73

v
5.4.1 Two Dimensional Method ....................................................... 73
5.4.2 Three Dimensional Method ...................................................... 84
5.5 Conclusion .............................................................................. 84

Chapter 6 - The Control Algorithm

6.1 Introduction ............................................................................. 93


6.2 Establishing the Minimum Cost Operating Point ................................... 94
6.2.1 Maxima and Minima Theory .................................................... 94
6.2.2 Partial Differentiation
6.2.3 Computer Search Methods ..................................................... 103
6.2.3.1 The Search Algorithms ................................................... .104
6.2.3.1.1 Vector Methoo ....................................................... 106
6.2.3.1.2 Uni-Directional Increments ......................................... 106
6.2.3.1.3 Multi-Directional Method ........................................... 108
6.3 The Complete Cost Optimisation System .......................................... 108
6.4 Change in Rock Strata ................................................................ 110
6.5 Operational Constraints ............................................................... 114
6.6 Conclusion ............................................................................. 114

Chapter 7 - The Results of the Optimisation System

7.1 Introduction ............................................................................ 117


7.2 The Cost Optimisation Computer Programme ..................................... 117
7.3. The Testing of the Optimisation System .......................................... 118
7.3.1 Test Work Using the I.B.M Alone ........................................... 119
7.3.1.1 Maximisation of Penetration Rates ...................................... 119
7.3.1.1.1 Establishment of the Search Routine .............................. 121
7.3.1.1.2 Conclusion of the Search Method Tests .......................... 128
7.3.1.1.3 Data Variance ......................................................... 134
7.3.1.2 Minimum Cost Optimisation Using the mM Alone ................... 137
7.3.1.2.1 Simulator Wear Rate and Simulator Penetration Rate ........... 141
7.3.1.2.2 Simulator Wear Rate and Penetration Rate Variance ............ 141
7.3.1.2.3 Randomly Generated Wear Values and Simulator
Penetration Rate .................................................................. 145

VI
7.3.1.2.4 Randomly Generated Wear Values and Penetration Rates
Variance ........................................................................... 146
7.3.1.2.5 Conclusion of Minimum Cost Tests ............................... 151
7.3.1.3 Conclusion of IBM Alone ................................................ 153
7.3.2 The IBM and BBC Drill Simulator ............................................ 153
7.3.2.1 Maximisation of Penetration Rate ........................................ 153
7.3.2.1.1 Non-fluctuating Penetration Rates ................................. 153
7.3.2.1.2 Penetration Rate Variance ........................................... 155
7.3.2.2 Minimum Cost ............................................................. 155
7.3.2.2.1 Simulator Wear and Non Fluctuating Penetration Rates ........ 155
7.3.2.2.2 Simulator Wear and Fluctuating Penetration Rates ............. 155
7.3.2.2.3 Randomly Generated Wear Values and Non Fluctuating
Penetration Rates ................................................................. 155
7.3.2.2.4 Randomly Generated Wear Values and Fluctuating
Penetration Rates ................................................................. 160
7.3.3 The IDM and the Laboratory Drill ............................................. 166
7.4 Conclusions ............................................................................ 166

Chapter 8 - Conclusions
J Con c Ius ion s ................................................................................. 169

Chapter 9 - Recommendations for Future Work


9.1 Improvements to the Optimisation System ......................................... 175
9.1.1 Rock Boundary Indication and Strata Prediction ............................ 175
9.1.2 Rolling Depth .................................................................... 176
9.1.3 Multi-Hump Surface Prediction and Improvements to the Ripple
Methoo ................................................................................... 176
9.1.4 Wear Rate Variance ............................................................. 177
9.1.5 Examination of the Optimisation Programme ................................ 177
9.2 General Improvements and Recommendations ................................... 178
9.2.1 Machine Test Trials ...................................................... ······ .178
9.2.2 Wear Data ................................................... ····.··.············ .178
9.2.3 Field Test Trials .......................................... ······· ... ·.·········· .179

JRefer en ces ................................................................................. 180

Vll
List of Figures

Figure Page
No. No.
1.1 A Graph Showing the Trade Off Between Penetration Rates ................. 6
and Wear Rates.

1.2 A General Scheme for a Drill Optimisation System ............................. S

2.1 General View of the Drill Electronics System ................................. 11

2.2 The 1 MHz B us Pin Connections and the Electronics Rack ................. 12

2.3 Address Decoding Circuit for &FCEO - &FCES .............................. 13

2.4 Voltage Shifting Circuit From (-1 V to + 1V) to (OV to + 2V) ................ 15

2.5 The Drill Stop / Start Circuit Diagram .......................................... 17

2.6 A Darlington Pair Used to Drive the On / Off Solenoids ..................... 17

2.7 A Driving Circuit for a Three Phase Stepper Motor .......................... 19

2.S End Effects Caused by the Old Type of L.V.D.T. Pointer .................. 22

2.9 Improvements to the Measured Profile by the New L.V.D.T.


Pointer System .................................................................... 22

2.10 The Old Method of Locating the Drill Bit in the Wear
Measurement Jig .................................................................. 23

2.11 The New Method of Locating the Drill Bit in the Wear
Measurement Jig .................................................................. 23

2.12 Profiles Before Superimposition ................................................ 25

2.13 The Superimposed Profiles ...................................................... 25

2.14 A Typical Volume / Area Loss Calculation Result ............................ 25

3.1 A Graph to Show the Influence of Weight On Bit on


Penetration Rate ................................................................... 29

3.2 A Graph Showing the Influence of Rotational Speed on


Penetration Rate ................................................................... 30

3.3 A Graph Showing the Influence of Flow Rate on Penetration


Rate ................................................................................. 31

3.4 A Simple Optimisation Algorithm for Maximising


Penetration Rates .......................................... ························ 33

Vl11
Figure Page
No. No.
3.5 A Plot of the Laboratory Drill's Response to the
Maximisation of Penetration Rate Algorithm .................................. 34

3.6 A Typical Plot of the Penetration Response Used by the


Drill Simulator ..................................................................... 38

3.7 A Multi-Hump Surface ........................................................... 39

3.8 or
The R~~e Standard Deviation Values Measured During
an OptImlSanon Test .............................................................. 40

4.1 Fixed Costs in a Drilling Programme ........................................... 49

4.2 Variable Costs in a Drilling Programme ........................................ 49

4.3 The Influence of Tripping Time when Bit Costs (£1500) are
Greater than Rig Costs (£100) .................................................. 54

4.4 The Influence of Tripping Time when Bit Costs (£1000) are
Equal to Rig Costs (£1000) ...................................................... 55

4.5 The Influence of Tripping Time when Bit Costs (£1000) are
Less than Rig Costs (£ 10000) ................................................... 57

4.6 The Influence of Bit Costs Compared to Constant Rig Costs


(£100) and Constant Tripping Time (20 s/m) ................................. 58

4.7 The Influence of Bit costs Compared to Constant Rig Costs


(£10000) and Constant Tripping Time (20 s/m) .............................. 60

5.1 An illustration to Show the Effect of Giving the Mean


Values to Unknown Points ...................................................... 68

5.2 A Two by Two Matrix which has Undergone Rippling


After a Single Value has Been Entered ......................................... 70

5.3 A 3-Dimensional Plot of the Equation Used for the First


Set of 2-Dimensional Ripple Tests .............................................. 74

5.4 A Typical Result from the First 2-Dimensional Test


Showing the Resulting S.L.P.M. Values after 20 Random
Values have Been Entered ...................................................... · 76

5.5 A Typical Result from the First 2-Dimensional Test


Showing the Resulting S.L.P.M. Values after 60 Random
Values have Been Entered ....................................................... 77

5.6 A Typical Result from the First 2-Dimensional Test


Showing the Resulting S.L.P.M. Values after 150 Random
Values have Been Entered ....................................................... 78

IX
Figure Page
No.

5.7 A 3-Dimensional Plot of the Equation Used for the Second


Set of 2-Dimensional Ripple Tests .............. :............................... 79

5.8 A Typical Result from the Second 2-Dimensional Test


Showing the Resulting S.L.P.M. Values after 20 Random
Values have Been Entered ....................................................... 81

5.9 A Typical Result from the Second 2-Dimensional Test


Showing the Resulting S.L.P.M. Values after 60 Random
Values have Been Entered ....................................................... 82

5.10 A Typical Result from the Second 2-Dimensional Test


Showing the Resulting S.L.P.M. Values after 150 Random
Values have Been Entered ....................................................... 83

5.11 A 3-Dimensional Plot of the Equation Used for the Third


Set of 2-Dimensional Ripple Tests .............................................. 85

5.12 A Typical Result from the Third 2-Dimensional Test


Showing the Resulting S.L.P.M. Values after 20 Random
Values have Been Entered ....................................................... 87

5.13 A Typical Result from the Third 2-Dimensional Test


Showing the Resulting S.L.P.M. Values after 60 Random
Values have Been Entered ....................................................... 88

5.14 A Typical Result from the Third 2-Dimensional Test


Showing the Resulting S.L.P.M. Values after 150 Random
Values have Been Entered ....................................................... 89

6.1 Wear Rate and Penetration Rate Data, the Corresponding


Graph and Polynomial Fit for Gniess .......................................... 97

6.2 Wear Rate and Penetration Rate Data, the Corresponding


Graph and Polynomial Fit for Sandstone ...................................... 98

6.3 Wear Rate and Penetration Rate Data, the Corresponding


Graph and Polynomial Fit for Limestone ...................................... 99

6.4 Wear Rate and Penetration Rate Data, the Corresponding


Graph and Polynomial Fit for Fine-Grained Sandstone ..................... l00

6.5 The Minimum Cost Results Obtained for Maxima and


Minima Cost Optimisation ...................................................... 102

6.6 A Simplified Control Scheme with the Prediction


Mechanism Included ............................................................. 105

6.7 An illustration of the Vector Selection Method ............................... 107

6.8 A Diagram to Show How the Vector Method May be Used


to Aid Detection of Multi-Hump or Multi-Trough Surfaces ................ 107

x
Figure Page
No.
6.9 A Schematic Representation of the Cost Optimisation
Control System ................................................................... 109

6.10 An Example of Strata Boundary Indication Through the


Monitoring of the Drill Parameters ............................................. 111

6.11 The Relationship Between Penetration Rate and the


Dimensionless Index U.C.S / e (Ultimate Compressive
Strength / Specific Energy) ..................................................... 113

6.12 The Complete Cost Optimisation Scheme with Lithology .................. 115
Predictor

7. 1 A 3-Dimensional Plot of the Simulated Penetration Rates


Used in the IBM Alone Tests ................................................... 120

7.2 An Early Stage of the Vector Method Locating Maximum Penetration


Rate Process ...................................................................... 122

7.3 An Intermediate Plot of the Vector Method Test Showing the


Progressive Learning by the S.L.P.M.of the Penetration Rate
Process ............................................................................ 123

7.4 The Final Path Taken by the Vector Method to Locate


Maximum Penetration Rate ..................................................... 124

7.5 An illustration of the Improvement in Efficiency of the~


Search by Increasing the Size of the Search Cross .......................... 125

7.6 A Plot of the Final Path Taken by the Uni-Directional


Method to Locate Maximum Penetration Rate ................................ 126

7.7 The Change in the Efficiency of the Uni-Directional Method


Caused by a Different Starting Direction ..................................... 127

7.8 The Final Path Taken by the Combined Vector and Uni-
Directional Method to Locate Maximum Penetration Rate .................. 129

7.9 The Path Taken to Locate Maximum Penetration Rate by the


Uni-Directional Method from a Different Starting Position ................ 130

7.10 The Benefit of the Combined Method Over the Uni-


Directional Method Starting from the Same Location ....................... 131

7.11 An Intermediate Plot of the Multi-Directional Method


Showing the Progressive Learning of the Penetration Rate
Process by the S.L.P.M ........................................................ 132

7.12 The Final Path Taken by the Multi-Directional Method


to Locate Maximum Penetration Rate ......................................... .133

7.13 The Improvement in Efficiency to the Multi-Directional


Method by Increasing the Size of the Search ................................ 135

Xl
Figure Page
No.
7.14 A Plot of the Optimisation System Attaining Maximum
Penetration Rates with a ~20% Variation in Penetration
Rate Values ....................................................................... 136

7.15 The Optimisation System Attaining Maximum Penetration


Rates with a :-60% Variation in Penetration Rate Values .................. 138

7.16 The Simulated Wear Rate Process ............................................. 139

7.17 The Cost Surface Generated by the Cost Equation, U sing the
Two Simulators and the Defined Cost Variables ............................. 140

7.18 A Plot of the Optimisation System Locating the Minimum


Cost Position with No Data Fluctuations ..................................... 142

7.19 The Cost Surface Generated During Testing with Simulator


Wear Values and Penetration Rate Variance .................................. 143

7.20 The Associated Penetration Rate S.L.P.M. Surface


Generated During Testing with Simulator Wear Values and
Penetration Rate Variance ....................................................... 144

7.21 An Intermediate Plot of the Cost Surface During Minimum


Cost Optimisation Using Randomly Generated Wear Values
and Simulator Penetration Rates ............................................... 146

7.22 An Intermediate Plot of the Penetration Rate S.L.P.M.


During Minimum Cost Optimisation Using Randomly
Generated Wear Values and Simulator Penetration Rates ................... 147

7.23 A Later Plot of the Cost Surface with the Minimum Cost
Position Obtained ................................................................ 149

7.24 The Cost Surface Generated by the Test Using Randomly


Generated Wear Values and Penetration Rate Variance ..................... 150

7.25 The Effect of Passing On of Knowledge From Test to Test. ............... 152

7.26 The Response of the Drill Simulator to the Control Provided


by the Optimisation System Set for Maximisation of
Penetration Rates ................................................................ 154

7.27 The Drill Simulator's Response to the Control Provided by


the Optimisation System Set for Maximisation of Penetration
Rates but with Penetration Rate Variance ..................................... 156

7.28 The Drill Simulator's Response to Optimisation by Minimum


Cost, No Data Fluctuations Present ................................ ·········· .157

7.29 The Drill Simulator's Response to Optimisation by Minimum


Cost, with Penetration Rate Variance .......................................... 158

XlI
Figure Page
No.
7.30 The Response of the Drill Simulator to Optimisation by
Minimum Cost, with Randomly Generated Wear Values, the
Random Value Set at 5 .......................................................... 159

7.31 A Plot of the Cost Surface Generated by Optimisation


Through Minimum Cost with Randomly Generated Wear
Values, the Random Value Set at 5 ............................................ 161

7.32 The Drill Simulator's Response to Optimisation by Minimum


Cost, with Randomly Generated Wear Values, the Random
Value Set at 20 ................................................................... 162

7.33 The Response of Drill Simulator to Optimisation by


Minimum Cost, with Randomly Generated Wear Values
and Penetration Rate Variance .................................................. 163

7.34 The Cost Surface Generated by Optimisation Through


Minimum Cost Using the Drill Simulator, with Randomly
Generated Wear Values and Penetration Rate Variance ..................... 164

7.35 The Wear Rate S.L.P.M. Generated by Minimum Cost


Optimisation Using the Drill Simulator, with Randomly
Generated Wear Values and Penetration Rate Variance ..................... 165

Xlll
List of Tables
Table Page
No. No.

1.1 Development of Rotary Drilling .......... '......................................... 2

1.2 Parameters which Hinder and Effect Drilling Optimisation ................... .4

5.1 2-Dimensional Ripple Test 1 .................................................... 75

5.2 2-Dimensional Ripple Test 2 .................................................... 80

5.3 2-Dimensional Ripple Test 3 .................................................... 86

5.4 3-Dimensional Ripple Test 1 .................................................... 90

5.5 3-Dimensional Ripple Test 2 .................................................... 91

6.1 The Cost per Metre for Varying Penetration and Wear
Rates, for the Rig Parameters and Rock Types Shown ..................... 101

XIV
Thesis Outline

The design and development of the optimisation system described in this thesis is fairly
complex and thus difficult to describe in a logical step by step progression. Every effort
has been made to do so. However, many concepts are interrelated and as a
consequence, they may require detailed explanation, before a real appreciation is
gained. Some of these concepts are complicated in their own right and may require
detailed explanation before being interrelated. Consequently, this can lead to some
concepts being initially difficult to understand or their relevance seen, which later
become clear once the whole system has been developed.

Therefore a brief description of the layout of the thesis has been included, highlighting
the contents of each chapter, to give the reader a general overview and aid initial
understanding.

Chapter 1 provides an introduction to drill optimisation and outlines the proposal of this
research project, to develop an 'on-line' drill optimisation system. To be able to
develop and test the optimisation system in a real situation, a laboratory drill rig,
already in existence was utilized. However, some modifications were required, mainly
for computer control and these are covered in Chapter 2. To test these modifications, as
well as generate both an understanding of the drill, and initial ideas on the optimisation
scheme, a series of tests were performed and are described in Chapter 3. These tests
highlighted a number of problems with the control of the drill rig and its inefficiency for
initial software testing. This proved invaluable in subsequent work.

The idea's gained from Chapter 3, allowed the initial development of the optimisation
scheme could be conducted. In any optimisation system, decisions have to be made on
what is the aim of the optimisation scheme, and by what method / parameters this may
be achieved. This is covered in Chapter 4, and describes several methods which were
investigated, the optimisation scheme finally being based on minimum cost per metre
drilling.

The problems of attaining 'on-line' measurements for the parameters required for the
optimisation scheme was covered in Chapter 5. A prediction method was developed,
(mainly for wear rate prediction), which would enhance the known data to aid
prediction of unknown points giving the system some self-learning capability.

xv
In Chapter 6, the developments of Chapter's 4 and 5, were utilized to generate methods
for a minimum cost optimisation scheme. Several methods were investigated, with a
continuous computer search method being selected. The compete layout and operation
of this optimisation system is described, along with the problems of multi strata
operations.

Chapter 7 describes the results of the testing programme to validate the optimisation
system, with conclusions and recommendations for further work being covered in
Chapters 8 and 9 respectively.

XVI
Chapter 1 - Introduction

1.1 Drill Optimisation

Holes are drilled in geological strata for a wide range of applications. For example,
exploration of an ore body or oil reserve, for extraction purposes, blast hole generation,
de-watering, and for many other reasons. Virtually every mineral extraction.project
will require some form of drilling. As a consequence, drilling expertise is very
important for successful operations. I;?rilling knowledge has progressed through time
from a fairly simple operation to a highly complex one, involving a variety of different
drilling techniques and methods, each for a particular purposes or situation. This
development of technology is aptly summarised by J.L. Lummus (40) which shows the
progression of rotary drilling - see Table 1.1.

With increasing knowledge and more advance technology, current drilling limits are
being pushed further each year (65), with deeper wells in more adverse geology.
· Combined with increased competition and the harsh economic climate, the demand for
more sophisticated equipment and computerization is growing. This is not only to
improve performance, but also to reduce cost.

The increase in computer technology, has allowed great advances in the monitoring and
control of the drill rigs and drill parameters. This has resulted in more data being
available to the drilling engineer for use in performance prediction and optimisation.
However, the data is only worthwhile if it truly represents what is actually happening
down hole.

In the mining industry, where the majority of holes are shallow and of small diameter,
it is believed that monitored data does represent down hole conditions. This is reflected
by the degree of monitoring / control equipment on the latest types of drill rig. This not
only benefits the drilling engineer, but is useful for many other processes such as
reserve calculation and blast design (32,33,39,46,66).

In the oil industry, the picture is slightly different as the complexities of monitoring in
deep holes are much greater. Some experts dispute the levels of accuracy of
measurements attained and do not believe that the monitored data represents down hole
conditions. However with progressive research and the increase in the use of
measurement while drilling (MWD) tools (13,24), it is hoped that this discrepancy will
be soon eliminated.

1
Rotary Drilling Development

Conception Period Development Period


1900-1920 1920-1948

Rotary Drilling Principle More Powerful Rigs


1900 (Spindle Top)

Rotary Bits Better Bits


1908 (Hughes)

Casing and Cementing Specialized Muds


1904-1910 (Halliburton)

Drilling Mud Improved Cementing


1914-1916 (National Lead Company)

Scientific Period Automation Period


1948-1968 1968-

Full Automation of Rig


Expanded Drilling Research and Mud Handling

Better understanding of "Closed-loop" Computer


Hydraulic principles Operation of Rig

Significant Bit Improvements Control of Drilling Variables

Optimised drilling; Complete planning of Well


Improved Mud Technology Drilling from Spud to Production

Table 1.1 Development of Rotary Drilling (After Lummus)

2
In any optimisation process, a decision has to be made on what parameter is to be
optimised. In an operation such as drilling, there are a multitude of parameters which, if
optimised could help improve the overall efficiency of the operation, e.g. the percentage
core recovery, penetration rate, cost, time to completion, ml!d circuits, etc. The choice
of the parameter comes down to what the hole is actually required for. For exploration
holes, in certain sections, the most critical constraint may be the percentage core
recovery. In many cases, the desired optimums will conflict with each other e.g.
percent~ge core recovery and maximum penetration rate. However for the majority, or
the major part of most holes, the predominant criteria will be to produce the hole as
quickly and as cheaply as possible.

In choosing the optimisation parameter, precaution should be exercised as there are


many parameters or factors which hinder the drilling operation. Some of these based
on a table by McDanial and Lummus (44) , are shown in Table 1.2.

If an optimisation system is employed, it can be seen from Table 1.2 that a number of
the parameters are unalterable and restrict the optimisation scheme. Others (also
unalterable) are unpredictable and interfere with and reduce the efficiency of the
optimisation scheme (1,19,20,39,40,42,49,69). Only a small number are under the
'- ,"-

direct control of the drilling engineer. The extent to which an optimisation scheme can
be applied is therefore limited by the unalterable variables. The system also requires a
degree of flexibility to cope with the unpredictable ones.

How then can drill optimisation be performed? Many of the currently employed
systems in the drilling industry use off-line techniques, where historical data from
previous wells are correlated with others, in an attempt to predict the nature of a similar
hole to be drilled in the same area. These vary from optimum casing design, methods of
evaluating rig perfonnance, to the use of drilling simulators. Many publications and
methods, featuring both manual and latterly computerized methods, have been
developed throughout drilling history which try to improve the performance of the
drilling operation. Each has had varying degrees of success (4,8,22,36,41,42,
48,55,59,63,66,67).

Other methods rely on drill off tests to predict the likely response of the drilling
operation in certain rock strata (5,15,26,27,61). The results are fed back through
various equations, to detennine optimum operating conditions. The most famous is
"How to achieve minimum cost drilling?" by Galle and Woods (26,27), which

3
Variables which restrict Variables which Compromise
the optimisation scheme Implementation

i
General Rig General Rig

Program Pump power Location Deviation

Total depth Rotary power Logistics Lost circulation

Geological predictions Pump pressure Weather Abnonnal pressures

Hole sizes Pump liners Planning Sloughing shale

Evaluation Flexibility Supervision Hole trouble

Casing program Pit system Mud solids

Directional Drill string Crew efficiency

Tripping time Geological correlation

Table 1.2 Parameters which Hinder and Effect Drilling Optimisation

4
estimates the correct bit weight and rotary speed to attain minimum cost drilling, taking
into account parameters such as bit wear.

With the advances in computer technology, the potential of on-line optimisation


techniques has increased dramatically. To the authors knowledge, the first paper
published on such a system was in 1968 (74), and to date remains the only known one
to be published. The paper describes the attempts by the Humble Oil and Refining
company, to develop an on-line drilling optimisation system using a full scale drill rig.
The system was developed around a Honeywell DPP 116 digital computer that logged
all the main drilling parameters and controlled weight on bit and rotational speed. The
system optimised through a cost equation, similar to the one used in this optimisation
scheme. Penetration rate characteristics for different combinations of bit and rock types
were determined by drill off tests and the wear characteristics of the bit were derived
from wear curves form the manufacturer. The trials proved successful, but the current
status of this project is unknown to the author.

More recently other "optimisation" systems have been developed (37,52,64)


particularly by Tamrock, but these do not directly optimise drilling performance. They
concentrate on automation of rod handling etc. None however, have been on same the
scale of the Humble Oil experiment. Therefore scope exists for the development of such
a system and it is understood that within the drilling industry an increasing amount of
attention is directed towards optimising drill performance. However much of this
research is propriety and currently remains unpublished.

1.2 The Research Project

Drilling research has been conducted at the University of Nottingham for 6 years,
predominantly using Diamond Impregnated Core Bits, but latterly with Poly-crystalline
Diamond Compact Bits. For the context of this project, the Diamond core drill rig
would be used, being more accessible and with more information available.

Initial research on diamond impregnated bits was conducted by Ambrose (2), who
analysed bit performance. In this research a number of different rock samples were
drilled with a variety of different bits and the results monitored. Amongst the results
produced we~ graphs of penetr~tion rate. against bit wear rtte. A. typical plot produced
is shown in FIgure 1.1. From thIS graph It can be seen that lrere IS a trade off between
penetration rates and bit wear rates. However how do wecfcide where the optimum

5
Graph of Penetration Rate and Wear Rate against
Weight on Bit

400 0.18

0.16

-s
.-=
-
S 300
--
S
---
S
QJ
S
S
--
0.14
~
~

.-=
-
QJ
cu

--
~
0

r..
QJ
~
r..
cu
QJ
~
0.12

~
=
QJ
200
.... WearRate
(mm/m)
0.10 -e- Penetration Rate
(mm/min)

100 0.08 -+------r--'-_,._--r------r---....-.----,

o 100 200 300

Weight on Bit (kg)

Figure 1.1 A graph showing the trade off between Penetration Rates and
Wear Rates

6
operation point is? What factors effect this operating point, and is it actually
achievable?

This was the basic rationale for this research project, i.e. to develop a system to locate
and operate a drilling system at the optimum performance point. The idea of such an
optimisation system is shown by Figure 1.2. Parameters from the drilling operation are
fed to a central location known as an Intelligent Knowledge Induction System. This has
the power to process these results and decide which parameters to alter to bring about
an improvement in performance. In this hypothetical case, rotational speed and weight
on bit (thrust) are indicated. This process would continue until no change of the
parameters was seen, hence at this stage the drilling operation would be operating at its
optimum performance point.

The project was split into two distinct parts :-

i) To establish feedback control loops on the existing radial ann core drill, and
to use them to aid the development of a system to maximise penetration rates.

ii) Find and develop a method to operate the drill at its optimum performance
point.

7
ROCK CLASSIFICATION

.....
,..-----...." . .
IKI
SYSTEM

I. K. I. S. _ Intelligent Knowledge Induction System

Figure 1.2 A General Scheme for a Drill Optimisation System

8
Chapter 2 - Modifications to the Laboratory Drill Rig and
Wear Measurement Jig

2.1 Introduction

Drilling research at the University started with the conversion of an old radial arm
machine shop drill to a laboratory rock coring drill. The drill used diamond impregnated
bits to produce rock cores for testing purposes as well as being used for drill research
purposes. During the fIrst two years, the laboratory rig was instrumented and computer
software generated to allow a variety of measurements to be taken, while the machine
was operating. These include parameters such as rotational speed, weight on bit
(thrust), penetration rate, flow rate, motor voltage and motor current. From these, a
variety of other parameters may also be derived. Consequently, this has allowed the
study of the drill's response and performance while drilling a variety of rocks under
different conditions.

In addition, a wear measurement jig was also built, enabling the measurement of wear
profiles developed on the diamond impregnated bits. This allows a wear history of drill
bits to be determined.

The author was involved in much of this work, mostly on the software side, but this
work and a greater explanation of the Laboratory drill rig and wear measurement jig
has been covered in previous publications (2,14,56).

It should be mentioned that at the start of this project the drill monitoring and control
system was thought adequate for the development of the optimisation system. However
as the project developed, the processing power of this system proved insufficient and
the use of an IBM type machine was required for the optimisation algorithm. This
posed a number of problems which were surmounted, but complicated the structure of
the optimisation scheme.

With the benefit of hindsight, and the technological advances in electronics and
computer systems that have occurred during this project, a different monitoring system
would have been initially developed, based solely upon a IBM type machine. However
despite this. a working drill optimisation system was developed, indicating the
feasibility and potential for such a system It is expected that for full scale test trials
an upgrade to a Micro Vax or Sun Work Station will be required.

9
2.2 The Laboratory Drill Rig

2.2.1 Hardware Modifications

From the onset of this research project, it was apparent that the laboratory drill rig
would need some major modifications to achieve the ultimate aim of optimising drill
perfonnance. Documentation on the drill monitoring electronics already in existence
was sparse. In addition, there had been no provision made for future expansion, which
would be required for the inclusion of feedback loops to several of the drills
parameters. Therefore it was decided to rebuild the electronics, salvaging what was
necessary.

The drill electronics were based around a BBC Micro Computer. The BBC was an ideal
computer with which to develop a low cost monitoring system, having many additional
features compared with other computers. The Basic language was enhanced from those
nonnally available, allowing a degree of structured programming. It was also readily
suitable to electronic interfacing, having ports readily available to do so i.e. the User
Port and the 1 MHz Bus. These and many other features make the BBC an extremely
versatile machine readily suitable to interfacing projects.

Figure 2.1, shows the electrical system in its entirety and serves to give the non-
electrical reader an idea of how the complete system works. From Figure 2.1, it can be
seen that the main communication line between the computer and the external drill
electronics is the 1 Megahertz bus. The 1MHz Bus is extremely useful for interfacing as
it has up to 502 memory addresses specially allocated to it in the BBC. Consequently it
is possible to service a large number of peripheral devices, and hence its use as the back
bone of drill electronics system. To allow for ready expansion of the drill electronics, a
common pin configuration was linked across the entire back plane of the electronics
rack, Figure 2.2. This configuration consisted of the I MHz Bus lines and the common
voltages used on the boards. In so doing each board could have direct access to the 1
MHz bus if required. Those not, such as the signal conditioning board could be located
next to their controlling boards for easy direct linkage.

With the addition of hardware to I Mhz Bus, address decoding is essential to ensure
that only the required chip / peripheral device is accessed by a pre-set range of address
values. A typical circuit is shown in Figure 2.3.

10
.....-"--.....I..----'--....L......., ...._ _ _ _ _ _ _ _ _-1 Wear Jig
EI ectron ics
Wear Measurement
Jig

L.V.D.T
rfBill
Selector

User Port

Stepper
Signal
Motor
Conditioning Control
Control

R.P.M
Air Pressure
Flush Rate
Motor Voltage
Motor Current

Penetration Rate

Figure 2.1 General View of the Drill Electronics System

11
Eurocard Back Plane Rail Etiquette

PIN Connection PIN Connection PIN Connection

1 +5 V 12 D2 23 AS
2 +12V 13 D3 24 A6
3 -12V 14 D4 25 A7
4 R /W 15 D5 26
5 1MHz 16 D6 27
6 NIQR 17 D7 28
7 NFC 18 AO 29
8 NFD 19 Al 30
9 NRS 20 A2 31 +24V
10 IX) 21 A3 32 OV
11 Dl 22 A4

Electronic
Electronics
Prototype
Rack
Board
(Eurocard)

1 MHz B us hung
across the Back Eurocard
Plane Connectors

Figure 2.2 :- The 1 MHz Bus pin connections and the electronics rack.

12
-FC
OV

A3

A4

AS

A6 CS

A7
R/W
+5V

Figure 2.3 Address Decoding Circuit for &FCEO - &FCE8

13
2.2.1.1 Monitoring Hardware

The majority of the various transducer signals were fed to a central electronics board
which conditioned and converted the signals to a range of 0 - 10 Volts required by the
Analog to Digital (AID) convertor. As no changes were made to the monitoring
transducers, the original signal conditioning board was directly incorporated into the
new system. The conditioned signals were passed to a 8 bit 8 channel AID converter
located on a separate adjacent board, directly accessing the 1 MHz Bus.

Penetration Rate is measured using a Linear Variable Differential Transformer


Transducer (L.V.D.T.). L.V.D.T.'s utilise the coupling generated by a soft iron
plunger in a series of coils to relate the position of the plunger. A similar device was
also used for the wear measurement jig. As both L.V.D.T.'s were never used at the
same time, it was decided to bring them under a common circuit. A four pole relay was
used to select the appropriate L.V.D.T. output, and this was fed to an Oscillator
Demodulator, which translates the L.V.D.T.'s signals to a linear response ranging from
-1 to +1 volt. The internal 12 bit AID of the BBC was used for converting this signal,
but as this required a 0 - 2 volt input, the Oscillator Demodulator output was passed
through a voltage shifting circuit shown in Figure 2.4.

The measurement of flow rate, had previously been based on an orifice plate and a
differential pressure transducer. This had resulted in a multitude of problems, generally
due to large pressure transients and the transducers incompatibility with water. This
caused the failure of a number of such transducers. A new method has been developed
(on the PDC rig) using a low cost water turbine, which while installed on the radial arm
drill rig, the electronics have not yet been commissioned.

2.2.1.2 Control Hardware

The control of external electronics from computers, generally has to be made through
interface adapters either Versatile Interface Adaptors (VIA) or Peripheral Interface
Adapters (PIA). Both are very similar, the VIA containing such features as internal
clocks, useful for more complex interfacing. These devices allow both the reading and
writing of data to a series of data lines ( collectively known as a port) which can either
be used for monitoring or control purposes. The BBC has an internal VIA for use by
the programmer, accessing both the printer and the User port. The User port in this
case however, has been dedicated to the wear jig and therefore and additional VIA or
PIA was necessary. As no external timing was required in the control circuits, a PIA,

14
1 8
2 -.J 7 + 12 V
L.V.D.T in ( -1 V to + 1V) 3
+>-
.- 6
-12V 4 5

1 14
2 13 +12 V
3 12
-12 V 4 11
5 10 L.V.D.T. out
(0 - 2 V)
OV - ..
10K 6 9 +12V 00

7 8
-
0
~
~
tv

10 K Preset

10 K Preset

-12 V won ov
(1 Watt) 3.9 Zenor
Diode

I:;gure 2.4 Voltage Shifting Circuit From (-IV to +IV) to (OV to +2V)

15
was selected and hung on the 1 MHz bus. This gave access to an additional 16 data
lines for use in control or monitoring purposes.

2.2.1.2.1 Stop / Start Control

To enable a fair degree of automation, drill stop / start control from the micro processor
was incorporated. This also added a safety feature as the machine could be instructed
to switch off if torque levels were too high or if the drill had reached the bottom of the
hole.

Figure 2.5 shows a circuit diagram for the drill stop / start circuit. By pressing the start
button, the resulting brief contact closure energises the primary coil, closing the drill
contacts and the primary contacts and hence primary circuit. The primary and machine
contacts remain closed until such time as the stop button is pressed, breaking the
circuit, de-energising the coil and allowing the contacts to open.

By adding two relays as shown in Figure 2.5, direct control from the computer can be
established, utilising two data lines from the PIA chip, one for start control the other for
stop control. As the PIA lines are only at TTL logic levels (i.e. 0-5 volts) and of limited
current capacity, they are incapable of directly driving a relay. Figure 2.6 shows the
driving circuit for the two relays triggered by positive logic levels on the control lines.
This is known as a Darlington pair.

2.2.1.2.2 Feedback Loops

From previous drilling tests it was known that both rotational speed and weight on bit
had a major influence on drill performance, and thus both these parameters were to be
placed under micro processor control. It was also known that flow rate also had an
effect, although not significant on this laboratory rig. However due to the difficulties
experienced in monitoring the flow rate, and its negligible influence, it was decided not
to control flow rate at this stage. However, the systems electronics were designed to
allow easy inclusion at later date.

Manual speed control utilised a potentiometer, to vary the demand signal to the motor
speed control electronics. As a low budget solution was necessary, it was decided to
motorize this unit, such that when the motor was energised the computer could control
the rotation of the potentiometer, and when de-energised manual control would remain.

16
240 V I
Main Motor

~
A/C Cwrent
,./
I
I

I
I
I Stop by breaking
I circuit
I

Start by briefly
~~ ~
closing circuit

I
-
I- Key

-~ I
§- Relay Solenoid

~
Figure 2.5 The Drill Stop / Start Circuit Diagram

12 Volts

Solenoid
nI

lK

oVolts

Figure 2.6 A Darlington Pair Used to Drive the Stop / Start Solenoids

17
To control the rotation of the potentiometer with any accuracy would require stepper
motor control. These motors turn a specific angle of rotation for each pulse sent to their
control circuit. The motor used in this case was an old three phase stepper motor which
had been salvaged from another piece of equipment. However, most modern stepper
motors are now four phase, and thus no off the shelf control circuit chips could be
found. A circuit designed by M.D.Waller for a the conversion of a three phase to six
phase, was examined and simplified, and proved satisfactory - Figure 2.7. An on / off
control for the motor was also added to reduce the time for which the motor was
energised, and thus minimise power supply drain.

A prototype system for the motorised potentiometer was built out of Metal Mechano to
prove the principle, using several gear ratio's to attain the desired resolution. This
system was so successful, it was used for the actual drill speed controller.

With the success of the speed controller, a similar approach was sought to control the
weight on bit. Weight on bit was provided by a piston in which air pressure was varied
to give differing loads. Unfortunately the existing pressure regulator was old and too
complex to motorize in the same way. However an additional regulator was found,
which could easily be adapted. This was placed in series with the other. By opening the
original pressure regulator fully, pressure regulation could be passed to the motorised
regulator and hence under computer control.

2.2.2 Software Modifications

The drill rig software had previously been developed to a fairly high degree (56).
Consequently only several modifications were made.

Throughout the history of the rig, the accuracy of penetration rate measurements have
caused problems due to the electrical noise associated with the internal AID converter on
the BBC. Many attempts to smooth the resulting fluctuations had improved the reading.
However, as these were written in Basic, it was thought that an interrupt routine
(through assembler) could be used to improve the current system.

Interrupts take priority over the general running of the computer e.g. the execution of a
programme, and thus give a means of attaining a higher priority over the programme.
Many devices work using interrupts e.g. disc system, keyboards etc, and all have
designated priorities. Care therefore must be taken at which level the interrupt is set at
as the results can otherwise be disastrous.

18
\
~
_\ ) I
) ,//

2
U 14

13_
5V On/Off 1

2
U 16 5V

15

3 12 OV 3 14

r:
\ Z 6
'"
"""
CD
en
11

10

9
X OV 4

OV 5

OV 6
-.....I
.r:..
......
<.0
.r:..
13

12

11 Clock
(
\.
~ L-
(

OV 7 8 Y 7 10
I--
OV 8 9 I M L- Direction

>
It)
>
It)

r
(
.........
l\-
C") C")
f ~
~ ~
C") C")

___ 1

!\ T ( L~ L - 2
V 14

13
5V

-
L----- 3 12 I 3K3 5V
I
4 -.....I 11
.r:..
o
01
5 10

6 9
X,Y,Z = Logic signals for the three motor phases 8
OV 7

Figure 2.7 A Driving Circuit for a Three Phase Stepper Motor

19
For this application, the lowest priority is used, as only priority over programme
execution is required, in BBC terms, this is the IRQ2V vector. The interrupt routine
written utilises the internal VIA timer, which causes an interrupt after a certain time has
passed. In this case the time set was a twentieth of a second. On each interrupt, the
timer would be reset and the analog port (penetration measurement) would be read and
stored accumulatively. On every twentieth reading i.e one second intervals, the analog
port values would be averaged and stored to disc along with the other drill parameters.

Previous problems had also occurred with the spin up time of the disc system causing
the programme to hang, while the disc attained the correct speed and the disc buffer
emptied. This was also catered for in the interrupt routine by flushing the disc buffer
each time data was written to the disc, i.e. every second. Consequently, the disc was
constantly spinning and immediately accessed.
The interrupt routine proved highly successful and smoothed the penetration rate
measurements to an acceptable level. However, the only real way to improve the
system would be to build an external 12 bit Nd convertor. For the marginal benefits
gained, it was not thought worthwhile at this stage.

As the project developed, the optimisation routine ran on a P. C for reasons discussed
later. In so doing, the storage of drilling data on the BBC was abandoned. The data
was transferred directly to the P.C for storage on a hard disc. Consequently the disc
system access pans were removed from the BBC monitoring programme.

For general drilling and specific energy research, data is still stored on the BBC disc,
and these programmes i.e. Monitoring, Plotting routines etc, have been brought to a
menu driven option for ease of operation.

2.3 Wear Measurement Jig

A prototype wear jig had been already built, with the software being written by the
author. The current wear jig electronics were housed in a number of boxes making
transportation difficult. With the re-building of the drill electronics, it was thought
worthwhile combining the wear jig electronics into a single board and adding it to the
drill electrical rack.

This modification was conducted at the same time as an additional wear jig was being
manufactured for DeBeers Industrial Diamond Division. Consequently, because of the
commercial nature of this project, two printed circuit boards (one for the University

20
and the other for DeBeers) were produced, rather than using prototype boards. While
the DeBeers project proved highly successful, later modifications to the University
board through progressive developments, resulted in the PCB becoming untidy. In
hindsight would have been better to produce the University board on eurocard allowing
greater flexibility.

2.3.1 Improvements to the Wear Jig

With regular use, it was found that the L.Y.D.T. pointers tended to flatten with use, the
flattening being most pronounced when the pointers were new. This was mainly due to
the pointers being made of mild steel with no heat treatment being applied. In addition
to this, the shape of the pointers had an adverse effect on the measured profile. The
difficulty of machining such small items, resulted in the pointers tending towards a
conical shape. This had the effect of exaggerating the rounding at the crown extremities
as shown in Figure 2.8.

Therefore, a solution to eliminate both problems was necessary, i.e. a pointer needle
like in shape, good wear/ impact resistance and readily available. The solution was
found in household picture nails. These were readily available in any hardware shop,
made of hardened steel and needle like in shape. A suitable holder for these nails was
made to allow for regular replacement. The improvement on the measured profiles is
shown in Figure 2.9.

With the development of the volume loss calculator described later, it was found that
the method of locating the drill bit in the jig was too inaccurate. Originally the drill bit
was butted up to a boss which was set at the beginning of an experiment. If the drill bit
was too small, a blank would be screwed into the bit, effectively elongating it - Figure
2.10. An additional drawback, was that once the apparatus had been set up for one bit
type, other bits could not be measured without moving the position of the boss, which
would invalidate any further measurements of the initial bit.

A new system (Figure 2.11) was designed in which the location method was directly
onto the core bit. A small counter sunk hole was drilled into the shank of the drill. care
being taken not to drill right through. The hole was used to locate a pin which ran
through the mounting "Y" block. This method also insured that the same portion of the
segment was measured each time. Subsequent testingby the re-measuring of the same

21
L.V.D.T Pointer
Crown
Height

Measured Profile

Crown Width

Exaggerated End
Effects

Figure 2.8 End Effects Caused by the Old Type of L.V.D.T. Pointer

Holder for picture nail,


screws into L.V.D.T.

Crown
Height

Hardened Steel
Picture Nail

Crown Width

Figure 2.9 Improvements to the Measured Profile by the New L.V.D.T.


Pointer System

22
Clamping Mechanism

L.V.D.T Locating Boss

I
1o:o:.:.:.a.-4--------------------_-_~__ - - l---4-_....J

Blanking Piece
Drill Bit
'V' Block

Figure 2.10 The Old Method of Locating the Drill Bit in the Wear
MeasurennentJig

Clamping Mechanism

L.V.D.T

- - - - - - - - +-----'

t;:;:;:;:;I--+
_ _ _ _ _ _I f_' ,____ J

Small counter sunk hole


to locale pIn
Drill Bit Locating Pin

Figure 2.11 The New Method of Locating the Drill Bit in the Wear
MeasurennentJig

23
segment several times and comparing the profiles, proved that a greater degree of
accuracy in location and hence wear loss was achieved by this method.

2.3.2 Modifications to the Software

The commercial nature of the DeBeers wear jig ensured the software was developed to
a high degree of efficiency. All programmes were menu driven and as user-friendly as
possible. The only major development undertaken during this research project was the
volume loss calculator. This programme would calculate the amount of wear which had
occurred between two particular profiles. During profile measurement, the L.V.D.T
was traversed at a set distance for all measurements. Therefore, the total volume lost
could be derived by summing the total profile heights, and subtracting one from the
other. However as the L.V.D.T. may not necessarily start in the exact same position
each time, discrepancies in the calculated volumes were observed as shown in Figure
2.12.

To compensate for this, a routine was written in which the two profiles could be
superimposed on each other until the best fit was observed - Figure 2.13 . To remove
end effects and rogue values, inner and outer boundaries could be positioned, and only
the centre portion sumated. The profile with the new boundaries was re-drawn after
calculation. A typical result is shown in Figure 2.14.

At present only the area lost is calculated. For volume to be established, each reading
would be required to multiplied by the appropriate radii. However, for a series of
measurements, errors would arise from not being able to accurately locate an exact radii
position to the crown profile. For this reason, true volume loss was not followed
further, but left as the area multiplied by the average radius. While not being the exact
volume lost, it was felt to be sufficient to show the wear trends.

2.4 BBC - IBM Data Link

During the development of the optimisation system, the processing power required,
soon outstripped the capabilities of the BBC. An upgrade was made to an IBM
compatible, which not only improved memory capacity, but equally important
processing speed. However, this did cause one major problem due to the drill
electronics being geared to the BBC.

24
1 2 3 4 5 6 7 8 9 o Crown Widlb/ rom

Figure 2.12 Profiles Before Superimposition

Crown Width / mm
2 3 4 5 6 7 8 9 o

Figure 2.13 The Superimposed Profiles

Volume Loss of ProfIle 11 - Profile 20

Height/nun

12
t ; ... ' ; .. ' .. , , ' . : ' ' . ,
..... '\ ...... t .. '.•... ,. •• " .. \, •. '.•. ' .. ., .• " .. " .......... " . '\ ...... " ..
~ ~. \:.,
...... ' ...... ~ ..... ~ ...... I....... ~ ~ .. 1_ .. ..: ...... ~ ......
\

~ ......
\

~
: "
~
..... ". . . . . . . .....
\

~ ......
~
~
.
..... :
~
.............
".,
_1_ ... ~ ...... ~ ...... : ...... ,
!~:I~:::~~:,~~::I~::
"'-~"'--"'--"'l"''''~''''''~''''''''''''--''''''-~--''''-~----'-'''~'''-~'''-r--'''''''''''-~---
11 :: : : . ~ ~ ." ~:,
<...>. ~ .. : .. ~ .. ~... ~ .. ~ .. ~ .. l .. ~ . -: ...~ ...
\ \ \ \ \ \

.. ~ •. ~ .. :. . J •• ~ •• : •• ~ .
\ . I \ : . " ~ : . , ' ~ . : I \
' • '\ •• " ............ "\' • , ••••• '\ •• " • '.' • '" •• , .......... '.' • " •• '1 ........... : •••
10
,::I\\~~~~'
/ -.. ~' . . -'-,,'. . . -.. . . Jj~~~
~\'"
. .~..........:.-....-.
\ ." ..... \ ,~'-'....,.,
. . . .~ . . . ,,.'
............. \,. . . .
I"

9 l ... _'_ ... ~ ...... ~ ...... 1.


~
~ .. _'_ ... _:_ ... ~ ...... ~ ...... ~ ... _'_ ......'_ ... ~ ...... ~ ...... : ... _\" ........... ~ .... : .... : .....
~: ~::::\~:~ ':'
... ,~.....:.....~ ...... ~ .. ~ ..;...~ .. : .. : .. : .. ;.....~ .....:..... ~ .... f ......:.. ~ .. ; .. .
8 . ~ .... " .. .. .. ,... .. , ..... ~ ~ \ .. : ..... " .. . ~ \

," ~ .. , , ~ ...... ~ .... -.. ',- --


~--~ ...... -" .. -~- :--~--\ ...... , ........ '--:- ~ -~ .... -.- ~--
\,. \ \ \ " "-
.... .... .... ....
, ,

2 3 4 5 6 7 8 9 o Crown Width / nun

LcngthofCut =28 to 181 = 154 (6.5 nun)


2
Area loss = 0.963 mm

Figure 2.14 A Typical Volume / Area Loss Calculation Result

25
Interfacing from the IBM is much more complex than from the BBC as it does not have
many of the built in features akin to interfacing as seen on the BBC. Consequently,
special boards etc have to be brought in. Even with these boards, large scale
modifications would be required to adapt the electronics to work from a mM. The time
taken to do this and generate a suite of monitoring programmes similar to those on the
BBC, was thought to great to be worthwhile at this demonstration phase of the research
project. Therefore a different approach was sought.

The concept of using the BBC as a front end processor was developed. In this, the
BBC would be designated solely to data acquisition and control of the drill. The main
optimisation system would run on the IBM and send control commands to the BBC, to
access data, or to instruct a parameter change. The two machines would communicate
through the RS232 interface common to both machines.

The establishment of the RS232 data link was originally thought to be a relatively
straight forward and easy task. Unfortunately, it proved to be to the contrary. The
control of the RS232 from the BBC was relatively easy and well documented. The mM
on the other hand was the complete opposite, with infonnation on the hardware side
extremely sparse. With a great deal of experimentation using a breakout box, the correct
pin configuration was established.

However the use of an unorthodox solution was necessary. Problems occurred at the
IBM end, where data could not be read from the RS232 port. The only solution found
to this problem was to short the CTR and RTS lines on each computer, leaving only the
two data lines and ground connected between the two computers. Obviously this is not
an ideal solution but it proved to be the only solution.

The shorting of the respective lines, removed any handshaking capabilities between the
two computers for data transfer control. This made data transfer very complicated as
data could be easily lost. This would typically happen if the IBM sent data to the BBC
and the BBC was busy. The input buffer on the BBC would progressively fill as the
IBM sent more data. If the BBC remained busy and the buffer not accessed, the buffer
would eventually fill. As the crs line had been shorted no instruction could be sent to
the IBM to stop it sending data. Consequently any subsequent data sent by the IBM
would be lost.

To overcome this problem, some sort of software computer handshaking was required,
such that one computer would indicate that it was ready to send data, and would hang

26
until such time the other indicated it was ready to receive the data. This has the
immediate disadvantage that the machines have to wait until the other is ready to receive
to transmit data, compared to normal transfer where the data is buffered and dealt with
when the machine is less busy.

To complicate matters further, it was found also that spurious data appeared on the data
lines. Unless excluded, this would be be treated as nonnal data and thus corrupt values
sent subsequently. To solve this problem, the transferred data was marked with a stop
and start character to ensure the correct data value was deciphered by the other machine.

It had originally been intended that the IBM would control the BBC by an interrupt
routine, such that a character sent by the IBM would cause an interrupt routine to be
called on the BBC. Depending on the character set, this routine would either transfer
data to the IBM or accept control information. This would allow the BBC to spend
most of its time monitoring, and transfer data only when requested. Several assembler
programmes were written to do this, but all proved unsuccessful. All the data transfer
programmes however worked well in Basic, and therefore the requests from the IBM,
had to be serviced by a polling routine on the BBC, which regularly checked the RS232
port to see if a request had been sent.

2.5 Conclusion

In concluding this chapter, the laboratory drill rig and wear measurement jig underwent
a series of modifications to allow the development and testing of the optimisation
system. Many modifications have also improved the general use of the drill, when
utilised for general drilling or coring purposes. An RS232 link to a LB.M was also
established which enables the transferring of data between the two machines. This
allowed the BBC to take the role of a front end processor.

27
Chapter 3:- Preliminary Drilling Tests and Developments

3.1 Introduction

A series of preliminary drilling tests were conducted with the aim of gaining experience
on the drilling apparatus, as well as generating any ideas on optimisation techniques
and highlighting any problems that may occur during optimisation tests.

3.2 Drill Response Tests

From Ambrose's work, it was apparent that both rotational speed and weightQnbit had
an influence on penetration rate. Therefore it was decided to conduct a series of
experiments to validate these results and establish general trends. The tests would
involve the manipulation of rotational speed and weight on bit to see the response of
penetration rate. Flush would also be manipulated. As only general trends were
required only one rock type was used. The results of these tests are shown in Figures
3.1-3.3.

From these graphs it can be seen that weight on bit has the greatest influence on
penetration rate. Rotational speed has some influence, but not as pronounced.
- .---

However, from the tests, it was apparent that a certain speed was required for differing
weight on bit values to ensure the drill did not stall. From the graph showing flush
variation, it can be seen that for the majority of mid range values, its influence (on this
machine) is negligible. However at the extremities it does have quite an adverse affect,
too little causing a build up of cuttings and too great causing hydraulic lift of the bit,
both causing a reduction in cutting efficiency and hence, a reduction in penetration rate.
Therefore all subsequent experiments were conducted with the flush rate being
maintained in the range of 6-12 litres per minute.

t (It should be noted that while in this laboratory system the influence of flush is
negligible, in field operations, especially in deep well drilling, the influence of hole
'hydraulics is critical, not only for hole generation but also well control. Therefore, for
any field application, hole hydraulics must be taken into account within the optimisation
. scheme.

28
Graph of Penetration Rate Vs Weight On Bit

400

300

.-
.-c
s
ee
---
~

~
200

c:::
c
--
.-0
eu
s..
~
100

c
~
Q.

0 100 200 300

Weight On Bit (kg)

Figure 3.1 A Graph to Show the Influence of Weight On Bit on


Penetration Rate

29
Graph of Penetration Rate Vs Rotational Speed

600

500

-..
.-=
-S
S
--
5
.....
< lj

~
400

cz::
=
.-
0
.....
~ 300
.....I.
<lj

=
<lj
Q..

600 800 1000 1200 1400 1600 1800

Rotational Speed (R.P.M.)

Figure 3.2 A graph Showing the Influence of Rotational Speed on


Penetration Rate

30
Graph of Penetration Rate Vs Flow Rate

50

40

.- 30
.-c
s
5
s
--
~

~
~
C
- 20

--
.-
0
~
J...
~
10
C
~
~

O~______~______r -____~______~

o 10 20

Flow Kate llitres/min)

Figure 3.3 A Graph Showing the Influence of Flow Rate on


Penetration Rate

31
3.3 The Development and Testing of a Simple Drill Optimisation
Scheme

With the lessons learnt from these initial experiments, a control strategy for optimising
penetration rates was designed and is shown in Figure 3.4. For a given starting
condition, weight on bit and rotational speed were increased in steps until such time as
the speed started to drop and a stall condition was occurring. The weight on bit was
then reduced to regain the speed of the drill. A stall marker would be set and the
process returned to the initial mode of increasing weight on bit and rotational speed
once again. If this was performed correctly the stall marker would be cleared.
However, if on the next increase of weight on bit, stalling occurred again, the drill was
deemed to be near its optimum point. At this point, the speed was manipulated in either
direction to see if any improvements in penetration rate could be attained. Once
completed, the optimisation process would repeat itself but with reduced step incr-
ements. In this way, the computer would search out the optimum operating point for
maxim urn penetration rate.

The algorithm was programmed into the BBe computer used for drill monitoring, and
a series of test perform in the laboratory to validate the algorithm. A typical plot is
shown in Figure 3.5. It can be seen from the plot that the penetration rate has increased
to near optimum.

3.3.1 Problems Arising from the Initial Tests

From 3.5, it can also be seen that the computer has recovered from two potential stall
conditions. However in so doing, the amount the computer has reduced weight on bit is
too great, causing the penetration rates to drop dramatically.

This is obviously not desirable and requires some modification to the control algorithm.
The problem arises through the weight on bit actuator. Load on the bit is applied
through a compressed air piston controlled by a motorized pressure regulator. The
regulator was treated as having a linear response, i.e. a certain number of steps cause
a certain increase, and the same number in the opposite, the same pressure decrease.
This was clearly not the case.

When a stall condition was encountered, pressure would be reduced as fast as


possible. However, large number of motor steps /turns were required before the
pressure began to drop, after which small numbers of turns would cause large pressure

32
r--
....... , JC
~L-. . . ,.,I. ,.,.n_cr. ,.,.ane_asT"e_increment
Io. ,.,.a,.,.,.,d.,.,.,.b. ,.,.Y__
......
Speed drops 1 .....
.I
Increment stall
marker

Speed OK )
"
t Reduce pressure
to regain speed
Clear stall
marker .. ':

, "
Increase speed Not equal
'" If stall marker
- by set amount
:;..
....
to two equals two
~..._... _...._..._... .,.,.,.
~" .... _. ____.........l ..

"
Hole near optimum
pressure

Reduce
increments

Increment speed to see


if penetration increases

Hole
optimised ..

Figure 3.4 A Simple Optimisation Algorithm for Maximising Penetration


Rates

33
Weitt on R.P.M Pen. Rate
Bi eN) {mm/min}

3000~ 300

iI
1000
( -- V-
/ \ /
VJ
-+::>.
2000 J 200
I
I

I
I
I
1000~ l 100

I I

O.....J oj 0 I

100 Time/(sec)

Figure 3.5 A Plot of the Laboratory Drill's Response to the Maxinisation of Penetration Rate Algorithm
drops. This seemed to indicate that the regulator was following an exponential
response.

To establish the true response of the regulator, a series of tests were performed. By
varying the starting pressure, a range of steps were taken in both directions and the
final pressure in each case noted. The results were quite surprising. The data produced
indicated that the response was in fact fairly linear! However there was a certain
number of steps necessary to change direction, before any pressure change was seen
i.e. a dead band response was being exhibited.

As pressure measurements tended to fluctuate slightly, a tolerance band was


established, inside which no change would be required. If the previous change had
been in the opposite direction from that now required, the required number of steps to
change direction would be performed. With the correct direction set, if changes were
just outside the tolerance limit, the pressure would be slowly adjusted towards the
desired value, otherwise the number of steps to attain the pressure would be calculated
and performed.

The setting of the regulator to various pressures could take several seconds and with
some routines requiring access every second, care had to be taken that on each pass,
the pressure setting routine did not take longer that 30 seconds to run. This required the
splitting of the routine into various parts e.g. changing the pressure direction, such that
the task was accomplished by several calls of the routine. By polling the routine along
with the other major procedures, the routine would be called as regularly as possible.

A similar routine was also written for the rotational speed controller, but as this
exhibited a true linear response, no reversal routine to facilitate direction change was
necessary.

The experiment also established another quirk of the drilling system which baffled the
author for a period of time. During the testing of the control procedures, it was found
that if both control motors were used at the same time, the rotational speed
measurements fluctuated widely. However if either procedure was run individually,
speed measurements remained nonna!. Naturally the first conclusions were that the two
procedures were interfering with each other by a common variable. However extensive
checks and the renaming of variables, did not alleviate the problem. No actual reason to
this problem was found, but it is believed that the current required when both motors
were power up, caused such a drain on the power supply that voltage drops occurred.

35
This caused other circuits including the rotational speed measurement to malfunction.
Therefore, each motor was switched off after use and thus the two were never
energised at the same time, preventing the associated voltage drops.

These changes could have been incorporated into the maximisation of penetration rate
algorithm previously described and further improvements made. However, as this
algorithm was only generated for test purposes, it was not developed further despite
being successful. The main reason for this, was that no self-learning process was
involved. Consequently, no benefit from the information gained from previous holes
could be used to aid the drilling of the next hole. However the optimisation system
described in the subsequent chapters has this capability and could readily be applied to
the maximisation of penetration rates.

3.4 The Drill Simulator

These tests, also highlighted the difficulty of debugging and developing the software
while using the drilling machine. Software errors often required the drilling tests to be
re-run, consequently involving drill re-positioning, collaring, etc, which when
debugging is very time consuming and frustrating. In addition to this problem, regular
access to the drilling machine was needed, the drill often being required for other
purposes, such as laboratory coring as well as its use being bounded by University
hours.

To alleviate this problem and aid software development, a simulator was developed
which would simulate to some degree the action of of the drilling machine. The
simulation procedure would be held in the monitoring computer, and allow a realistic
representation of the drilling rig. Therefore, development work could be conducted
away from the drilling machine in a clean and pleasant atmosphere with increased
efficiency, rather than in the laboratory environment.

Furthermore, as mentioned in Chapter 2, the large processing power required by the


main optimisation system, had resulted in the programme being developed on an
I.B.M. system, with the BBC monitoring computer being solely used as a front end
processor. The two computers were linked by an RS232 data link. While this link had
already been established, the data transfer concepts had to be combined into the
monitoring (BBC) and optimisation (IBM) routines, to allow control to take place. The
concept of this data transfer mechanism is similar to that of parallel programming, each
computer running a part of its own programme, before requiring data to be transferred.

36
Consequently, debugging the software is complex as either machine could be the
initiator and not necessarily the recorder of an error. The testing of the data transfer
software would have been extremely time consuming if not impossible, if a simulator
had not been produced.

The design of the simulator was such that different processes i.e. drill responses, could
be interchanged. This would allow the optimisation system to be tested under a variety
of conditions. To enable a variation in processes, the simulator infonnation was
contained in a matrix fonn, which could be loaded into the computer from a data file. A
typical response is shown in Figure 3.6. This represents a hypothetical penetration rate
against the variables, rotational speed and weight on bit. Therefore for a certain value
of rotational speed and weight on bit, a certain penetration rate would be returned by the
simulator. It can be seen from Figure 3.6 that this only describes a two variate process,
but it could easily be extended to incorporate other parameters such as flush rate by
adding extra dimensions to the matrix. This allows the optimisation algorithm to be
developed to a high degree involving many different parameters. However conceptually
these become impossible to visualise.

As the processes are generated artificially, complex surfaces may be tested to establish
whether the algorithm will cope with a wide range of complex surfaces. This is
particularly useful for testing of multi-hump surfaces such as shown in Figure 3.7.
Methods for this type of prediction are described in later chapters.

With the simulation data being contained in a matrix fonn, real or historical data
generated either from general core drilling or from optimisation test work could be
transferred into a matrix fonn. This could be used to load the simulator and thus allow
the simulator to imitate a real process for the particular rock type being drilled.

This highlighted several other improvements which could be made to the simulator.
When using artificial data, the response of the simulator was always exact, i.e. at a
certain rotational speed and weight on bit, a certain rate of penetration rate was
observed. However, in the real drilling process, the response of the drill is more
varied, giving data fluctuations, as shown in Figure 3.8 showing the range of standard
deviation measured during an optimisation test.

Therefore to enhance the efficiency of the simulator and aid the development of the
optimisation algorithm, a variation in the sim ulated data was produced. This was left as
an optional mode, as with initial development exact known values were beneficial.

37
F i 1ename : PENSIM2

0,0

Figure 3.6 A Typical Plot of the Penetration Rate Response Used by the Drill
Simulator

38
0,0

Figure 3.7 A Multi-Hump Surface

39
F i 1ename · Maxpen3

0,0

Figure 3.8 The Range of Standard Deviation Values Measured During an


Optimisation Test

40
However as the system becomes more developed, a range of standard deviation values
could be incorporated, either imaginary, or real values such as those measured form a
panicular test. In so doing, this would allow more authentic simulation.

Another feature also incorporated into the simulator was that of a time lag between the
setting and attaining of the required parameter value as exhibited by the rig. The
simulator originally gave an immediate response which was initially beneficial to the
development of the optimisation scheme, as the time lag aspect would only serve to
slow down the development. However as the optimisation system becomes more
developed, the time lag between setting and attainment, is useful to simulate the real
drilling situation.

The development of the drill simulator proved highly successful, allowing the easy
testing of the optimisation. The simulation data can take the form of imaginary
processes, initially used for development work, or to test the system under conditions
such as multi peak or trough surfaces. Real processes from those of previously
monitored drilling tests can also be incorporated, to test the optimisation system under
real conditions, with real or imaginary variations in standard deviation. This has
consequently allowed the optimisation algorithm to be developed to a high degree, with
the minimal time spent drilling, in a clean and productive environment.

3.5 Conclusion

In concluding, for the drill rig to be used for drill optimisation research, the complete
electrical system was re-built. As a consequence, a series of tests were performed,
using a simple optimisation algorithm for maximising penetration rates to ensure that
the new electrical system functioned properly, as well as generating further ideas on
drill optimisation. While this algorithm proved successful, it was not progressed any
further, as it was surpassed by the scheme developed later. However it did highlight
several problems both in control and optimisation techniques, which aided development
of the main optimisation scheme.

The testing of the maximisation of penetration rate algorithm also highlighted the
problem and inefficiency of developing software in connection with a real process i.e.
the drilling rig. For this reason, a simulator was established which greatly enhanced the
development of the optimisation scheme. The simulator also allowed the optimisation
process to be studied under a wide range of complex surfaces which would otherwise
be difficult to produce on the drill rig.

~ 1
Chapter 4 - Optimisation Theory

4.1 Introduction

In Chapter 1, the basic rationale for the optimisation system developed during this
research project was introduced. It mentioned that from the work done by Ambrose,
~ optimum point for a given set of conditions existed between penetration rates and
wear rates. This point, in the drilling operation was deemed to be most efficient. To
find this optimum point, a parameter and governing relationship must be developed
which combines the trade off between penetration rate and the associated wear rate.

This trade off is well known by drillers that at the start of the hole, Renetration rates
are generally maximised to ensure rapid hole generation. However as the hole
becomes progressively deeper, the time taken to pull and change the bit when worn ie
the tripping time, increases. IE-deep holes the trip_ping time can be considerable and
therefore more emphasis is placed on conserving bit life to reduce the time spent
tripping.

From this, it can be seen the optimum operating point between penetration rate and
wear rates is not constant throughout the length of the hole. The gepth affects the
position of the optimum operating point. It is also well known that wear and
penetration rates are not solely interdependent i.e. penetration rate is not a sole
function of wear rate and vice versa, but many other parameters effect one or both.
For example, rock type will effect both whereas abrasivity will only directly effect
wear. Furthermore, there are other parameters, possibly unknown, who's effects will
also be unknown. With the complex interaction of all these possible parameters, a
prototype system designed to cope with them would be extremely complex if not
impossible to develop.

Therefore, simplifications have to be made to enable a system to be developed. Once


this has been proved, other parameters which were otherwise ignored and held
constant, can be introduced into the optimisation system. Bearing this in mind. the
choice of a governing parameter and its relationship with others, should aim to
remove / accommodate as many parameters as possible, without becoming too
complex. Those that are excluded from it. but still have influence on the result, must
be dealt with separately.

42
4.2 The Governing Parameter and Its Relationship

As has been mentioned previously, to enable any optimisation system to be


developed, the desired aim of the optimisation system must be defined along with its
relationship to other parameters involved in the operation. In this case, the
optimisation system must seek out and maintain the most efficient operating point for
a drilling system. Therefore a parameter must be established to achieve this.

4.2.1 Maximised Penetration Rates

If penetration rate was selected, as described in the previous chapter, the optimisation
system could maintain maximum penetration rate throughout the entire length of the
hole. However as the hole becomes deeper, the increase in tripping time would
eventually become detrimental reducing the overall efficiency of hole generation. The
use of such a system to maximise penetration rate may be of great benefit where
tripping times remain low i.e. in short holes. These are more common in the Mining
Industry for blast hole production.

4.2.2 Time
The time taken to drill the hole was initially investigated as a possible control
parameter, as it is every drillers aim to complete the hole as quickly as possible.

-*,4.2.2.1 Minimum Time For Each Bit Run

The time taken for each bit run was initially investigated. This comprises the rotating
time of the bit, and the time taken to trip the bit out of the hole at the new depth. This
can be written as follows:-

(4.1) ',/

where Tb = Total time for each bit run.


Tr = Rotating time.
Tt = Tripping time

The time rotating can be expanded in terms of penetration rate(P) and expected
distance.

43
T - Expected Distance ___ (4.2)
r - Penetration Rate

The Expected Distance is governed by the wear rate (W), and K the expected bit life,
at the declared penetration rate(P), such that,

Expected Distance = ~ --- (4.3)


Therefore

Tr J;) - - (4.4)

1 K
Tr= P . W --- (4.5)

Similarly, the tripping time may also be expanded in to terms of Present Depth (D)
and the expected distance drilled with the bit, and the average round trip time per
metre (Tm).

T t = (Present Depth + Expected Distance) . Tm --- (4.6) f

Thus by substitution,

K
Tt=(D+ ) .Tm --- (4.7):
W

Therefore, substituting into equation (4.1), gives

1 K K
Tb = P . W + (D + W) .Tm --(4.8) I

Multiplying out, gives

1 K K
Tb = P . W + D. Tm + W . Tm --- (4.9)

By excluding the Present Depth term from the equation (being a constant for each
particular bit run), once an optimum point has been established for the first bit run,
then since no other influences will effect the equation, this operating point will remain
constant for subsequent bit runs. Therefore the drill parameters will remain constant
for the rest of the hole.

44
Clearly this is not desirable and would not provide a successful optimisation
algorithm. The failure of this method is that it does not include the effects of other bit
runs. Only by doing this i.e, looking at the hole as in entirety, does a time
optimisation system become possible. Therefore time to hole completion was
investigated.

4.2.2.2 Total Time to Completion

Equation 1 gives the time taken to complete one bit run. Therefore by sumating all the
predicted bit runs, a total time to completion can be estimated. This can be written as

Time to Completion (Tc)= L (T r + T t) --- (4.10)

Expanding this equation in a similar fashion,

Tc - L in K K
Wn + (Dn + w ) .Tm --- (4.11)
n
1

But the Present Depth (Dn) is the previous depth, plus the distance drilled by the bit
1.e.

K
Dn = D (n-1) + W --- (4.12)
(n-l)

Therefore we can sumate to the total distance tripped now in terms of the previous bit
runs, such that the total depth Dn is given by ;-

p=n
Dn - L Dp --- (4.13)
1

--- (4.14)

45
n-J
Dn =k~ (j-n+1) . _1 --- (4.15)
£..J wn
n=l

Therefore total time spent tripping is

n=]
T t = Tm· K L (j-n+1) . ~n --- (4.16)
n=1

and thus the total time to completion is given by,

n=]
K
Wn + Tm· K. L (j-n+ 1) . ~n --- (4.17)
n=l

This gives us an equation which will allow the sumation of total time to completion,
from various predicted penetration rates and wear rates, until hole bottom is reached.
However to find an optimum, the best combination of all the bit runs must be
ascertained to achieve minimum time to completion. Therefore, every combination
may have to be searched. For just a simple example, having only 5 combinations of
penetration rates and associated wear rates, it can be seen that there is a tremendous
build up in the number of possible combinations.

Number of Computations
1st Bit Run 5
2nd Bit Run 25
3rd Bit Run 125

Mathematically this can be described by the equation

Number of combinations = Number of associations (Number of bit runs)

For a real situation, the number of associations would be much higher. It may be
possible to eliminate some of the combinations, as the calculation proceeds, but what
has to be remembered is that some combinations may require less bit runs to reach the
required depth and hence less time is spent tripping. Therefore care has to be
exercised in eliminating these, to ensure that the true minimum time is derived.

46
In addition to this, the system is highly inflexible. Any changes in circumstances
resulting in early bit failure and / or not attaining the predicted distance drilled, such as
differing geology or hole problems, would disrupt the optimisation system.
Consequently it would have to be re-run to calculate the ~ew optimised time to
completion. Furthennore in the equation developed for optimising time to completion,
an assumption was made that the parameters would not change for the whole of the bit
run. This would obviously not be the case unless drilling was occurring in
homogeneous rock. To generate a system to cope with changing geology within each
bit run, would require an exact knowledge of hole geology and increase the
calculation time dramatically.

Clearly the inflexibility of this system and the computer time involved to predict the
best parameters to attain minimum time to completion makes this method of
optimising drill perfonnance unsatisfactory.

4.3 Cost

Other parameters were looked at but were quickly dismissed apart fonn some fonn of
cost optimisation. This parameter had the immediate benefit that virtually all other
parameters could be related to a cost function. The proposed criteria for the
-

optimisation system of trading off penetration rates and wear rates, can also be easily
accommodated. Penetration rates have a direct relationshi p to rotating and therefore rig
running costs. Wear rates not only influence bit consumption but also influence
tripping time intervals and hence rig costs. Provided an equation can be established
relating these parameters to cost, the criteria for the proposed optimisation scheme
could be met.

The other big advantage in using cost is that at the end of the day, profit is one of the
most important factors and reducing hole production costs is obviously beneficial. In
addition, it may be found that by using an optimisation system, wells originally
thought un-economic to drill, may prove to be economic.

However for a drilling operation, there are many cost centres, some of which have a
direct relation to the generation of the hole and others that do not. To ascertain those
that have some direct influence, a brief cost analysis was perfonned. Two categories
were produce: fixed cost (independent to the drilling operation) and variable costs (
those having some influence).

47
The fixed costs are mainly due to the preparatory work done before drilling can occur,
or afterwards when the site etc must be restored. Some of these are shown in Figure
4.1. These costs are mostly independent of the drilling perfonnance and cannot as
such be incorporated into a cost optimisation scheme. However these costs could be
reciuced, e.g by improving the efficiency of site restoration, or rig transportation, but
although they do bring benefits to the total cost to completion they do not reduce the
drilling cost directly.

The inclusion of the casing programme as a fixed cost is for simplicity, as the effects
of running casing in this optimisation scheme have been ignored. Generally the depth
at which casing is run is governed by the hole geology. Thus while not effecting the
drilling costs directly, an optimisation scheme that incorporates the casing programme
into its algorithm would be beneficial. This would allow the optimisation system to
ensure that when the required depth for casing was reached, the bit would be ready to
be pulled anyway.

I The variable costs do directly relate to the cost while drilling and these are shown in
. Figure 4.2. It can be seen, there are a large variety of cost centres, each having a
. differing influence on the overall drilling cost. Some are also dependant on external
, influences e.g. manpower costs vary geographically, fuel cost may vary with
, economic/ political influences.

It would be desirable to establish a relationship to include all of these variable costs


and develop an optimisation system, to minimise total drilling costs. However the
lessons learnt from the total time to completion, also apply here in that an inflexible
system would be developed, which would not be able to cope with problems and
unforeseen influences.

Therefore a simpler optimisation method is necessary. To do this, cost per metre was
proposed. This would allow the development of a system which could incorporate all
of the variable cost mentioned above, including external influences such as geographic
location, but allow the system to be very adaptable, as the criteria was to minimise the
cost of each metre drilled. While this may not achieve the minimum cost to
completion, it does ensure that drilling cost are minimised for the current situation.

An equation for the cost per metre already existed in most drilling text books, (Drilling
Practices Manual by P.L. Moore (47) ), and it was decided to use this as the basis
from which to develop the cost optimisation system.

48
( Rigging up/ Rigging down)
Casing Programme

Site Restoration

. .:"
. ..

Figure 4.1 Fixed Costs in a Drilling Programme

.... -;..

__- - - - 4 1 Fuel Costs


:-:::,::.:,:-:-:,: ','
.:::;. :-::::',

Figure 4.2 Variable Costs in a Drilling Programme

49
C= ... (4.18)

where C = Cost per Metre


B = Bit Costs
R = Rig Costs per Time Interval
Tr - Rotating Time
Tt = Tripping Time
F = Distance Drilled with Bit

It can be seen from the equation, that the variable costs have been further simplified,
and are combined into just two costs, bit and rig costs. While this generalises the
variable costs, it does have the great benefit of being a very simple equation to work
with, and to develop an optimisation system around. Once the system has been
developed, this equation can be expanded to include the other variable costs described
previously e.g geographic location may also have an influence on labour costs or a
fuel element may be required.

These could be included into the equation and expanded as follows.

B + R (T r + T t) + M (T r + T t) + P ·in {S. T r· }
c - F
---(4.19)

where M - Manpower Costs per Time Interval


P - Unit Power Costs
S - Rotational Speed
in = An Unknown Function

With the correct system design, the upgrade to a more complicated cost equation
would only require a straight equation swap, rather than a complete redesign of the
optimisation system.

Returning to the original aim of the optimisation system, the basic rational was to
locate and maintain the optimum operating performance of a drilling system by using
the trade off between penetration rates and wear rates. To do this, cost per metre has
been chosen as the optimisation parameter and its relationship given in equation
(4.18). However it can be seen that none of these terms directly relate to any of the

50
controllable drill parameters. Therefore some mathematical manipulation of this
equation must take place to convert this equation into one which will be suitable for
the establishment of an optimisation system.

Multiplying out equation (4.18), gives us :-

B R . Tr R . Tt
C = F + F + F --- (4.20)

As with the equations for minimisation of total time to completion, the distance drilled
by each bit is determined by the total bit life (K) and its wear rate (W), we have:-

Distance Drilled = ~ --- (4.21)

Also,

Distance Drilled (F) ___ (4.22)


Penetration Rate (P)
Time Taken (Tr)

Substituting in we have,

B R R .T t
C=--- + p + (~) --- (4.23)
(~)
Therefore,

C = B .W + R + R.T t .W ___ (4.24)


K P K

If the tripping time is expressed on terms of the average time taken for a round trip
per meter, using D for depth, we can calculate the tripping time at any depth, thus

T t =D. Tm --- (4.25)

where T m = Average Round Trip Time per Metre

51
and therefore,

Simplifying gives,

= (B + R.T m . D) .W R
C K + P --- (4.27)

This equation provides a means of comparing combinations of drilling parameters


through minimum cost per metre, with the ultimate intention of finding a minimum.
Returning to Figure 1.1, this equation would form the rule base for which decisions
would be made to improve the current operating position.

However to do this, both parameters i.e. penetration rates and wear rates must be
readily available as the absence of one would totally invalidate the optimisation
system. Furthermore, with the data available, how is the equation used to determine
the optimum operating point, and is the method used the most efficient? These two
issues are covered by the following chapters.

4.4 Cost Equation Sensitivity Analysis

To understand the effects and trends that the various parameters have in the cost
equation (4.27), a sensitively analysis was performed. For each test, four different
combinations of penetration rate and wear rate were used to cover a broad range of
possibilities. They comprised of worst and best scenarios as shown below :-

1) Low Penetration Rate with High Wear Rate.


2) High Penetration Rate wi th Low Wear Rate
3) Low Penetration Rate With Low Wear Rate
4) High Penetration Rate with High Wear Rate

Although some of these scenarios are unlikely in reality e.g. high penetration rate with
low wear rate, they are included to show the boundaries of the system. The values for
each of these conditions was based on work by Ambrose. They are shown below and
were kept constant for each test.

52
Low Penetration Rate = 50 mm/min
High Penetration Rate =350 mm/min
Low Wear Rate = 0.05 mm/m
( assuming a 10mm crown Height =200m of drilling)
High Wear Rate = 0.2 mm/m
( assuming a 10mm crown Height =50m of drilling)

Many different graphs were produced relating various parameters to others, but plots
of cost against depth illustrate the observations well.

4.4.1 The Effect of Bit and Rig Cost with Respect to Tripping
Time

The first series of tests conducted were to establish the effect of Bit and Rig costs,
with respect to the Tripping time per Metre. For each test, three tripping times would
be used 10,20 and 40 s/m, shown by the three graphs in each of these tests.

The Rig and Bit costs used are shown below:-

1) Bit Costs Greater than Rig Costs (Bit Costs = £1500, Rig Costs = £100)
2) Bit Costs Equal to Rig Costs (Bit Costs = £1000, Rig Costs = £1000)
3) Bit Costs Less than Rig Costs (Bits costs = £1000, Rig Cost = £10000)

4.4.1.1 Bit Costs Greater than Rig costs

From the three graphs shown in Figure 4.3, it can be seen that the tripping time has
very little effect at all on the cost per metre increasing the cost only marginally with
increasing depth. It can also be distinctly seen the remarkable division of the four
penetration and wear rate scenarios into wear rate zones i.e. high and low wear rates.
If high wear rates were allowed to occur, then the cost per metre would rise
dramatically. Therefore in this situation, low wear rate are the predominant criteria.

4.4.1.2 Bit Cost Equal to Rig Costs

Figure 4.4 shows that with rig and bit costs even, the tripping time is beginning to
effect the cost per metre as depth increases. With tripping times slow i.e. 40 slm, the
cost at the greatest depth is nearly double of that when faster tripping times are
obtained. The division of wear rates still exists, indicating that wear rates should be

53
Tripping Time = 10s/m
40

+ Low Pen. Rates and


30~~======~======~ High Wear Rates

~ High Pen. Rates and


Low Wear Rates
20

o
- co
o
. . Low Pen. Rates and
Low Wear Rates

10Jt1=======~========~ ~ High Pen. Rates and


High Wear Rates
OT-~-r---,~ __ ~~~ ________
o 2000 4000 6000 8000 10000 12000
Depth (Metres)

40
Tripping Time = 20s/m

. . Low Pen. Rates and


High Wear Rates

~ High Pen. Rates and


...G Low Wear Rates
c.. 20
-
o
o
co . . Low Pen. Rates and
Low Wear Rates

10Jt-=======~========~ ~ High Pen. Rates and


High Wear Rates

o 2000 4000 6000 8000 10000 12000


Depth (Metres)

Tripping Time = 40s/m


50

;:)-
Low Pen. Rates and
40 High Wear Rates

-...
G

G ~
High Pen. Rates and

..
:E 30 Low Wear Rates
...G
0.. Low Pen. Rates and
-
co
0
()
20
Low Wear Rates

. . High Pen. Rates and


10
High Wear Rates

o 2000 4000 6000 8000 10000 12000


Depth (Metres)

Figure 4.3 The Influence of Tripping Time when Bit Costs (£1500) are
Greater than Rig Costs (£100)

54
Tripping Time = 10s/m
60

... Low Pen. Rates and

-..
., 50 High Wear Rates
.,
...,
::::e 40 ... High Pen. Rates and
Low Wear Rates
a.. 30

-•
0
(.)
20
. . Low Pen. Rates and
Low Wear Rates

... High Pen. Rates and


10 High Wear Rates

0
0 2000 4000 6000 8000 10000 12000
Depth (Metres)

100
Tripping Time = 20s/m
. . Low Pen. Rates and

-..
., 80 High Wear Rates
G)
::::e ... High Pen. Rates and
...
G)
60 Low Wear Rates
a..
-•
0
(.)
40
. . Low Pen. Rates and
Low Wear Rates

... High Pen. Rates and


20 High Wear Rates

o~~-.--~.-~-.--~~~~~~
o 2000 4000 6000 8000 10000 12000
Depth (Metres)

Tripping Time = 40s/m


140
Low Pen. Rates and
120 + High Wear Rates
.,...
-G)
::::e
100

80
..... High Pen. Rates and
Low Wear Rates
.,... . . Low Pen. Rates and
a.. 60
-•
0
(.)
40
Low Wear Rates

... High Pen. Rates and


High Wear Rates
20

0
0 2000 4000 6000 8000 10000 12000
Depth (Metres)

Figure 4.4 The Influence of Tripping Time when Bit Costs (£1000) are
Equal to Rig Costs (£1000)

55
kept as low as possible, but within these divisions, the differing penetration rates is
also causing a separation. It can also be seen that at shallower depths, the separation
between the two wear rate zones has been reduced substantially.

4.4.1.3 Bit Costs Less than Rig Costs

It can be seen from Figure 4.5 that the continuing separation of the penetration rates
and the reduction in separation of the wear rate zones, has occurred to such an extent
that two of the lines cross, i.e. Low Penetration Rates and Low Wear Rates, and High
Penetration Rate and High Wear Rate. This indicates that at different depths of the
bore hole, different criteria are required. The initial part of the hole requires high
penetration rates, out weighing the associated high wear rates, and in the latter part,
low wear rates become more important.

It can also be seen that increasing tripping time again increase the cost per metre
within increasing depth but more importantly, moves the position of the change over
point between high penetration rates and low wear rates, thus indicating benefits are
realised if tripping rates are kept as high as practicably possible.

4.4.2 The Effect of Bit Costs with Constant Rig Costs

The last two tests performed, were to show the effect in cost per metre of differing bit
costs with constant rig costs ( one at low rig cost £100, the other high £10000), and
constant tripping times, again plotted against depth. The two bit cost used were £500
and £1500.

4.4.2.1 Low Rig Costs.

Figure 4.6 shows that as with the test shown in Figure 4.3, it is essential that when
bit costs are greater that rig costs, wear rates should be kept to a minimum, otherwise
the cost per metre increases dramatically. This is highlighted further by the higher of
the bit cost graphs, where the cost per metre would be tripled is high wear rates were
to occur.

56
Tripping Time = 10s/m
400
+ Low Pen. Rates and
., High Wear Rates
-
.....,
::E
300
.... High Pen. Rates and
.,
a- Low Wear Rates

k::=-----:~?---
a.
-
."
o
o
200 . . Low Pen. Rates and
Low Wear Rates
100 .... High Pen. Rates and
High Wear Rates

O~~~~~~~_r~~~ __~~__
o 2000 4000 6000 8000 10000 12000
Depth (Metres)

Tripping Time = 20s/m


700

600
+ Low Pen. Rates and
High Wear Rates
-CD
a-
CD
::E
500
.... High Pen. Rates and
a- 400 Low Wear Rates
CD
a.
-."
0
300 . . Low Pen. Rates and
Low Wear Rates
0 200
... High Pen. Rates and
100 High Wear Rates

0
0 2000 4000 6000 8000 10000 12000
Depth (Metres)

Tripping Time = 40s/m


1200
Low Pen. Rates and
+ High Wear Rates
1000
.,...
-
CD
::E
800 ... High Pen. Rates and
Low Wear Rates
...CD GOO
D. .... Low Pen. Rates and

-
."
0
0
400
Low Wear Rates

... High Pen. Rates and


200 High Wear Rates

0
0 2000 4000 6000 8000 10000 12000
Depth (Metres)

Figure 4.5 The Influence of Tripping Time when Bit Costs (£1000) are
Less than Rig Costs (£10000)

57
Bit Costs = £500
40

... Low Pen. Rates and

-.....
High Wear Rates

::E
...
30
... High Pen. Rates and

..
Low Wear Rates
Q. 20

0
-..
0
Low Pen. Rates and
low Wear Rates
10 ... High Pen. Rates and
High Wear Rates

O~~~~---r~~~--~--~~~~
o 2000 4000 6000 8000 10000 12000
Depth (Metres)

Bit Costs = £1500


20

.-..
... Low Pen. Rates and
High Wear Rates

Q)
... High Pen. Rates and
::E

0.
.... 10
Low Wear Rates

-o'"
o
. . Low Pen. Rates and
Low Wear Rates

High Pen. Rates and


.... High Wear Rates

04-~~~ __-r~~~-r-,--~~~--,
o 2000 4000 6000 8000 10000 12000
Depth (Metres)

Figure 4.6 The Influence of Bit Costs Compared to Constant Rig Costs
(£100) and Constant Tripping Time (20 s/m)

58
4.4.2.2 High Rig Costs

Figure 4.7 illustrates that when rig costs are much higher than bit costs, that the
influence of bit costs on the cost per metre is negligible. This is shown aptly by the
superimposed graph at the bottom.

4.4.3 Conclusions of the Sensitivity Analysis

From the analysis, we can draw several conclusions :-

i ) With rig costs less or equal than bit costs, emphasis is placed on low wear rate, as
the effect of higher wear values dramatically increase the cost per metre. This is
increasingly important with high bit costs. Tripping time has very little effect of the
cost per metre.

ii) With Rig costs higher than bit costs, bit costs have very little effect on the overall
cost per metre. Tripping times however, significantly effect the cost per metre.
Increase tripping time, cause increases in the cost per metre, especially if high wear
rates are seen. The effect is much less with low wear lates. A cross over point is also
seen where, after an initial requirement of maximising penetration rates, the emphasis
is change to low wear rates. The depth at which this interchange occurs is effected by
the tripping time, being dramatically reduced by increasing tripping times

4.5 Conclusion

In concluding, to achieve optimisation of a process, a governing parameter and its


relation to that process must be developed. In selecting the optimum parameter, it is
beneficial to select one that eliminates as many variables as possible, to enhance the
development of the optimisation system. In this case, a trade off between penetration
rates and wear rates was selected as the rational for drill optimisation. Several
parameters were examined e.g maximised penetration rate, time per bit run, total time
to completion, but most were disregarded, due to non-practicality, and / or the
inflexibility of the likely optimisation system.

Optimisation by cost per metre was selected as the governing parameter and a well
established cost equation was used. By mathematical manipulation of this equation,
parameters directly related to the drilling operation were derived to establish the rule
base for the optimisation system.

59
Bit Costs = £500
800
700 + Low Pen. Rates and

....,.,
High Wear Rates
-
::i:
600
500 . . High Pen. Rates and
...., Low Wear Rates
Q. 400

0
-In
0
300
. . Low Pen. Rates and
Low Wear Rates
200 . . High Pen. Rates and
100 High Wear Rates

0
0 2000 4000 6000 8000 10000 12000
Depth (Metres)

800
Bit Costs = £1500

Low Pen. Rates and


+
.....,,
..
High Wear Rates
-
~
600
High Pen. Rates and
.,...
..
Low Wear Rates
Il. 400
-In
0
Low Pen. Rates and
Low Wear Rates
0
200 . . High Pen. Rates and
High Wear Rates

o 2000 4000 6000 8000 10000 12000


Depth (Metres)

Superimposed Graphs
800
Low Pen. Rates and
"0-
High Wear Rates

.....,, 600
-
::i:
.... High Pen. Rates and
Low Wear Rates
...CD 400
Il. . . Low Pen. Rates and

-
In
0
200
Low Wear Rates

High Pen. Rates and


0
. . High Wear Rates

O~~~~~-r~~~--~--~~~-'
o 2000 4000 6000 8000 10000 12000
Depth (Metres)

Figure 4.7 The Influence of Bit costs Compared to Constant Rig Costs
(£10000) and Constant Tripping Time (20 s/m)

60
To develop general understanding of this equation, a sensitivity analysis was
perfonned on this equation to indicate the trends etc, associated with it. This
highlighted several points when dealing with differing rig costs, tripping times etc,
which were useful when developing the minim urn cost optimisation system.

61
Chapter 5 - Bit Wear and Wear Prediction

5.1 Introduction

The monitoring of most drill parameters in the laboratory is relatively easy. In the field
however, difficulties do arise as a result of current technology limitations and
engineering constraints. As a consequence, drill parameter monitoring becomes
difficult. Furthermore when considering deep holes, the surface monitored data may
not depict the true down hole conditions. Despite this however, it is in the authors
opinion that with increased exploitation of Measurement While Drilling Tools (MWD)
and other techniques, accurate and reliable data will be available for most parameters in
the future. The one exception to this is possibly bit wear.

Bit wear is a complex subject involving many different processes, which determine the
type and extent of bit wear. Different bits wear in different ways, but some processes
are common to all such as abrasion. The correct selection of bit type, may play the most
crucial part in minimising bit wear and enhancing performance. Despite correct
selection, wear and damage will always occur, and therefore it is important that
research is conducted into bit wear to understand the processes that govern it. This is
not only important to the bit manufacturer, for improving cutting performance and
reducing wear characteristics, but also to the driller who needs to know when the bits
life is exhausted and thus reduce the chance of pulling a green bit.

Throughout drilling history, research has been conducted into bit wear, and
consequently many schemes and systems have been produced to minimise wear each
having differing degrees of success (2,5,9,11,13,14,19,25,28,72). Tri-cone roller bits
have probably been the most extensively researched. They were originally regarded as
the work horse of the oil drilling industry, still extensively used but now being
surpassed by PDC bits. Tri-cones are also predominantly used for large diameter blast-
hole production for surface mining operations.

Prediction of the bit wear mechanism for a tri-cone bit is complicated having both a
wear mechanism common to all types of bits, as well as a bearing failure mode.
However previous research, has lead to a good understanding of the wear modes and
prediction of failure. Consequently several formulas and prediction methods are in
current use with both the oil industry and surface mining industries.

62
Research at the university, has f?cused just on two types of bits, primarily diamond
impregnated core bits, and latterly polycrystaline diamond bits for use in roof bolting.
Diamond impregnated bits are generally thought of as having fairly uniform wear
characteristics (11,70,73), the cutting matrix wearing away to expose new diamonds at
a relatively constant rate, for constant drill parameters. This enables wear characteristics
to be determined fairly easily. Much of the work into diamond bit wear was conducted
by Ambrose. The author continued some of this work, mainly in the improvement of
the accuracy of the wear measurement jig. Determination of further wear characterises
were not undertaken due to time constraints. Research on the PDC side is fairly new
and the author has had no direct involvement. PDC itself, has good wear characteristics
being comprised of sintered diamond, but it is prone to impact damage and is intolerant
of maltreatment. Its introduction into the drilling industry has been fairly recent.
Consequently the wear mechanisms are still under investigation to determine modes of
wear and best performance scenarios (9,68).

Attaining wear data in the laboratory is relatively easy. However to measure real field
wear data and to construct and validate wear models is very difficult as in a commercial
operation, the tripping of the bit every set distance to measure the incurred wear is
highly impractical. The actual measurement of the wear is also complicated and often
prone to subjective interpretation, such as in the case of the lADe wear code for tri-
cone roller bits.

Furthermore, having attained a wear value, how do we depict what parameters, have
influenced the wear which has occurred. Obviously in the field, this is very difficult as
we do not know exactly what bottom hole conditions are. The laboratory environment,
while not imitating the real situation fully, does enable controlled experiments to be
performed, and wear characteristics to be taken with relative ease. Once adequate data
has been gathered and a wear theory developed, comparisons can be made with those
from the field and validation made.

'5.2 The Wear Predictor

The aim of this research project was not to enhance knowledge about the mechanism of
wear for a particular bit, but to develop a system which would enhance the data already
available and predict likely wear values for any type of bit.

Most current wear predictors work on a formula bases, or use rules established from
data bases. A typical example is shown below for Tri-cone bit wear, F.S.Young (74).

63
Rate of Bit Bearing Wear

:=~ .N.Wcr - (5.1)

Rate of Bit Tooth Wear

dH Af (PN + QN3)
dtr = (-D 1.W+D2)(1+C 1.H) --- (5.2)

where B - Normalised bit wear


tr - Rotating time
N - Rotary Speed (r.p.m)
W - Bit Weight, (pounds)
cr Weight exponent in bearing rate equation
H - Normalised tooth wear height
P & Q & C1 = Constants dependent on bit type
Dl'and D2 = Constants dependant upon bit size
Af = Formation abrasiveness factor

By entering the required values, an estimation can be made of the extent and type of
wear / failure likely to occur. However these equation are rigid, and if inaccurate will
always remain so.

The emphasis placed on this predictor was not to be tied to any particular formula or
rule base, but to predict solely on data collected previously, from other holes. An
analogy can de drawn, to a brain learning a process for itself. With the large number of
possible influences on wear, a system that could be expanded to accept these is also
desirable.

To produce such a system data points must be stored with distinct set of reference
parameters, and stored in such a way as to allow the inclusion of additional reference
parameters. This is best done in a matrix format. For different combinations of
parameters, each has a unique accessible data point. Additional parameters can be
accommodated by adding another dimension to the matrix. These matrix tables were
named S.L.P.M.'s (Self Learning Prediction Matrices), the self-learning aspect is
described later in this chapter. Therefore, by measuring wear values, these can be
loaded into the S.L.P.M.'s with respect to their reference parameters. They may be
either historical data or from on line measurements.

64
5.2.1 Data Validity

With progressive use of the predictor it is likely that over time several values will be
generated for one particular point. Therefore, some sort of averaging process is
necessary. One of the more common methods for continual data streams is the running
average given by :-

New Reading + Previous Average


Running Average = 2 --- '(5.3)

This method can also be used to generate more a general average by using say the sum
of the previous five or ten readings. However this method does not give a true
average, and more importantly, it does not give any indication of the accuracy/ range of
the data such as given by the standard deviation. To calculate the mean and standard
deviation, requires the storage of all the previous data values. With the matrix system
proposed, this would be undesirable, as over time the amount of data stored would be
colossal. Therefore ways were investigated to reduce the number of values or variables
which would be required for each point.

The Mean is given by

- Sum of all values


x = Number of values --- (5.4)

-
x =
Sum of previous values + New value n_ (5.5)
Number of values

-x - (No· Xo)+ New value


--- (5.6)
No + 1

where No = Previous number of values


-
Yo = Previous mean

It can be seen that to calculate the mean, only two values have to be stored, i.e. the
mean and the number of data points entered. Small errors do occur from progressive

65
multiplication of the mean value but these are negligible. The standard deviation may
be calculated in a similar fashion.

Standard Deviation = -V Variance --- 5.7

L( x _x)2
Variance = - - - - - --- 5.8
n

By multiplying out and reducing,

Variance I n x2 _ ( x) 2 --- 5.9

From this it can be seen that we need to store the 'sum of the squares' of each point
along with the mean and the number of data points. Thus to calculate both the true mean
and the standard deviation, we need only need store three values.

The mean
The num ber of data Values
The sum of the squares

Compared to the original suggestion of storing every data point, this method provides a
usable and accurate means for determining data reliability.

Summarising what has be proposed so far; as wear is a difficult parameter to measure,


we require some sort of prediction mechanism to provide estimation of wear for the
optimisation scheme. This predictor has been developed in the fonn of a matrix system
(S.L.P.M.) which may be readily expanded to multi dimensions for the various
parameters required. To provide some indication of the accuracy of the data, the Mean
and standard deviation are calculated for each point, but only necessitating three storage
variable for each point.

5.2.2 Data Enhancement by Interpolation

When using the predictor, to start with data values will be sparse, and consequently
predicted values will be inaccurate but with progressive use the predictor becomes more
and more accurate. However as the reader has probably already questioned, for a large
matrix, there are a large number of data points to fill. For example a 20 by 20 has 400

66
points! This is clearly impractical to fill both in the laboratory and in the field. Therefore
we need some way of enhancing the data which we load into the S.L.P.M.'s, i.e. a
self-learning capability, such that we can estimate values for unknown points. This can
be achieved by interpolation.

Interpolation of data is always subject to risk as the interpolated values may generate
totally meaningless data which does not represent the true process. However with the
S.L.P.M.'s having a continual learning process i.e. data will be continually entered, it
is felt that if interpolation takes place after each addition, a reliable system will be
produced.

5.2.2.1 Two Dimensional Interpolation

There are many ways in which we could 'interpolate' the data with in the matrix. A
simple system could give every unknown point the average value of all the known
points. This however, would give unrealistic values as no account is taken of the trend /
surface variation of the data. This is shown in Figure 5.1.

Therefore some sort of curve fitting or general averaging method is required. As an


averaging process is more readily applied and easier to develop, initial work
concentrated on this type, with the idea of investigating curve fitting etc, at a later date
once the main optimisation system had been developed.

For each value with in the matrix if the point was an unknown, the four values directly
adjacent to the selected value, were averaged. This value then replaced the original
selected value, as shown below.

2 2

I I
3- .
?- 8 3- 5- 8

I5 I5

Using this as the bases of the averaging process. The original method of applying this
technique was to interpolate the whole matrix starting from one comer and ending in the
opposite comer. The process would continue until no change was seen in the matrix.

67
True Process

-- .. - - --

.. ..
Es timated Process . ..
.. Mean of all Known Values
.
.. .
• •
• ••
••

.

.. .
.. .. ..
.. ..

.. •

~ _ _ _ _-+-+--_ _....:=::._ Known Values

. .
. . ..

Figure 5.1 An Illustration to Show the Effect of Giving the Mean Values
to Unknown Points

68
This however is extremely wasteful of processing time as generally the influence of the
new value was restricted to a smail part of the matrix, and therefore did not require the
whole matrix to be interpolated.

To improve the efficiency of the interpolation system, a technique which later became
known as the ripple technique was developed. When a new value was entered into the
matrix, the averaging process, would radiated outwards in concentric squares (if 2
dimensional) until the effects were minimal. This would repeat itself starting from the
new value once again, until no change was seen.

The original method was to interpolate the whole of each growing square. However, as
the squares radiate outwards the inner average values would never change, and thus
computing time would be wasted. Therefore only the periphery values of each square
would be used as shown.

First
Ripple

14-_ _ Second _ __
Ripple

Old Method New Method

A similar action is seen by ripples ( hence the name ) radiating outward from its
disturbance point in water. Figure 5.2 shows a small two by two matrix undergoing
rippling. It can be seen from this that the closer the points are effected to a greater
extent, than those further away.

5.2.2.2 Three Dimensional Ripple.

The ripple method described so far has only been for two dimensions and is thus easy
to visualise and develop. For three dimensions, a cubic like structure is see. Therefore
an average of the six adjacent values to the selected point must be taken forming a three
dimensional cross, as shown overleaf.

69
a 0 0 1 2 3 4 5 6 6 6 5 4 4 3 2 1 0 0 0
0 a 0 1 2 3 5 6 7 7 7 6 5 4 3 2 1 0 0 0
0 0 0 1 3 4 6 8 9 9 9 7 5 4 3 2 1 0 0 0
0 0 1 2 3 5 8 10 12 13 12 9 7 5 4 3 1 0 0 0
0 0 1 2 4 7 10
13 17 19 17 13 9 7 5 3
1 0 a 0
a 0 1 3 5 8 12
17 24 31 24 17 11 8 5 1 0
3 0 a
a 0 1 3 5 8 13
19 31 58 31 19 12 8 5 1 a
3 0 a
0 0 1 3 5 8 12
17 24 31 24 17 11 7 5 3
1 0 0 a
0 a 1 3 5 7 10
13 17 19 17 13 9 6 4 2
1 0 0 0
0 0 1 3 4 5 79 11 12 11 9 7 5 3 2
1 0 0 0
0 0 1 2 3 4 56 7 8 8 7 5 4 3 1
0 0 0 0
0 a 1 2 2 3 34 555 5 4 3 2 1
0 0 0 0
0 a 1 1 1 1 22 3 3 3 3 3 2 2 1 a 0 a 0
0 0 a 0 0 0 1 1 1 1 1 1 1 1j 1 1 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 a 0 0 0 0 0 0 0 0 0 0 a 0 0 0
a 0 0 0 0 0 0 0 a 0 0 0 0 0 a 0 a 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Figure 5.2 A Two by Two Matrix which has Undergone Rippling After a
Value has Been Entered

70
8 8

7 ___ 1/4 7 __ 1/4


?
• - 6- -6
9/1 9/1
3 3

The actual mode of rippling will take a cuboidal shell. However, this may be split down
into six two dimensional squares similar to those of the two dimensional ripple.
However unlike the 2-D ripple where just the periphery is averaged, the whole square
must be averaged.

e.g Side 1 = Constant -Y, Average (X,Z)

5.2.2.3 Multi- Dimensional Interpolation

It may be necessary for dimensions greater than three, but these are beyond the capacity
of the human brain. Therefore a pattern must be established which links the addition
of extra dimensions.

Considering the two dimensional ripple, rippling occurs in a square like function , with
the respective areas being held either constant or rippled. This is summarised below:-

Y
,I 1
4 2
1. Constant +Y, Ripple -X to +X

2. Constant +X, Ripple +Y to-Y

L3 ~ 3. Constant - Y, Ripple +X to -X

4. Constant -X, Ripple -Y to +Y


x

71
In the three dimension ripple, the steps are listed below.

1. Constant -Y, Average (X,Z)


z 2. Constant +Y, Average (X,Z)
3. Constant -Z, A verage (X, Y)
4. Constant +Z, Average (X, Y)
Y
5. Constant -X, Average (Y,Z)
x 6. Constant +X, Average (Y,Z)

From the two dimension case, it looks as though no pattern exists. However as
mentioned previously, in the 2D version only the peripheral values were used, as
averaging the whole of each growing square was inefficient. However if this method
was used a pattern could be established.

2D :- Average (X,Y)

3D :- Constant 7Z Average (X, Y)


Constant 7Y Average (X,Z)
Constant 7 X Average (Y,Z)

Thus for a four dimensional ripple, the selected would require the average of the eight
adjacent values, W 71 , X 71, Y 71, Z 71, and ripple in a fashion shown below.

Constant 7 W Constant 7Z A verage (X, Y)


Constant 7 Y Average (X,Z)
Constant 7 X Average (Y,Z)

Constant 7 X Constant 2" Z Average (W,Y)


Constant 7 Y Average (W,Z)
Constant 2" W Average (Y,Z)

Constant 7Y Constant 7Z Average (X,W)


Constant 2" W Average (X,Z)
Constant 2" X Average (W,Z)

Constant 7Z Constant ~ W A verage (X, Y)


Constant ~ Y Average (X,W)
Constant ~ X Average (Y, W)

72
5.3 Testing of the Predictor

For the initial development of the predictors only two and three dimensional ripple
methods were developed. To test the accuracy of prediction, the matrices were loaded
with a set number of known points, with rippling occurring after each addition. To
generate the known data points, random co-ordinates were fed into a known equation
and the value calculated. Once the required number of points had been entered, using
the known equation, the accuracy of the interpolation could be determined.

Several test were performed, by varying the number of entered values as well as the
form of the equation. This allowed equations / surfaces of varying complexity to be
tested.

5.4 Results

For a measure of the prediction accuracy a limit was set in which the data had to be
within the ~ 10 % of the real value, the greater number in this range, the greater the
accuracy.

5.4.1 Two Dimensional Method.

Figure 5.3 shows a three dimensional plot of the equation / function which was used to
perform the fIrst series of tests. Table 5.1 shows the results of these tests, with Figures
5.4 - 5.6, giving typical examples. It can be seen that initially with only limited data in
the S.L.P.M.'s, the predicted results are highly inaccurate. However after the
S.L.P.M. is 5% full, there is a dramatic improvement in the accuracy, i.e. with the
S.L.P.M. only 10% full, 77% of the data is acceptable, and with the S.L.P.M. 37.5%
full, 97% is acceptable. This is also reflected in the time taken to ripple each of the
tests, the majority of time taken with the first 5%, and only small increases there after.
Therefore for simple surfaces, the ripple method proved a good predictor.

Figure 5.7 shows the three dimensional dome like structure generated by the equation
for the second series of experiments. The results are shown in Table 5.2 and Figures
5.8 - 5.10. From these results it can be seen that the accuracy of the prediction method
is reduced. Despite this, the predictor still performs adequately once 15-20% of the
S.L.P.M. is full, with 70% of the data being with in the tolerance. As the surface
changes more rapidly than in test one, the influence of each point to the matrix as a
whole is reduced and hence the very low rippling times are seen.

73
Equation: 2X + 3Y

Figure 5.3 A 3-Dimensional Plot of the Equation Used for the First
Set of 2-Dimensional Ripple Tests

74
Table 5.1· 2-Qimensional Ripple Test 1

Matrix Size :- 20*20 Number of Runs per Test 20

Equation :- 2*X + 3*Y.

Test Number of Error in Average Average Average time


Number known interpolated number percentage taken to ripple
values values. ill error ill error all values
range range. (seconds)

<10% 25.25 63.51 %


<20% 44.10 11.02%
1 20 29.10
(5%) <50% 76.60 19.15%
>50% 254.05 6.31%

<10% 308.55 77.14%


<20% 26.65 12.94%
2 40 31.25
<50% 51.75 6.66%
(10%)
>50% 13.05 3.24%

<10% 356.95 89.24%


<20% 23.20 5.80%
3 60 32.95
(15%) <50% 13.10 3.27%
>50% 6.75 1.67%

<10% 355.60 88.90%


<20% 25.40 6.35%
4 80 34.90
<50% 14.05 3.51%
(20%)
>50% 4.95 1.24%

<10% 372.10 93.02%


<20% 16.00 4.00%
5 100 34.85
<50% 8.30 2.07%
(25%)
>50% 3.60 0.90%

<10% 386.85 96.71 %


<20% 8.55 2.14%
150 36.85
6 3.65 0.91%
(37.5%) <50%
>50% 0.95 0.24%

75
Equation :- 2X + 3Y

62 64 60 59 59 60 62 64 66 68 71 74 78 76 7 73 73 73 73 73
59 59 58 5 5Q 60 62 64 66 68 71 75 83 77 7 1 I 73 73 73
56 56 56 57 58 60 62 65 66 68 70 73 76 75 73 73 73 7 73 73
54 54 54 55 56 58 61 67 66 67 69 72 73 73 72 72 73 7 I

52 52 52 53 5~ 56 59 62 64 66 68 72 71 71 71 72 73 74 76 79
50 50 50 51 52 54 57 59 61 63 65 67 68 69 70 72 74 75 78 85
48 48 48 49 50 52 54 56 58 60 62 64 56 68 69 71 76 74 75 77
46 46 46 47 48 50 52 54 56 58 60 62 64 67 67 68 70 70 71 71
43 43 44 45 46 48 50 52 54 56 58 60 61 63 64 65 66 66 66 66
40 40 41 42 43 45 47 49 51 53 55 57 58 59 60 61 62 62 62 62
3 34 37 38 40 42 44 46 48 50 52 54 55 56 57 58 59 59 59 59
35 35 36 37 38 40 42 44 46 48 49 51 53 54 55 56 58 57 57 57
34 34 34 34 36 38 40 41 43 45 46 48 50 51 52 54 58 55 54 54
32 32 32 29 34 36 37 37 40 42 43 45 47 48 49 51 52 52 51 51
31 31 31 1 33 35 36 37 38 39 40 42 44 45 46 47 48 49 48 48
31 31 31 32 33 34 35 J 36 35 38 39 41 42 43 44 45 46 46 46
31 31 31 32 32 33 34 34 33 34 36 37 39 40 41 42 43 44 45 44
31 31 31 32 32 33 33 32 27 34 35 37 38 39 40 42 43 44 44
31 31 31 32 32 32 32 32 31 32 32 32 34 36 38 38 41 43 44 44
31 31 31 31 31 31 31 31 31 31 30 27 32 35 31 38 40 42 43 44

Random entered values = 20


Interpolated values with <101. error= 244
Inter 01 ed s i 10-2 r r= 7
Interpolated values with 20r.-50r. error= 33
Interpolated values with >50r. error = 28

Figure 5.4 A Typical Result from the First 2-Ditnensional Test Showing
the Resulting S.L.P.M. Values after 20 RandomValues have Been Entered

76
Equation :- 2X + 3Y
62 60 60 62 63 6S 70 76 73 73 75 77 80 84 90 90 91 93 96100
59 59 60 62 63 65 68 71 71 72 74 76 79 82 86 89 91 92 95 96
56 57 59 62 63 64 66 68 69 71 73 76 78 80 83 85 87 89 92 94
53 55 57 59 61 63 64 66 67 69 72 75 76 78 80 82 84 86 88 89
52 53 55 56 58 60 62 64 65 67 70 72 74 76 78 80 82 84 86 86
50 51 52 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 82 82
~~~OO~~$~~~~6S~ron~~~ ~ ~
46 46 46 47 49 51 53 55 57 59 61 S3 65 67 69 71 73 75 76 76
43 43 43 44 46 48 50 52 54 56 58 60 62 64 66 S8 70 72 74 73
40 40 40 41 43 45 47 49 51 53 55 57 59 61 63 65 66 67 68 68
37 37 36 38 40 42 44 46 48 50 52 54 56 58 60 61 62 63 6~ 64
33 34 34 35 37 40 42 43 45 47 49 51 53 55 57 58 59 60 61 61
29 30 30 32 34 37 39 40 43 45 46 48 50 52 54 56 57 58 59 59
23 27 28 29 31 34 37 39 41 42 44 46 48 49 51 53 55 56 57 57
24 25 25 27 29 32 34 36 38 38 41 43 45 46 48 50 52 54 56 55
23 23 21 23 25 28 31 34 36 38 40 41 42 43 45 47 49 51 52 52
21 21 20 20 22 24 28 31 34 36 38 39 38 40 43 45 47 49 50 50
18 18 18 17 19 23 26 28 31 33 36 38 39 40 42 43 45 46 47 49
14 14 16 18 20 22 23 22 27 29 33 36 38 40 40 38 42 42 45 46
11 7 13 17 20 22 23 23 24 23 30 35 38 39 40 40 42 3 44 45

Random entered values = 60


Interpolated values with <10% error= 283
Interp dues wi h - ~ e r = 4
Interpolated values with 20X-50X error= 17
Interpolated values with >50X error = 6

Figure 5.5 A Typical Result from the First 2-Dimensional Test Showing
the Resulting S.L.P.M. Values after 60 Random Values have Been Entered

77
Equation :- 2X + 3Y
62 64 65 68 70 72 74 74 76 80 82 82 83 86 90 89 90 91 90 90
59 61 63 65 67 69 71 73 75 77 79 80 82 85 87 88 91 93 90 89
57 58 60 62 64 66 68 70 72 74 76 78 80 82 84 86 88 90 B8 87
54 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85 86 86 85
50 52 54 56 58 60 62 64 66 68 70 72 74 76 78 80 82 83 83 83
49 50 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 80 81 81
46 47 48 50 52 54 56 58 60 62 64 66 68 70 72 74 76 7B 80 80
41 44 45 47 49 51 53 55 57 59 61 63 65 67 69 71 73 75 77 79
38 41 42 44 46 48 50 52 54 56 58 60 62 64 66 68 70 72 74 76
35 38 39 42 44 45 47 49 51 53 55 57 59 61 63 65 67 69 71 72
32 35 37 39 41 42 44 46 48 50 52 54 56 58 60 62 64 66 68 70
29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67
26 28 30 32 34 36 38 40 42 44 47 49 50 52 54 56 58 60 62 64
25 26 27 29 31 33 35 38 40 42 44 46 47 49 51 53 55 57 59 61
22 23 24 26 28 30 32 35 36 38 41 42 44 46 48 50 52 54 56 58
17 19 21 23 25 27 29 32 34 36 38 40 42 43 45 47 49 51 53 55
14 16 18 20 23 24 27 29 31 33 34 36 39 41 43 44 46 48 50 52
11 13 15 18 21 23 24 25 28 30 31 33 36 38 40 41 43 45 47 49
8 10 12 15 18 20 21 23 24 26 28 31 33 34 37 39 40 43 45 46
9 9 9 11 16 18 17 2 23 23 27 29 29 32 33 36 37 41 44 45

Random entered values = 150


Interpolated values with <10% error= 242
lnterpolat d v lu i 0- O~ error=
Interpolated values with 20r.-5Q% error= 2
Interpolated values with >50r. error = 1

Figure 5.6 A Typical Result from the First 2-Dimensional Test Showing
the Resulting S.L.P.M. Values after 150 RandomValues have Been Entered

78
Equation:

,/
/

OIO~

Figure 5.7 A 3-Dimensional Plot of the Equation Used for the Second Set
of 2-Dimensional Ripple Tests

79
Table 5.2 :- 2-Dimensional Ripple Test 2

Matrix Size :- 20*20 Number of Runs per Test :- 20


Equation : - 40*«sin(xl20*Pi))*(sin(y/20*Pi)))

Test Number of Error in Average Average Average time


Number known interpolated number percentage taken to ripple
values values. ill error ill error all values
range range. (seconds)

<10% 78.10 19.52%


1 20 <20% 70.50 17.62% 14.20
(5%) <50% 136.75 34.19%
>50% 114.65 28.66%

<10% 156.45 39.11%


<20% 83.00 20.75%
2 40 18.60
(10%) <50% 56.30 14.07%
>50% 104.25 26.06%

<10% 205.70 51.41 %


<20% 69.20 17.30%
3 60 20.11
(15%) <50% 38.65 9.66%
>50% 86.45 21.61 %

<10% 246.95 61.74%


<20% 43.75 10.94% 22.30
4 80
(20%) <50% 31.65 7.91%
>50% 77.65 19.41 %

<10% 276.40 69.10%


<20% 33.45 8.36%
5 100 23.65
<50% 29.65 7.41%
(25%)
>50% 60.50 15.21 %

<10% 318.90 72.22%


<20% 18.20 4.55%
150 26.05
6 5.25%
<50% 21.00
(37.5%)
>50% 41.90 10.47%

80
Equa t ion : - (40* ( (s i n (X/20*P i ) ) * i (Y/20*P i ) ) ) )+1
(8 n

6 6 7 8 9 10 11 11 11 10 12 13 13 12 11 10 9 9 9 9
5 6 7 8 9 10 11 12 11 7 12 14 14 13 12 11 9 8 9 9
3 6 8 8 10 11 12 13 13 13 15 16 15 14 13 11 9 5 8 8
7 8 10 11 12 13 14 15 16 17 19 18 17 16 14 12 ,0 8 8 8
9 10 12 15 15 15 16 17 18 19 20 20 19 18 16 14 12 10 9 9
10 11 13 15 16 17 18 19 20 22 23 24 22 20 18 16 13 10 10 10
11 12 4 16 18 19 20 21 22 24 27 32 26 23 20 17 14 11 11 11
12 13 15 18 20 22 23 23 24 25 27 29 27 25 22 19 16 14 12 11
13 14 16 19 23 26 26 25 26 27 28 28 28 28 24 21 18 15 13 12
14 15 17 20 25 33 28 27 27 28 28 28 29 33 26 21 18 15 13 12
14 15 17 20 24 27 27 27 28 29 29 28 28 7 24 20 15 13 12
14 15 17 20 22 24 26 27 30 32 30 28 26 25 22 19 16 14 12 10
14 15 17 19 21 23 25 27 31 39 31 27 25 23 21 18 15 13 10 7
13 14 16 18 20 22 24 25 28 30 28 25 23 21 19 17 5 12 8 1
12 13 15 18 20 21 22 23 25 25 24 23 21 20 18 6 4 12 10 7
11 10 14 17 21 21 21 21 22 22 21 20 19 18 17 15 14 13 11 10
10 11 13 16 18 20 19 19 19 19 19 18 8 7 16 14 4 13 12 11
9 10 12 14 15 16 16 16 16 16 16 16 16 16 15 12 13 13 12 12
6 7 10 12 13 14 14 14 14 14 14 14 14 14 14 13 13 13 12 12
2 3 8 11 12 13 13 13 13 13 13 13 13 13 13 13 13 13 12 12

Random entered values = 20


Interpolated values with <10% error= 51
In rpolated valu s i h 0- r. er r= 100
Interpolated values with 20X-50X error= 119
Interpolated values with >50i. error = 110

Figure 5.8 A Typical Result from the Second 2-Dimensional Test Showing
the Resulting S.L.P .M. Values after 20 Random Values have Been Entered

81
Equa t ion : - (40* ( (5 j n (Xl20*P i ) ) * i n(Y120*P i ) ) ) )+1
(5

9 8 6 6 5 1 4 1 6 8 6 1 4 1 8 11 11 11 10 8
9 8 4 7 8 8 9 9 10 11 10 8 9 9 11 12 12 11 9 6
9 9 8 10 12 13 14 14 14 13 13 13 13 13 13 12 9 1
10 10 9 12 13 15 17 18 18 18 17 17 17 16 14 14 13 11 8
11 11 12 14 16 18 20 23 22 ~2 21 21 20 19 17 16 15 13 12 11
11 12 14 6 18 20 22 24 5 25 25 25 24 24 20 18 15 10 12 12
11 13 16 19 20 22 25 27 29 29 29 29 27 25 22 19 16 11 12 12
10 13 17 22 22 24 27 31 36 34 34 35 31 28 25 21 17 I 13 12
7 13 17 21 23 26 29 33 39 37 36 35 33 32 28 22 18 13 12
11 14 18 21 24 27 30 34 37 38 38 37 34 32 27 22 18 5 13 12
11 15 19 22 26 29 31 34 37 41 41 39 35 33 27 ~ 18 1~ 13 12
7 14 19 23 29 31 32 3 37 39 38 36 33 30 ~ 22 18 1~ 13 12
12 15 18 22 27 32 33 34 36 39 36 34 32 29 21 17 13 13 12
13 15 18 21 25 29 33 33 34 37 34 33 33 28 24 20 16 13 12 12
12 13 16 19 23 27 30 30 31 32 31 30 28 25 22 19 15 11 11 11
10 10 14 17 21 24 25 26 27 28 29 28 25 22 20 18 14 10 9 9
8 10 12 15 18 20 21 22 23 24 24 23 21 19 18 15 12 9 5 7
4 8 1U 12 14 16 17 18 19 19 19 18 17 16 15 12 0 8 7 7
7 8 9 9 10 12 13 15 16 16 15 14 4 14 13 11 8 5 6 7
8 8 7 5 5 8 7 12 14 14 12 7 11 12 12 10 7 3 5 6

Random entered values = 60


Interpolated values with <10r. error= 157
Int rpola 1 e wi 1G- % :
Interpolated values with 20r.-50r. error: 21
Interpolated values with >50r. error: 94

Figure 5.9 A Typical Result from the Second 2-Dimensional Test


Showing the Resulting S.L.P.M. Values after 60 Random Values have Been
Entered

82
Equa t ion : -- (40* ( (s i n (X/20*P i ) ) * (s i n (Y 120*P i ) ) ) ) +1

2 1 465 1 44 1 44 1 44 1 1 33 2 2
2 3 6 8 8 6 7 8 7 8 7 7 7 7 5 4 4 2 1·
3 5 7 10 11 11 12 13 13 13 13 13 12 11 10 7 4 3
5 7 9 12 14 15 16 17 18 19 19 18 17 16 14 12 7 5 3
G 8 12 15 18 20 20 21 22 23 24 23 22 20 17 15 12 8 5 1
5 10 14 18 21 24 24 5 2 27 29 28 26 24 20 17 14 10 5 4
6 12 16 20 23 26 27 29 30 31 33 32 30 27 23 20 16 11 6 6
11 14 18 22 25 28 30 33 34 35 35 35 33 30 26 22 17 12 9 7
12 14 18 23 28 31 33 37 39 39 38 37 35 31 28 23 18 14 10 6
11 13 19 24 29 33 35 38 40 41 39 37 35 32 29 24 19 14 9 1
7 13 19 25 29 33 35 38 41 41 39 37 35 33 29 24 19 13 10 6
10 13 19 24 28 32 35 37 39 40 40 37 36 32 29 24 19 14 10 8
11 13 18 23 27 31 35 36 37 39 39 36 34 31 28 23 18 13 7 8
10 13 17 22 26 30 33 34 35 37 36 35 33 30 26 22 17 13 9 8
6 11 16 20 24 27 29 31 32 33 32 31 30 27 24 20 16 11 6 6
7 10 14 18 21 24 26 28 29 29 28 28 26 23 20 17 14 10 5 5
5 8 12 15 18 20 22 23 24 24 24 23 22 20 17 15 12 5 4
5 7 9 12 14 16 17 19 19 19 19 18 17 16 14 12 8 5 1
3 6 9 10 12 14 15 14 14 13 13 13 11 10 8 7 5 3
2 5 6 5 5 6 10 11 7 9 7 7 9 6 5 6 4 3 4 4

Random entered values = 150


Interpolated values with <107. error= 169
I nt rpo 1 e · - 1"= ?

Interpolated values with 207.-507. error= 19


Interpolated values with >507. error = 41

Figure 5.10 A Typical Result from the Second 2-Dimensional Test Showing
the Resulting S.L.P.M. Values after 150 Random Values have Been Entered

83
Figure 5.11 shows the last two dimensional surface tested, and as it is a fairly complex
one. The results in Table 5.3 and Figures 5.12 - 5.14 reflect this fact, with the accuracy
of the prediction method dropping dramatically. Only when the S.L.P.M. is 37.5%
full are the predictions acceptable i.e. 60% are within tolerance. This implies that for
more complex surfaces, other methods of interpolating must be used in conjunction or
instead of this method.

5.4.2 Three Dimensional Method

Table 5.4 gives the results of interpolation of a three dimensional simple equation. As
with the two dimensional simple equation, a good correlation between predicted and
actual values is seen (with the S.L.P.M. 20% full, 90% of the data is ~10%). The time
taken however, has increased dramatically, being some 6-7 time greater than for the
two dimensional ripple.

The results of a complex three dimensional surface ripple are shown in Table 5.5. It can
be seen that for this type of surface, other methods must be used such as curve fitting in
conjunction with rippling, as the prediction method is inaccurate i.e with the S.L.P.M.
30% full, only 46.55% is with in the tolerance range.

5.5 Conclusion

In concluding this chapter, wear is a complex subject with many different parameters
effecting the type and extent of wear. It is also a difficult parameter to measure, both
physically and practically. Therefore, the prediction of likely scenarios is required to aid
drilling performance. There are many equations which have been produced calculating
wear rates etc, some in every day use. However in using these equations,
improvements to prediction will never occur unless the equation is re-calibrated with
field results.

The wear predictor used for the drill optimisation system was designed to eliminate this
problem, and to learn the wear process, therefore continually improving its prediction
perfonnance. To do this data was stored in a matrix fonnat (named Self Learning
Prediction Matrices), in which wear or penetration rates could be referenced by specific
parameters. As wear results are hard to generate, on each addition of a new value. the
S.L.P.M. would be interpolated to estimate unknown data points. The interpolation
system used the "ripple" technique, which perfonns an averaging process within the
S.L.P.M. The test results of the predictor are encouraging, especially for

84
Equation:

0,0

Figure 5.11 A 3 -Dimensional Plot of the Equation Used for the Third Set
of 2-Dimensional Ripple Tests

85
Table 5.3· .2-Dimensional Ripple Test 3

Matrix Size:- 20*20 Number of Runs per Test: 20


Equation : - 20*((sin(x!7*Pi))*(sin(y/12*Pi)))

Test Number of Error in Average Average Average time


Number known interpolated number percentage taken to ripple
values values. ill error ill error all values
range range. (seconds)

<10% 89.05 22.26%


20 <20% 16.50 4.13% 5.00
1
(5%) <50% 85.45 21.36%
>50% 209.00 52.25%

<10% 115.70 28.92%


<20% 31.15 7.95%
2 40 7.95
<50% 117.90 29.47%
(10%)
>50% 135.25 33.81 %

<10% 135.55 33.89%

60 <20% 39.20 9.80%


3 10.30
(15%) <50% 150.60 37.65%
>50% 74.65 18.60%

<10% 162.55 40.64%


<20% 53.20 13.30% 11.95
4 80
(20%) <50% 137.00 34.25%
>50% 47.25 11.81 %

<10% 183.60 45.90%


<20% 64.60 16.15%
5 100 13.45
(25%) <50% 126.75 31.65%
>50% 25.05 6.29%

<10% 238.80 59.70%


<20% 77.30 19.32%
16.70
6 150 18.14%
(37.5%) <50% 72.55
>50% 11.35 2.84%

86
-6 -5 -4 -3 -2 -1 0 1 2 3 4 4 3 1 -2 -6-1 0- 4-13-12
-6 -6 -5 -4 -3 -2 0 1 2 3 4 5 5 2 -2 -6-11 -19-14-12
-7 -7 -7 -6 -4 -2 0 2 3 4 5 6 9 4 -1 -6-10-13-12 -9
-8 -9-10 -8 -5 -2 1 4 5 5 6 6 6 3 -1 -6-11-11-10 -
-9-11 -17-10 -6 -3 2 8 6 6 6 6 5 3 -1 -7-17-11 -9 -8
-8 -9-10 -8 -6 -6 0 4 5 6 7 6 5 3 o -4 -8 -8 -7 -6
-5 -6 -6 -6 -5 -4 -1 2 4 6 10 6 4 3 1 -2 -4 -5 -5 -4
-3 -3 -3 -3 -3 -2 -1 1 2 4 5 4 3 3 2 o -2 -3 -3 -2
0 o -1 -1 -1 -1 0 1 0 2 3 3 2 2 2 1 -1 -1 -1 -1
3 4 2 1 1 0 0 1 1 1 1 1 1 1 1 1 0 0 0 0
6 6 5 3 2 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
8 11 7 5 4 2 o -1 -1 -1 -1 -1 -1 -1 -1 0 0 0 0 0
7 8 7 6 5 3 o -2 -3 -3 -3 -3 -3 -2 -1 0 0 0 0 0
6 7 7 8 8 4 o -3 -5 -6 -6 -5 -4 -3 -1 0 0 0 0 0
5 6 7 10 16 5 -1 -5 -7 -9-11 -8 -5 -3 -1 0 0 0 0 0
4 5 6 7 7 2 -3 -8 -8-11 -19-10 -6 -3 -1 0 0 0 0 0
3 4 4 4 3 o -3 -6 -7 -9-11 -8 -5 -3 -1 0 0 0 0 0
2 2 2 2 2 -1 -3 -5 -6 -7 -7 -6 -4 -3 -1 0 0 a a 0
2 2 2 2 2 -1 -3 -4 -5 -5 -5 -4 -3 -2 -1 a 0 0 0 a
1 1 1 2 4 a -2 -3 -4 -4 -4 -3 -2 -2 -1 a 0 0 0 a

Random entered values = 20


Interpo1ated values with <10r. error= 67
In ~
Interpolated values with 20r.-50r. error= 69
Interpolated values with >50r. error = 234

Figure 5.12 A Typical Result from the Third 2-Dimensional Test Showing
the Resulting S.L.P.M. Values after 20 Random Values have Been Entered

87
Equat i on : - 20*<sin(x/7*pi )*sfn(y/12*pl))

-10-12-13-13-12 -8 -3 1 4 4 31 0 -2 -5 -8-10-11 -12-13


-9-12-14-15-15 -9 -3 3 7 5 31 0 -2 -6-10-11-12- 3-13
-9-13-16-19- 4 -9 -2 6 16 7 31 0 0 -7-1&-13-13-16-13
-8-15-17-19-13 -8 -2 5 8 5 20 -1 -3 -6-10-11-11 -1 -12
-11 -14-17-15-11 -8 -1 8 5 3 -'1 -1 -2 -3 -5 -8 -9-10-10-10
-10-11-12-11 -9 -6 0 2 l ' 0 -1 -2 -2 -3 -4 -6 -8 -9 - -8
-8 -9 -9 -8 -8 -6 -3 -2 -2 -2 -2 -2 -2 -2 -3 -4 -6-10 -7 -6
-5 -6 -6 -5 -5 -5 -4 -3 -3 -3 -2 -1 -1 -1 -1 -1 -2 -4 -3 -3
o -2 -2 0 -3 -4 -4 -3 -4 -4 -3 0 0 0 1 2 2 1 1 0
2 1 1 1 -1 -2 -3 -2 -5 -6 -6 -4 -2 0 2 4 5 5 4 3
4 4 4 4 2 1 -2 -4 -7-10-10 -7 -4 -1 3 8 9 9 8 4
6 7 9 7 6 6 -1 -5 -8-11-12-10 -6 -2 3 8 14 14 11 6
8 10 17 10 7 5 -1 -8-10-13-17-14 -8 -2 3 7 10 11 10 8
8 9 11 9 7 8 -1 -7-10-12-14-15 -8 -2 3 7 9 10 9 8
8 8 9 8 6 4 -2 -9-10-11-12-11 -7 -1 5 9 9 9 9 8
9 9 9 7 5 3 0 -6 -9-11-13-10 -6 0 8 15 10 9 9 B
10 10 10 8 5 2 -1 -5 -8-12-17-10 -5 0 5 9 9 9 9 8
10 11 14 10 S 2 -1 -4 -7-10-11 -8 -4 0 3 6 7 8 8 8
10 10 11 10 6 2 -1 -4 -7-10 -8 -6 -4 -1 2 5 6 7 8 8
10 9 9 8 5 2 0 -3 -5 -5 -5 -5 -3 -1 2 4 5 6 7 8

Random entered values = 60


Interpolated values with <10% error= 80
Int rpolated v ues i 10-2 ~ rr r= 34
Interpolated values with 20%-50% error= 116
Interpolated values with >50% error = 110

Figure 5.13 A Typical Result from the Third 2-Dimensional Test Showing
the Resulting S.L.P.M. Values after 60 Random Values have Been Entered

88
-14-15-16-15-14 -7 0 7 14 16 15 14 8 0 -8-11-1 -17-14 -8
-13-15-19-16-13 -8 0 8 15 19 15 12 7 0 -7- -15- 6-15-12
-9- 4-17-16-13 -7 1 9 14 1 14 12 9 0 -9-13-19-18-16-13
-11-14-17-19-15 -7 1 8 15 15 14 11 8 1 -6- 2-19-19-15-12
-11 -13-17-17-12 -8 0 6 12 15 17 11 6 1 -5-10- 5-17-13 -8
-9-11 -14-12 -9 -5 0 5 11 14 13 9 - 0 -6 - 2-14-11 -8
-4 -7 -9 -8 -8 -4 0 4 7 10 10 8 4 0 -4 -,-10-10 -, -4
-2 -4 -5 -4 -4 -2 0 2 4 5 5 4 2 0 -2 -3 - -4 -4 -3
1 00 0 000 00 1 0 1 1 00 1 1 1 0 0
5 5 4 4 4 2 0 -2 -3 -3 -3 -2 -1 0 2 4 5 5 4 2
8 8 8 8 8 4 0 -4 -8 -7 -7 -6 -3 0 4 8 9 9 8 5
11 11 12 11 11 6 0 -6 -9-11 -14-10 -6 0 6 11 14 2 11 6
13 14 17 4 11 . 8 1 -6-10- 4-17-14 -8 0 8 13 17 11:"' 13 10
14 15 19 16 12 7 0 -8- 2-19-19-13 -7 0 8 15 19 19 15 11
15 16 18 19 12 6 0 -6-11-15-15-11 -6 0 9 14 18 19 16 9
14 15 19 11 6 0 -6-11-14-14-11 -6 0 7 14 19 17 15 8
13 14 5 13 10 8 0 -7-14-15-17-14 -8 -1 6 14 15 14 13 10
11 12 14 11 8 5 -1 -6-11 -14-13-11 -6 -1 5 10 12 12 11 10
9 9 9 8 6 3 -1 -4 -8-10-10 -8 -4 0 4 8 10 10 8 8
8 7 5 5 4 2 -1 -4 -6 -7 -5 -5 -2 0 4 7 8 7 4 6

Random entered values = 150


Interpolated values with <10% error= 86
Interpol d val with 1 - ~ p ro = 7
Interpolated values with 20%-50% error= 63
Interpolated values with >50% error = 28

Figure 5.14 A Typical Result from the Third 2-Dimensional Test Showing
the Resulting S.L.P.M. Values after 150 Random Values have Been Entered

89
Table 5.4:- 3-Dimensional Ripple Test 1

Matrix Size :- 10*10*10 Number of Runs per Test 20


Equation :- 2*X + 3*Y.+2*Z

Test Number of Error in Average Average Average time


Number known interpolated number percentage taken to ripple
values values. ill error ill error all values
range range. (seconds)

<10% 567.90 56.79%


<20% 311.05 31.10% 167.30
1 50
(5%) <50% 98.45 9.84%
>50% 22.60 2.26%

<10% 793.05 79.30%


<20% 137.45 13.74%
2 100 196.40
<50% 53.74 5.38%
(10%)
>50% 15.75 1.57%

<10% 907.65 90.76%


<20% 63.10 6.31%
3 200 231.40
(20%) <50% 25.35 2.53%
>50% 3.90 0.39%

90
Table 5.5 :- 3-Dimensional Ripple Test 2

Matrix Size :- 10*10*10 Numberof Runs per Test 20


Equation : - 80*(sin(x/10*pi)*sin(y/10*pi)*sin(z/10*pi»

Test Number of Error in Average Average Average time


Number known interpolated number percentage taken to ripple
values values. ill error ill error all values
range range. (seconds)

<10% 137.80 13.78%


<20% 99.10 9.91%
1 50 126.65
(5%) <50% 311.85 31.18%
>50% 451.25 45.12%

<10% 202.90 20.29%


<20% 125.95 12.59%
2 100 182.35
<50% 270.70 27.07%
(10%)
>50% 400.45 40.04%

<10% 342.35 14.23%


174.60 17.46%
<20% 241.90
3 200
<50% 180.15 18.01 %
(20%)
>50% 302.90 30.29%

<10% 465.50 46.55%


<20% 182.60 18.26% 267.20
4 300
<50% 120.95 12.09%
(30%)
>50% 230.95 23.09%

91
simple functions. However as the functions become more complex, the accuracy of
prediction drops. This indicated that some sort of polynomial curve fitting technique
should also be developed to aid the rippling technique. In view that the predictor
worked fairly well on moderately complex surfaces, work on the polynomial fitter was
left, to return to once the initial optimisation system had be developed and was running.

92
Chapter 6 - The Control Algorithm

6.1 Introduction

In Chapter 1, an introduction was given to drilling optimisation and an indication of the


requirements for the development of such a system. This concept was shown by
Figure 1.2, where drilling parameters were fed to an Intelligent Knowledge Induction
System, which was used to model the current drilling environment. It then selected and
manipulated those parameters required to bring about an improvement to the current
situation.

To enable an optimisation system to make such judgements, a decision has to be made


on what is the requirement of the optimisation system, and through which controlling
parameter or parameters, this can be achieved. Its relationship with other drilling
parameters must also be considered, to ensure that they do not conflict with the overall
optimisation scheme. This selection process was developed in Chapter 4. The
optimisation criteria had previously been defined as achieving optimum operating
perfonnance through the trade off between penetration rates and wear rates. Many
drilling parameters were proposed for the controlling parameter, but most were
rejected. Cost per metre was selected, as it enabled both a relatively simple and flexible
system to be developed. A simple cost equation was used incorporating running costs
such as bit costs and rig charges. However, as this equation contained no parameters
directly relating the drilling operation, it was manipulated to produce such an equation,
with parameters more specific to the drilling operation. This equation 4.27 (shown
below) would fonn the basis for the optimisation scheme.

(B + R. Tm . D) . W R
C - ----------"K,..:.:..:....---- + P ---- ( 6.1)

To use this equation however, data must be readily available for both penetration rates
and wear rates. While penetration rates may generally be measured on line, wear rates
are difficult to measure and often require extensive laboratory testing to obtain. A
method for storing and enhancing known data was described in Chapter 5, where the
concept of S.L.P.M.'s was developed and used to progressively learn and predict a
process such as bit wear. Utilising these S.L.P.M.'s in the cost optimisation scheme,
enables reliable and readily available data to prime the cost optimisation equation.

93
However, how is this data and the cost optimisation equation used, and by what
process can the minimum cost operating point be determined? Once the optimum point
has been established, can the system cope with a changing environment such as
changing lithology? The answers to these questions forms the basis for this chapter
and it describes the various methods which have been developed to locate the minimum
cost operating point and overcome problems likely to be encountered by the
optimisation system.

6.2 Establishing the Minimum Cost Operating Point

6.2.1 Maxima and Minima Theory.

In any process requiring either a maximum or minimum value to be found, one of the
most simple and direct methods is that of maxima and minima theory. This states that
when the fIrst derivative is equal to zero, then a maximum, minimum or inflection is
found. By substituting values back into the equation, the type of feature found may be
established.

To apply this, the fIrst derivative of equation 6.1 must be found. However it can be
seen that there are two independent variables with respect to cost, that of penetration
rate and wear rate. Therefore, substitution is required to eliminate one of them, i.e. a
relationship between wear rate and penetration rate must be found such that :-

W = f(P) --- (6.2)

Substituting this into equation 6.1 ,gives

(B + R . T m . D) . f(P) R
C = K + P --- (6.3)

and differentiating gives,

dC B + (R . T m . D) . rep) R
--- (6.4)
dP= K p2

To fInd the minimum, the equation is equated with zero, such that,

B + (R . T m . D) . f'(P) R
0= K p2 ---- (6.5)

94
and thus

f(P) . p2 - (B + : .' T~ . D) --- (6.6)

However, to be able to solve this, the relationship W-f (P) must be found.

The work reported by Ambrose, gave a series of relationships between varying


rotational speeds and weight on bit values, with corresponding penetration rates and
wear rates. For each rock type tested, values were entered into a database, allowing
retrieval and also subsequent addition in the event of new values being generated. A
curve fitting routine utilising least squares method was used on each set of data (i.e. for
each different rock type) to detennine a polynomial equation. The best fit was
detennined by visual observation. For simplicity the equations were kept to forth order
or below. The resulting polynomial equation 6.7, thus gave the desired relationship
between wear rates and penetration rates.

W= f (P) = E . p4 + D. p3 + C. p2 + B . P + A ---(6.7)

where A,B,C,D,E are constants


of the polynomial equation.

The resulting polynomial can be readily differentiated such that,

f(P) = 4. E. p3 + 3. D. p2 + 2. C .P + B ---(6.8)

Substituting this back into equation 6.7, gives

(4.E. p3 + 3. D. p2 + 2. C .P + B). p2 _ (B + RR . . TKm . D) ---(6.9)

Hence by solving this, a minimum can be detennined. This can be achieved using an
iterative technique such as Newton Raphson.

The method described, formed the bases of a computer programme, which was
developed at an early stage in the research project. It was developed to test the idea of

95
cost optimisation and to learn and generate ideas which would aid the design of the
main cost optimisation system.

From the work performed by Ambrose, four rock types had been tested, and thus
penetration rates and associated wear rates from this work, were used to generate the
polynomial relationship described. However, it should be noted that these relationships
only hold for the data shown. They are not general rules or laws, but are solely used to
show how such a relationship could be developed with progressive testing, and how
these relationships may be used for minimum cost prediction.

The relationships developed are shown in Figures 6.1 - 6.4. Using these relationships,
tables could be generated giving the cost per metre for differing values of the
parameters in the cost equation, as shown in Table 6.1. It can be seen from this table
that the penetration rates have been evenly space, having ten values ranging from
maximum penetration rate to zero. The wear rates have been derived from the
polynomial expression and these penetration rate values. From the minimum cost
values for each depth (highlighted in the table), it can be seen that in certain cases as
depth increases, the required penetration rate and wear rate change.

By utilising the principle of maxima and minima, and the substitution via the
polynomial relationship, all previously described, the computer programme can be
made to determine the best wear rates and penetration rates for a particular drill rig
scenario, at any depth. Furthermore, as polynomial relationships have been derived for
four rock types, the programme was developed to allow differing combinations and
thickness of the rocks, hence giving self-designer bore holes.

Figure 6.5 shows the results of such a test on a hypothetical hole with the same rig
conditions as shown in Table 6.1. It can be seen that on several occasions, changes of
penetration rate and wear rate have been made within a particular lithology. Tripping
depths and the cost per metre at the start and finish of each horizon are shown.
Therefore by using this format, the proposed hole could be drilled at minimum cost.

However, while this method does give an easy solution, it should be remembered that
penetration rates and wear rates are not truly interdependent, but there are many other
variables which effect either one or both. For this reason, the assumption that W = f(P)
is invalid, and therefore this method was not persued any further. This method also has
another disadvantage when considering development of a versatile optimisation
system. To enable differentiation, the exact process must be known, either by a

96
Rock Type :- Gniess

Penetration Rate Wear Rate


(mm/min) (urn / m)

50 400
60 300
70 200
90 185
100 150
135 200

500

..- 400
-€
E
::I
'-"
~
~
~
300
~
~
200

100~~~----~------------~------

40 60 80 100 120 140

Pen. Rate (mm/min)

y = 1003.3038 . 16.2692x + 0.0767x"2 R = 0.97

Figure 6.1 Wear Rates Vs Penetration Rate Data, Graph and Polynomial Fit for
Gniess

97
Rock Type:- Sandstone

Penetration Rate Wear Rate


(mm/min) (um/m)

50 180
100 150
150 110
200 100
250 95
350 90

200

180

~
160
-€
E
='
'-"
140
£
~
ez:::
120
~
:::
100

80
0 100 200 300 400

Pen. Rate (mm/min)

y = 222.6429 - 0.9324x + 0.0016x"2 R = 0.99

Figure 6.2 Wear Rates Vs Penetration Rate Data, Graph and Polynomial Fit for
Sandstone.

98
Rock Type :- Limestone

Penetration Rate Wear Rate


(mm/min) (urn /m)

20 300
30 140
40 80
70 25
90 30
100 90
110 190

400

--
.§.
E
300

::::2
"--'
.....
(]) 200
~
~
@
(])

~ 100

O~--~--~--~--------~------~
60 80 100 120
o 20 40

Pen. Rate (mm/min)

y = 1017.9478 - 55.8907x + 1.2155x"2 - 0.012x"3 + 0.000e+Ox"4 R = 1.00

Figure 6.3 Wear Rates Vs Penetration Rate Data, Graph and Polynomial Fit for
Limestone

99
Rock Type :- Fine Grained Sandstone

Penetration Rate Wear Rate


(mm/min) (um/m)

100 40
150 41
200 42
300 45
350 50
380 70

80

---
E
E
70

:::l
'-'
60
....
Cl)

""
c.:::
~
Cl) 50
~

40

30~--------~----------~------

o 100 200 300 400

Pen. Rate (mm/min)

y = 59.9214 - 0.2385x + 6.555e-4x"2 R = 0.91

Figure 6.4 Wear Rates Vs Penetration Rate Data, Graph and Polynomial Fit for
Fine Grained Sandstone.

100
Bit Costs=£1500
Rig Cost Per Day =£5000

Average Round Trip Time Per Metre = 20 seconds

Rock Type :- Gnlea

Depth (Metres)
Penetration Wear Rate
Rate (mm/s) (mm/m) 10 100 500 1000 1500

2.250 0.205 56.71 58.84 68.33 80.20 92.06


2.025 0.159 §2.U U.l'7 61.63 70.83 80.03
1.800 0.141 53.46 54.93 U.U n.n '7 '7 • '78
1.575 0.151 59.57 61.14 68.13 76.87 85.61
1.350 0.189 71.44 73.40 82.15 93.09 104.03
1.125 0.255 85.45 87.79 98.21 111.23 124.25
0.900 0.349 117.05 120.69 136.85 157.04 177.24
0.675 0.470 156.78 161.67 183.43 210.63 237.83
0.425 0.620 229.88 263.34 265.05 300.92 336.80
0.225 0.798 377.83 386.14 423.08 469.26 515.44

Rock Type :- Sandstone

Depth (Metres)
Penetration Wear Rate
Rate (mm/s) (mm/m) 10 100 500 1000 1500

5.833 0.092 H.I] l~.n It.~§ 34.37 39.69


5.250 0.087 24.17 25.08 29.11 H.U n.n
4.667 0.087 25.55 26.46 30.48 35.52 40.55
4.083 0.090 27.78 28.72 32.88 38.09 43.30
3.500 0.097 31.20 32.21 36.70 42.31 47.92
2.917 0.108 36.16 37.29 42.29 48.54 54.79
2.333 0.123 43.40 44.68 50.37 57.49 64.61
1.750 0.142 54.53 56.01 62.59 70.80 79.02
1.167 0.165 74.53 76.25 83.89 93.44 102.98
0.583 0.192 128.29 130.29 139.17 50.29 161.40

Rock Type:- Limestone

Depth (Metres)
Penetration Wear Rate
Rate (mm/s) (mm/m) 10 100 500 1000 1500

1.833 0.135 51.98 53.38 59.63 67.45 75.26


1.650 0.070 45.46 46.38 49.62 53.67 57.73
1.467 0.040 ~§.U ~§.~1I. ~'7.'7' §~.n §l.n
1.283 0.027 49.19 49.47 50.72 52.28 53.84
1.110 0.022 55.46 55.69 56.71 57.98 59.25
0.917 0.027 67.19 67.47 68.72 70.28 71.85
0.733 0.054 87.l1 87.68 90.18 93.30 96.43
0.550 0.121 123.51 124.77 130.37 137.37 144.38
0.367 0.260 196.99 199.69 211.73 226.78 241.82
0.183 0.511 393.47 398.80 422.45 452.03 481.60

Rock Type :- Fine Grained Sandstone

Depth (Metres)
Penetration Wear Rate
Rate (mm/s) (mm/m) 10 100 500 1000 1500

6.333 0.064 18.81 19.48 22.44 26.15 29.85


5.700 0.055 U.~'7 U.U 21.59 24.77 27.95
5.067 0.048 18.68 19.18 2lI. .~~ l~.n l'.~§

4.433 0.043 19.55 20.00 21.99 24.48 26.97


3.800 0.040 21.28 21.69 23.54 25.86 28.17
0.Q38 24.02 24.41 26.17 28.37 30.57
3.167
0.039 28.74 29.15 30.95 33.21 35.47
2.533
0.041 36.66 37.08 38.98 41.35 43.73
1.900
0.046 52.63 53.l1 55.24 57.90 60.56
1.267
0.052 99.28 99.82 102.23 105.24 108.25
0.633

Table 6.1 The Cost per Metre for Varying Penetration and Wear Rates.
for the Rig Parameters and Rock Types Shown

101
Depth Crown Cost per Penetration Wear Rate
Hieght Comments Rate
(Metres) Metre (mm/m)
(mm) (mm/s)

o 10.00
£23.73
5.833 0.092

£24.25
50 5.4
----------_._.- £18.72
---------------
---------------
---------------
:fue Gia1ieii-:: 5.700 0.055
-:Sandstone --
_._------------
--_._----------
-----------_ ....
--------------- £19.17
--------------- 120 1.55
£46.00

159 10.00 Trip


1.467 0.040

£46.47
----_ ... -------- 220 7.56
_.. ------_._ .. - £19.80
_--_._--------
..-_._ .. _---------
-_ .... _----------
---------------
..__------------- 5.700 0.055
.. _----------_ ..
-.Fine Grained .:_ 270 4.81
Penetration
Rate Change
.
-.Sandstone £20.12
_.. __ .. _.. _.. _-_ .. _-
---------------
---_
----_
.. __ .. ---_ ......
.... _-------
--_ .. _-----_
-------_ .. ------
.... _- 5.067 0.048
--_ .. _---------- £20.57
---------------
--_ .... _--_ .. __ .. _-
350 0.97
",, ",, ",, ",, ",, ",, ",, " £58.89
""""""""
",, ",, ",, ",, ",, ",, ",, " 356 10.00 Trip 2.025 0.159
", ", ", ", ", ", ", "
", ", ", ", ", ", ", "
", ," "' '." "' "" ," ," 415 0.62 Penetration £60.07
, . . Gruess , , Rate Change
", ", ., .,.,., ", "
",.... "........ ",.... ",, ".... ",.... ",, " 420 10.00
"""""""" ;
1.800 0.141
", ";; ", ", ", "; " "; ; ; ;

", " ", ", ", ", ", "


; ; ; 492 10.00
"""""""" ;

" "; " "; ", " " "


; .... ; ;
500 8.87
£61.46
.•.. ::. :<::::.
£29.05
::\/
596 10.00 Trip
5.833 0.092

610 8.712 Penetration £30.32


Rate Change
5.250 0.087
£31.12
700 0.88
", "; "; ", ", ", ", "
", ", "; ", ", ", ", "
£64 .72
706 10.00 Trip
", ".... " "; ", " "; "
; ;
1.800 0.141
""""""""
/ / ~.~ ~ / /
" ; " . . Gruess·
... "; " ;

", ", ", ", ", ", ", " 777 10.00 Trip
", ";; ", "; ",, ",, ",, "
; ;

"""""""" ;

", ", ", ", ", ", ", " 6.76


£66.36
"" """ 800 £23.07

-. Fine Grained .:- 10.00 Trip 5.067 0.048


941
-·Sandstone
£24.18
1000 7.168

Figure 6.5 The Minimum Cost Results Obtained for Maxima and Minima
Cost Optimisation

102
pre-defined equation or by data to which one can be fitted. Therefore, by using this
method, an optimisation system could never start form a null state, thus would require
test programmes from which the relevant data could be learnt. In an optimisation
process, this is not desirable.

Despite these limitations however, the method of substitution and applying maxima and
minima theory, is extremely useful and could be applied to aid the search algorithm for
locating areas of minimum cost, especially when multi-slope surfaces are seen. This
will be expanded later in Chapter 9.

6.2.2 Partial Differentiation

As the cost optimisation equation has two independent variables i.e. penetration rate
and wear rate, it is possible to partially differentiate the equation with respect to each
variable. The resulting equations may be solved in a similar way to find a minimum as
that described previously.

The two partial differential equations of equation 6.1 are shown below.

oC (B + R . Tm . D) oW R
---- (6.10)
dP - K . dP p2

oC (B + R . Tm . D) R oP
dW ---- (6.11)
dW - K p2

A solution to these equations was sought, but none were found due to the large number
of unknowns which make any solutions complex if not impossible.

6.2.3 Computer Search Methods

The two methods described above provide a direct method with which to locate a
maximum or minimum position. However both these methods were ruled out for use in
drill optimisation. Therefore a different way in which to use this equation to achieve
optimisation must be found.

Developing technology has produced computers with ever increasing memory capacity
and more importantly increased processor speed. This allows a large number of
calculations to be performed in a short space of time. Consequently it is now possible

103
to develop computer search methods which can cope with the large numbers of
repetitive calculations required to find the optimum value or position.

Applying this to a drill optimisation scheme, associated combinations of penetration


rates and wear rates could be fed through the cost equation and the combination
yielding the minimum cost, used to set the drill parameters. However to do such
calculations, the system requires the ready access of both penetration rate and
associated wear rate data. Such a storage method was described in Chapter 5, where
penetration rates and wear rates were stored by set reference parameters in S.L.P.M.'s.
By accessing parameters common to both, a means of inter-relating the two can be
established.

Therefore, the various com binations of wear rates and penetration rates can be fed
through the cost equation to yield a minimum cost. Furthermore, as drilling progresses,
the two S.L.P.M.'s will progressively learn their respective process and hence enhance
the prediction of the minimum cost position. This process is shown in Figure 6.6.

However, by using this method, it has to be remembered that the process of wear rate
and lor penetration rate may not be fully understood ie the S.L.P.M.s may only be
partially full. Furthermore, if the system is starting from the null state, then they will
contain no information. Therefore, as the computer searching method can only predict
the minimum cost based on the information it contains, and hence"" the predicted
minimum cost per metre may not necessarily be the true minimum cost per metre. This
may be hidden in an area where information is sparse or unknown, and consequently
not revealed by the S.L.P.M.'s interpolation method.

The control algorithm therefore, also has to be able to search for the true minimum cost
position starting from that given by the predictors. This may be achieved by
manipulating the drill's parameters and monitoring the response.

6.2.3.1 The Search Algorithms

The search algorithm has to be able to manipulate the drill parameters in such a way as
to bring about an improvement to the current situation. Several methods were tried,
each having a varying degrees of success. The results of the search methods are
contained in the next chapter, but the theory of each is described below.

104
:
:
:::::
..

0 rl-I I
:
. ., ..
: . .: : ·. ·.
( Drill " 10::::::::::::::::::::

........
·. .:.~

.. : ·.
~
:
.. :
.. .
· .
. . .. .
·· ..
. . . Data)
: ·. ·. · .
: .. · . ·.. ... ..
: ·. ·. ~
.. . ·. .. . . ·· .. ·· .. ~
..
. . . . . . . ·· .. · . '- -"

Database

( control]
\. Data /

:::::::::.,.::: ::::::::::::::::::::,.::::::,.; ,>:,:,,::::::::::::::::


::::::
"
::::::::::::::
"
:;:;;::.;:;~:T:::::::::::::::::/::::."::::::::::::::::::::::::::::;:::::::::.
::I::::
:::
11!!111:1 Optimisation Prediction ·ft
Algorithm ...... Mechanism
(Cost Equation) (S.l.P .M.'s) f
:1:,' 1
:::::::{7:: .:.,.: }ff:·:.,::)t:ff::·:· :·:·:· ~:.:.:.:
:::.:::::::::::::::.,.:::::::::::: .:::::: .::,.:::::::::I!::·:

Figure 6.6 A Simplified Control Scheme with the Prediction Mechanism


Included

105
6.2.3.1.1 Vector Method

This method was designed as an initial starting process, which would determine the
best direction in which to manipulate the drilling parameters to bring about the greatest
improvement over the current situation. From the selected starting point, four data
points are established by manipulating the required parameters in a cross format and
their differences compared to the central point established. The diagonal values are
detennined by summating the two adjacent cross values, if the point is an unknown. If
the vector of this value falls between 30 - 60 degrees of the two cross values and has a
positive value (i.e. an improvement in the current situation), the vector is deemed to
exist, otherwise it is disregarded.

The values are then compared to the central one, and the one giving the best
improvement selected, diagonal ones taking preference, as shown in Figure 6.7. If all
the values indicate no improvement can be made, then the central point must be near or
on the optimum operating point. Analysis of the other values calculated can also take
place to detennine whether multi-hump surfaces are present, as in the case shown in
Figure 6.8.

If this occurs, the appropriate co-ordinates can be stored to a stack for later retrieval
and examination. By altering the size of each of the crosses tested, a general search
method can be established. By using large values to start with, which progressively
decrease in size each time an optimum point is found, the system can 'home in' on the
true optimum operating point.

6.2.3.1.2 Uni-Directional Increments

This is a very simple method and involves the manipulation of only one drill parameter
at a time. Starting at the predicted minimum point, and either using a random starting
direction or one selected by the vector method, an increment in this direction is taken
e.g. an increase in rotational speed. The response of the drilling system is noted.
Progressive increments are added until such time that a deterioration is experienced.
The direction is then reversed, and backtracked until the optimum value for that
parameter established. The search algorithm then selects another parameter, and
optimises it in the same way. The parameter selection continues in a rotational fashion
until no improvements are seen, indicating the optimum operating point has been
reached.

106
12
18 25
~,

••
•• Selected Vector
••
••
••

0 •••
63.4 ••
6 •
13

I
I

~
-3

Figure 6.7. An illustration of the Vector Selection Method

10
Selected Vector
14
• •
• ••

• ••
•• • 0
•• ·68.1
-2 ,~-- .. ------.. 4

Point stored to stack


9 as possible multi-hump
surface.
5

Figure 6.8. A Diagram to Show How the Vector Method May be Used to Aid
Detection of Multi-Hump or Multi-Trough Surfaces.

107
6.2.3.1.3 Multi Directional Method

This method is similar to the uni-directional method but all parameters are searched at
the same time. This is perfonned on a rotational bases. In turn, each of the parameters
is manipulated in its respective direction and the response determined. If an
improvement is seen the search algorithm moves to the new optimum point. If not, the
search algorithm stays at the original position, and the direction of manipulation is
changed. The parameter is not manipulated further until all the other parameters have
been manipulated.

6.3 The Complete Cost Optimisation System

The complete optimisation system is summarised in Figure 6.9, which shows the
main parts of the control scheme. Infonnation about the drilling operation is fed directly
via transducers on the rig etc, to a central data base. The infonnation within this data
base, along with external data such as wear rates, geology, etc is used to load two
S.L.P.M., one for penetration rate, and the other for wear rate. The optimisation
algorithm uses the cost equation to predict the estimated minimum cost position,
from the data contained within these two S.L.P.M's. This estimated minimum cost
may not necessarily be the true minimum cost, as the penetration rate and wear rate
processes may not be fully understood i.e. the S.L.P.M.'s may only be partially full.
Therefore, a search algorithm is used to manipulate the drill parameters in an effort to
improve the current operating point and establish a lower minimum cost. This
manipulation continues until no improvements are seen, whereby the true minimum
cost operating point has been established.

As the cost optimisation scheme is running, new penetration rate and wear rate values
will be generated, both from attaining the initial minimum cost predicated point and
subsequent drill parameter manipulation. This data can in turn be fed back to the
S.L.P.M.'s to improve their understanding of their respective processes, and thus aid
subsequent minimum cost predictions.

Furthermore, with this fed back and progressive learning process, it would be
extremely foolish to loose the data held within the S.L.P.M.'s once the hole had been
completed. By saving the two S.L.P.M.'s to the computers disc system each time a
hole is completed, then reloading them at the start of the next hole, the optimisation
system's "experience and knowledge" can be passed on from hole to hole, in a similar
fashion to a human drilling engineer.

108
.. ..
'"
~.:

.. .. ( R.P .M ......
. . ~
..

II :1 .• •Drill •·• : . .
{Weight on Bit} ...
...... r .........

:11;····· .••.. ..... . •. . •..•. . ..


,r Penetration Rate "'\ ..
......
"- ~

.. J
I DATABASE I
~

(R.P.M (Weight on Bit


It :
:
5:iiYrt':'i'N G:.
.... . '
...

Self-learning . Matrix Self-learning Matrix


(S.L.P.M.) for (S.L.P.M.) for
Penetration Rate Wear Rate
Prediction Prediction

Penetration Rate) Wear Rate)

r
Search for
predicted
minimum cost

r r

Figure 6.9 A Schematic Representation of the Cost Optimisation Control


System

109
For research purposes, the cost ~nfonnation generated by cost equation and the two
S.L.P.M.'s is used to generate a third S.L.P.M. for cost. It plays no real part in the
optimisation system apart from supplying readily available cost data. For initial research
purposes it is extremely useful however, as it allows access to what the computer has
interpreted the cost function to be. When using the optimisation system in simulated
mode, the interpreted cost function can be compared directly with the simulated cost
function. This allows an accuracy of the interpretation to be detennined, as well as
ways to possibly improve the optimisation system.

6.4 Change in Rock Strata

The cost optimisation scheme discussed so far, has been made with the assumption
that it is operating under homogeneous rock strata conditions. This is obviously rare
and thus for differing lithologies, changes in penetration rate and wear rate will be
seen. Therefore, any optimisation system must be able to cope with changing lithology.

The present system could adequately cope with rock strata changes, as the S.L.P.M.'s
would slowly learn the new process associated with the new rock fonnation. However
this would not only be time consuming but also destroy the infonnation learnt about the
previous rock strata. Therefore it would be advantageous to have a set of S.L.P.M.'s
for different lithologies. These could be interchanged by the computer when a rock
strata was indicated.

For this to occur, the optimisation system has to be able to initially detect a change in
lithology and ultimately predict the new rock type being entered. The detection of strata
boundaries has been under investigation for a number of years particularly in the
surface mining environment to aid optimum blast design (32,33,51,52,53,66). A good
example of this is shown in Figure 6.10 where penetration rates have been used to
detennine the location of weak strata amongst strong overburden. This detennination
can then be used to allow the optimum placing of charges.

Increasing attention is also focusing on the parameter specific energy, which has also
been used to aid strata identification. Specific energy is the energy required to excavate
a unit volume of rock. It was originally proposed by Teale (62), and is calculated by the
equation 6.12.

1 10
)( )( ,. )( )( II II ,. X

)( ,. ,. ,. ,. ,. ,. ,. )( )
.~
IC )( ST~ 11)(
x ,. ,. X X ,.
)( )( MASSIVE II IC
X X ,. ,. X
X x CAPROCK x )(
~ X ,. X X X X X )(
~ .. ~. . .x. ..X•••~ •. ~ .. ~ . .~ . B
A X ······ ···· .. ·· · ·· · ·· ·
IC )( IC X ,. X )( )( • )(
.· .. .... . .
)(
• x • • X

,.
A

)(
foe)
~

)( . . . .. .
• X X
,. ,.
• • X ,.
>
J

""
.. . .. . LOOSE
SAND "
,. ,. J( )( )( ,. t----------------
• • • X )( • )( • ,. -

t
II ,. X X X X II
• II ,. X ,. )( II )( )( J(

,. ,. II )( )(
X ) OiARGE'
IC ,.
ST~G )( X ~
MASSIVE )()(
<
~
)()(IC.,.
RfJO< )( •
,.

II
II

)(

II
IC

IC
,.

II

BACKFIll.,.
J( )( ,. )( )( · .. .. . .,. ,.
)( )(

• )( X • • •
?

Figure 6.10 An Example of Strata Boundary Indication Through the


Monitoring of the Drill Parameters

111
"f" E nergy (e) = A
SpecIIc F + A
21t . N.U T --- (6.12)

where F - Force on the bit


A - Area of bit
N - Rotational speed
T - Torque
U - Penetration rate

All the parameters within this equation are generally easy to measure, and therefore a
constant indication of specific energy can be obtained. However it was mentioned
earlier in this thesis, that results are only worthwhile if the monitored data truly
represents down hole conditions. In oil well drilling, the data may not be so, but it is
the authors opinion that this problem will soon be resolved.

Thus for differing rock strata, differing ranges of specific energy are seen. It is unlikely
that specific energy will be solely be able to be used to predict rock type. With a
combination of other parameters such as torque, penetration rates etc, which all change
with differing rock strata, it may be possible to find some method of characterising
different rock stratas.

Research into rock strata prediction is still in is early stages (21,32,51,52,71), but it is
known that an increasing effort is being made to reliably predict rock strata types from
monitored drill parameters.

Some preliminary joint research was conducted into strata identification with another
post graduate student S. Rogers, the results of which were encouraging. A number of
rock samples consisting of different limestones and sandstones were cored and the
average specific energy of each core was calculated. Along with other such data such as
Uniaxial Compressive Strength of each core, the graph shown in Figure 6.11 was
produced.

It can be seen that there is a distinct zoning of the two rock types. At present, no
further work has been undertaken to explore the possibility of whether other rock types
fit into zones as well. If this is the case, this may allow an initial strata type predictor to
be established.

Another important parameter which is also currently undergoing investigation is


vibration analysis of the drill string. This is being used in an attempt to determine a

11 2
200
.:.;
.. ',:-;

-;./
.: ....
.:.:" .. .:
.....:

• :> .
.:.

,,-....
.~
t: 150

1
-..-
£ro
Sail stone

~
100
c
0
·a
ro
b
I· B
0
c B
0
~ 50
B

o 5 10 15 20 25 30

UCS/e (Dimensionless)

Figure 6.11 Relationship between Penetration Rate and the Dimensionless Index
V.C.S Ie (Ultimate Compressive Strength! Specific Energy)

11 3
variety of infonnation, ranging from bit wear to rock strata type, differing rocks having
different vibration signatures (51).

With a combination of such techniques, it is felt that a system capable of reliable rock
strata prediction from monitored drill data, will be available in the foreseeable future.

Returning to the optimisation system, strata identification will thus allow the correct
penetration rate and wear rate S.L.P.M.'s to be selected and used. Therefore
progressive improvements will be made to the respective S.L.P.M.'s each time a rock
strata is entered. Furthermore a check system can also be established to ensure the
correct rock strata has been selected. On initial selection of a particular set of
S.L.P.M.'s, the values contained within them can be compared to those actually being
monitored. If large discrepancies are seen and persist, it is probable that a wrong strata
type was predicted, and therefore re-selection can be made. In this way, the
optimisation system will be able to cope with changing lithological conditions.

The strata type predictor and S.L.P.M. selector can be incorporated in the over all
control scheme as shown in Figure 6.12.

6.5 Operating constants

There will frequently be situations in which the optimisation system will not be able to
. operate unbounded. For example directional controlled wells will necessitate the
maintenance of the weight on bit within defined limits. The optimisation system could
be developed to operate within certain operating constraints, in this example
maintaining 'constant' weight on bit yet still optimising cost by controlling the other
drilling parameters.

6.6 Conclusion

In conclusion, this chapter has developed a control strategy based on the cost equation
6.1, which was introduced in Chapter 4. A number of methods for detennining the
minimum cost operating point have been discussed. Initial attempts were focused on
maxima and minima theory. A system was developed using this theory which worked
successfully and gave a good insight to later work. However, in designing this system,
the assumption that W =f(P) was made, which in reality is not true. Therefore this
method was abandoned. This method may be of great use however in improving the

114
Vibration Information

Penetration Rate

I DATABASE I

Salt-learning Matrix Self-leaming Matrix


(S.L.P.M.) for (S.L.P.M.) ror S.L.P.M. Selector
Penetration Rata .. Wear Rata
Prediction : Prediction

Search for
predicted
minimum cost

Figure 6.12 The Complete Cost Optimisation Scheme with Lithology


Predictor

11 5
interpolation method of the predictors and locating multi-peak surfaces.Partial
differentiation was also attempted but no solutions were found.

A continuous computer searching method was developed, which searches for a


minimum cost operating point and continually manipulates the drilling parameters to
ensure minimum cost drilling is being achieved. To aid this searching process, a
prediction mechanism (the cost equation and the two S.L.P.M.'s) locates a minimum
cost operating position from information previously learnt. The search routine then
"walks" the drill parameters to the true minimum cost operating position. A simplified
diagram of this process was shown in Figure 6.6.

To allow the system to cope with changes in lithology, each strata type will have an
associated pair of S.L.P.M.'s. Using parameters such as specific energy, torque,
penetration rate and vibrational analysis, rock boundaries will be detected and a
prediction of the new strata type made, consequently allowing the appropriate
S.L.P.M.'s to be selected. The prediction of multi-lithology holes however, is still in
its early stage of development. The complete cost minimisation system was shown in
Figure 6.12. The results of tests using this system are covered in the next chapter.

116
Chapter 7 - The Results of the Optimisation System

7.1 Introduction

The previous chapters have been concerned with the design of the cost optimisation
scheme. This chapter discusses the development of the cost optimisation programme,
and the results of the test work undertaken to validate the optimisation scheme. With the
development of such a computer system, it is difficult to indicate to the reader in writing
why certain methods are better than others and the success in performance of the
optimisation scheme. However with the use of a 3-D plotting routine, it is hoped to
convey this information, which would otherwise be easily shown visually by a
demonstration of the optimisation system.

7.2 The Cost Optimisation Computer Programme

It has been previously been mentioned that the drill rig used in this research project, had
been instrumented to allow monitoring and control of its parameters through a BBC
Micro-computer. Due to the limited memory and processing power of this machine, the
main optimisation system was developed on an I.B.M. type computer. To avoid the
rebuilding of the drill electronics to suit the I.B.M., the BBC would be utilised in a
front end processor role, with monitored data and control information being passed
between the two machines via an RS232 data link.

The IBM supports many different languages, and therefore a decision on which one to
base this optimisation system around had to be made. Several languages were
considered 'C' and Pascal being the favoured two. Pascal is an extremely versatile
language with both good user and graphics capabilities, as well as an easy structure.
'C' is a language which allows as standard many unorthodox practices enabling more
innovative programming. However, this makes the code more difficult to understand.
As this was an initial development system in which ready understanding of the various
routines was necessary, Pascal was chosen.

The optimisation scheme in its entirety has been shown previously in Figure 6.12. It is
a large and complex system. The development of the programme was therefore split
into tasks which were developed, encoded and tested separately e.g. data transfer
mechanism between the BBC and IBM, the data storage and interpolation method
described in Chapter 5, the control scheme and search methods as discussed in Chapter

11 7
6. Many other minor tasks and utility programmes were also developed, such as a 3-D
screen plotter, methods of developing customized simulation data, and X-Y plotter
routines for hard copies.

If all these routines were contained in one programme, the size would be enormous
hindering, editing and debugging. Therefore many of the tasks were split into Units
(sub programmes which are compiled separately). These can be referenced (linked) to
the main programme, and their contents used as normal.

7.3 The Testing of the Optimisation System

The majority of the test work focused on two main parts of the optimisation system, the
data storage and interpolation method (S.L.P.M.'s) and the search routines for
'walking' the drill to the minimum cost operating point. The testing of the S.L.P.M.'s
was covered in Chapter 5, and the results given. With this process developed, the
testing of the search routines could be undertaken. The remainder of this chapter will
focus on the results of the various search methods developed in Chapter 6 and the
results of the overall cost minimisation system.

The testing of the optimisation system using the complete system i.e. the IBM, BBC
and laboratory drill rig would have been extremely complex and tedious. Therefore it
was decided to split the testing into several parts. An initial creditation phase would be
conducted using the IBM alone, with set imaginary processes for penetration rates and
wear rates. Once accomplished, the system could be tested by using the drill simulator
incorporating the data transfer mechanisms between the mM and the BBC, and fmally
the laboratory machine itself.

It was also felt that using the criteria of optimising by minimum cost would also add an
extra degree of complexity due to the requirement of two simulators i.e for penetration
rates and wear rates, to generate the cost data. The optimisation surface seen therefore,
would bear no resemblance to either of these two processes. Consequently, this would
hinder debugging and possible improvements to the search methods.

Therefore it was decided to test the various systems through maximisation of


penetration rates. While this at first seem contrary to the design of the optimisation
system, the criteria of both systems is to establish either a maximum or minimum
point, by finding progressive improvements to the current operating point until such
time that no improvements are made. Therefore by declaring whether an improvement

1 18
is a lower value (i.e for minimising cost) or a higher value (for maximising penetration
rates), the optimisation system can be used for both scenarios.

In this way, by using maximum penetration rates, only one simulation process is
required. This gives a much simpler optimisation system, in which its search
performance can be readily seen and interpreted. This would also aid later test work
when using both the drill simulator and the laboratory drilling rig, as visual
determination of maximum penetration rates is much easier than minimum cost.

7.3.1 Test Work Using the I.B.M Alone

Much of the development and test work was completed using the LB.M on its own.
This not only simplified the testing of the optimisation system as no data transfer
mechanism etc was necessary, but also and more imponantly reduced the time taken
for each test, which was essential for debugging and further development work.

The series of tests conducted initially sought to establish the best search method for
"walking" the optimisation system to the maximum penetration rate position. This
search method would then be tested under more severe conditions. With this complete
and the test results satisfactory, the optimisation system would be changed to minimum
cost and the system tested under progressively adverse but realistic conditions.

7.3.1.1 Maximisation of Penetration Rates

As penetration rates are generally available on line, their ready availability negated the
need for a prediction and interpolation mechanism, as this only serves to improve the
overall understanding of the penetration rate process, and does not help optimisation.
Therefore, when the optimisation scheme was tested using the maximisation of
penetration rate mode, the ripple method was switched off, increasing the speed with
which optimisation was achieved. For the description of the results however, it has
been used to show how the ripple system aids overall understanding of the simulated
penetration rate process.

Figure 7.1 shows the simulated penetration rate process used for all these tests. Each
test was also started at the same point to allow some comparison to be made.

11 9
Fi 1ename : Pensim2

0,0

Figure 7.1 A 3-Dimensional Plot of the Simulated Penetration Rates


V sed in the IDM Alone Tests

120
7.3.1.1.1 Establishment of the Search Routine

i ) Vector Method
This method evaluates four points in a cross format from its present position and selects
the best direction in which optimisation can be achieved. When no improvements can
be seen, the search method would be at the optimum operating point (refer to
6.2.3.1.1). The results of the Vector method are shown in Figures 7.2 - 7.4. From
these figures it can be seen how the optimisation system has attained maximum
penetration rate, and the progressive learning of the penetration rate process by the
S.L.P.M.'s through the Ripple method. The route taken by the method is a fairly direct
route. However, the number of parameter changes required to do this is great, as each
move requires four separate manipulations of the drill parameters. This is clearly not an
ideal system.

Improvements can be made by initially increasing the size of the search cross, and
reducing its size each time an optimum point is found. The increase in efficiency is
shown in Figure 7.5. It is interesting to note the change in path direction.

ii ) Uni-Directional Method
This search method involves the manipulation of one parameter in such a way as to
locate this parameters optimum operating position. When achieved, another parameter
is then selected and optimised in a similar way. This continues until there are no
changes in any of the parameters, yielding the optimum point. (refer to 6.2.3.1.2). The
result of this method is shown in Figure 7.6. From this plot, it can be seen that the
optimum point has been reached once again, but with a reduced number of parameter
manipulations compared to the Vector method. The efficiency of this search routine
depends however on the initial start direction, which in this case is a random value.
From Figure 7.7, a more direct and hence efficient route has been followed by starting
with a different start direction.

iii ) Combined Vector and Uni-Directional Method


To eliminate the faults of the last two systems i.e, the large number of holes required by
the Vector method and the varying efficiency of the Uni-Directional method, a
combined system was developed.

In this search method, the initial start direction would be determined by the Vector
method. The Uni- Directional method would then continue to optimise in this direction
until an optimum was found. The Vector method would be used once again to select

1 21
Test · - Vector

0,0

Number of changes of parameters = 35

Figure 7.2 An Early Stage of the Vector Method Locating Maximum Penetration
Rate Process

122
Test : Vector

0,0

Number of changes of parameters = 60

Figure 7.3 An Intennediate Plot of the Vector Method Test Showing the Progressi ve
Learning by the S.L.P.M.of the Penetration Rate Process

123
Test : Vector

Jy~ , ,
\.; '9ht
~.~ .~.
0,0

Number of changes of parameters = 85

Figure 7.4 The Final Path Taken by the Vector Method to Locate Maximum
Penetration Rate

124
Test · Vector

0.0

Number of changes of parameters = 56

Figure 7.5 An illustration of the Improvement in Efficiency of the Search by


Increasing the Size of the Search Cross

125
Test : Uni-Directional

0,0

Number of changes of parameters = 56

Figure 7.6 A Plot of the Final Path Taken by the Uni-Directional Method to Locate
Maximum Penetration Rate

126
Test •
Uni-Directional

0,0

Number of changes of parameters = 41

Figure 7.7 The Change in the Efficiency of the Uni-Directional Method Caused by a
Different S tarring Direction

127
another direction for optimisation by the Uni-Directional method. This would continue
until no further changes were seen. Figure 7.8 shows the result of this method.
Ironically, in this case it has followed the same path as the Uni -Directional method, as
on each vector assessment these directions proved the most promising. However by
using a different starting location the benefit of the combined system can be seen -
Figures 7.9 and 7.10.

iv ) Multi-Directional Method
This method is similar to the uni directional method but the parameters are manipulated
only once (rather than continually until an optimum is found), before the selection of
another parameter. The rotational selection repeats itself until no further change are
seen, (refer to 6.2.3.1.3). From these results, Figure 7.11 and 7.12, it can be clearly
be seen that this picks a direct route locating the optimum position, and is not effected
by its starting position or direction. This method also has the lowest number of
parameter manipulations.

7.3.1.1.2 Conclusion of the Search Method Tests

It has been shown that all the methods found the maximum penetration position, but
with varying degrees of efficiency. The Vector method while choosing the most direct
route, required a large number of parameter manipulations to do so. The Uni-
Directional method could also be one of the most efficient methods, but was sensitive to
starting direction. In an effort to enhance both systems, they were combined. This
produced quite a good system, which generally found the optimum operating point
efficiently. However in certain situations due to the nature of the surface, its efficiency
would be reduced. The Multi-Directional method proved to be the best routine tested. It
had no starting constraints and established a near direct route to the optimum position.
For these reasons, the Multi-Direction method was chosen as the search method for the
optimisation system.

While only one series of tests have been shown, several surfaces were used to test the
chosen search routine. These surfaces however, were restricted to uni-peak or uni-
trough surfaces, as at this initial development stage, no capability for multi-peak
surfaces has been included. All tests proved successful.

128
Test · Combined Vector and Uni-Directional

0,0

Number of changes of parameters = 71

Figure 7.8 The Final Path Taken by the Combined Vector and Uni-Directional
Method to Locate Maximum Penetration Rate

129
Test •
Uni-Directional

I ,

,
i
I
I I
Jrf:.\ '
v'9ht
0,0

Number of changes of parameters = 43

Figure 7.9 The Path Taken to Locate Maximum Penetration Rate by the Uni-
Directional Method from a Different Starting Position

130
Test · Combined Vector and Uni-Directional

i l
I I I

0,0

Number of changes of parameters = 77

Figure 7.10 The Benefit of the Combined Method Over the U ni- Directional Method
Starting from the Same Location

131
Test : - Mu 1t i -0 i rect i ana 1

0,0

Number of changes of parameters = 8

Figure 7.11 An Intennediate Plot of the Multi-Directional Method Showing the


Progressive Learning of the Penetration Rate Process by the S.L.P.M.

132
Test Multi-Directional

l! ;\ I i
!
I
i I I

,t j I :I

LllU ~ I

eIght
0,0

Number of changes of parameters = 29

Figure 7.12 The Final Path Taken by the Multi-Directional Method to Locate
Maximum Penetration Rate

133
Several other points are also worth noting at this stage:-

1) It can be seen that the Ripple method in all cases has enhanced the S.L.P.M.'s
knowledge of the simulation process shown in Figure 7.1, (i.e. penetration rate as a
function of weight on bit and rotational speed). While the general trend has been
established, discrepancies by the interpolation system do exist at the peripheries.
However as the system is searching for maximum penetration rates, this does not
matter as in the S.L.P.M., the area around the maximum penetration point is well
defined. Improvements to the ripple method (by curve fitting etc) would serve to
improve their prediction further.

2) In the testing of the Vector Method, improvements were made by increasing the
initial size of the search pattern and progressively reducing them as optimum points
were established. Figure 7.13 show the improvements made to the Multi-Directional
search by doing this. It can be seen after just a small number of parameter
manipulations, the search routine has nearly reached the optimum operating area. Due
to the progressive reduction of search sizes, the method requires actually more
manipulations than the original method. However, the latter are all in the maximum
penetration rate area and thus while requiring a greater number of manipulations, higher
penetration rates are achieved at a much earlier stage. This therefore provides an
improvement to the optimisation system, but for further development the search size
was kept to unity, being easier to follow the path of the search routine.

7.3.1.1.3 Data Variance

With the search method selected, it was decided to see how the system would cope with
fluctuations in penetration rate measurements. This would be more akin to the real
situation where for set drill parameters, (in this case weight on bit and rotational speed),
factors such as friction, minor lithology changes, etc would cause fluctuations in the
measure penetration rates. However by calculating the mean for each point as described
in Chapter 5, with the optimisation system visiting the point several times, a
representative value could be calculated, reducing the effect of rouge values and thus
allowing optimisation to continue.

Using the simulated process as shown in Figure 7.1, a random variation of ! 20% of
the simulation value was introduced. The results of the optimisation test were quite
surprising, with the search method attaining the maximum penetration rate area in quite
a short time, Figure 7.14. The word area is used because, unlike in the non variance

134
Test : Multi-Directional

0,0

Number of changes of parameters = 33

Figure 7.13 The Improvement in Efficiency to the Multi-DirectionalMethod by


Increasing the Size of the Search.

135
Test Multi-Directional

0,0

Number of changes of parameters = 86

Figure 7.14 A Plot of the Optimisation System Attaining Maximum Penetration


Rates with a !20% Variation in Penetration Rate Values

136
system where values are constant, the fluctuation in data has the effect of varying the
mean values, slightly causing the search method to "wander" around this maximum
value area.

Such was the success of the optimisation system, a second test was developed where
the data values fluctuated "2- 60% of the simulated value. The system once again
attained the maximum penetration rate area, Figure 7.15, but a considerable number of
parameters changes were required, with the path being fairly contorted. However, it
does show the optimisation system can work under fairly extreme conditions.

7.3.1.2 Minimum Cost Optimisation Using the IBM Alone

The previous section established the best search method to be used in the optimisation
scheme as well as testing it to a satisfactory degree, using maximisation of penetration
rates as the criteria. The system was therefore changed to optimise by minimum cost
drilling, to allow further test to continue.

To generate cost data, two simulators are required, one for penetration rates and the
other for wear rates. The two used in this series of tests is shown in Figures 7.1 and
7.16 respectively. The other variables in the cost equation were set to constant values as
shown.

(B + R. Tm . D) . W R
C - K +p

where Rig Costs (R) = £l000/day


Bit Costs (B) = £500
Round Trip time /metre (Tm)= 20 seconds/metre
Crown Height (K)= 10mm
Depth (D)= 1000m

Using these variables and the data held within the two simulators, a simulated cost
surface can be generated as shown in Figure 7.17. During the optimisation tests, unlike
those of maximum penetration rate tests, the cost S.L.P.M. will not resemble the
simulated surface as no rippling takes place. Therefore only the path of the optimisation
system will be shown along with the cost calculated values.

To test the optimisation system under the minimum cost drilling criteria. four scenarios
were used with increasing severity, but increasing realism. In each test. the search

137
Test •
• Multi-Directional

0,0

Number of changes of parameters = 419

Figure 7.15 The Optimisation System Attaining Maximum Penetration


Rates with a "! 60% Variation in Penetration Rate Values

138
F i 1ename · Wearinf2

0,0

Figure 7.16 The Simulated Wear Rate Process

139
F i 1ename •
• Cost1000

I I
I I J

I;I
1I I
r
I'
j

i ~~~~
I I '
I iI
~ i

eIght
0,0

Figure 7.17 The Cost Surface Generated by the Cost Equation, Using the
Two Simulators and the Defmed Cost Variables

140
method would remain the same, but the data from both simulators would be changed
according to the test. The tests are listed below.

1) Simulator wear rate and simulator penetration rate


2) Simulator wear rate and penetration rate variance
3) Randomly generated wear values and simulator penetration rate
4) Randomly generated wear values and penetration rates variance

7.3.1.2.1 Simulator Wear Rate and Simulator Penetration Rate

In this test, when a cost value was required for a point, the values of penetration rate
and wear rate used, were those given directly by the respective simulators, i.e. no data
variations were included. The result of this test is shown in Figure 7.18. It can be seen
from this that the optimisation system, has directly "homed" into the minimum cost
value position. The number of manipulations to do this is relatively small, indicating a
good degree of efficiency. However, it does have to be remembered that there are no
data fluctuations and thus optimisation is relatively easy.

7.3.1.2.2 Simulator Wear Rate and Penetration Rate Variance

To complicate the process slightly, a degree of variation in penetration rates was added.
When the cost values were calculated, the wear rate was that returned by the simulator,
where as the penetration rate value had a ~20% variance added to the simulator value
(the same process as used in the latter tests of maximisation of penetration rates). The
results are shown in Figures 7.19 and 7.20. Figure 7.19 shows the cost surface
generated. From the figure, it can be seen that the path is slightly more contorted than
the previous test. This is due to the variation of penetration rates altering the calculated
cost values. Only when a representative mean value for penetration rates for a particular
point has been established, do the cost fluctuations diminish and allow the
optimisation to progress further. This wandering and averaging process is reflected by
the large increase in the number of parameter manipulations required to attain the
minimum cost position. Figure 7.20 shows the penetration rate S.L.P.M., and it can
be seen that the penetration rate process has only been partially learnt. However as the
minimum cost position has been located, the knowledge that greater penetration rates
can be achieved is irrelevant as a movement in this direction would only cause an
increase in cost from the present situation. Therefore this shortcoming does not maner.

141
Test : - Mu 1t i -0 i rect i ona 1

0,0

Number or changes or parameters = 42

Figure 7.18 A Plot of the Optimisation System Locating the Minim urn Cost Position
with No Data Fluctuations

142
Test : Multi-Directional

0,0

Number of changes of parameters = 206

Figure 7.19 The Cos~ Swface Generated During Testing with Simulator Wear Values
and Penetration Rate Variance

143
Test · Multi-Directional

0,0

Number of changes of parameters = 206

Figure 7.20 The Associated Penetration Rate S.L.P.M. Surface Generated During
Testing with Simulator Wear Values and Penetration Rate Variance

144
7.3.1.2.3 Randomly Generated Wear Values and Simulator
Penetration Rate

In the previous two tests, each time a cost calculation was required, a known wear rate
value ( i.e. one held within the simulator) was returned. However in the real situation,
this would not be the case as wear values would be generated only when the bit was
tripped. Therefore, a more realistic scheme must be developed, to reduce the regularity
with which simulator wear values are entered into the wear S.L.P.M., and rely more
on the values generated by the interpolation system i.e. the ripple method. Thus when
cost data is required, the wear value is returned from the wear S.L.P.M. rather than the
wear simulator. To ensure that some known values are passed to the wear S.L.P.M.,
simulator values would be entered into the wear S.L.P.M. at random intervals. This
would mimic the generation of a newly measured wear values as with a real life
situation. In this way, the wear S.L.P.M. would be performing its role as a data
enhancement mechanism rather than a straight storage system. With the progressive
entering of wear values into the wear S.L.P.M., the predicted wear values returned will
increase with accuracy.

In this and the following test, the random variance was set at twenty, such that on
average, after every twenty parameter manipulations and thus cost calculations, a new
wear value would be generated from the wear simulator. This would be entered into
the wear S.L.P.M. and interpolated. To aid initial prediction of the wear S.L.P.M., it
was seeded with 5 randomly positioned wear values, to imitate some limited prior
knowledge.

The results of this test, using the randomly generated wear values and simulator
penetration rates are shown in Figures 7.21 -7.22. Figure 7.21. shows the cost
optimisation surface and the path taken. It can be seen that the system has initially
headed for the maximum penetration rate area, and then unexpectedly 'U' turned. This
is entirely due to the values returned by the wear S.L.P.M. Initially the wear process in
the S.L.P.M. is not well established, and therefore, the increases in penetration rate
values, currently out way the low wear values returned by the wear S.L.P.M. However
as shown in Figure 7.22, at the point of turn, a new wear value was generated. The
resulting interpolation alters many of the surrounding wear values, to ones closer to the
simulator wear values. In this case, much higher. Consequently, subsequent cost
calculations using the new wear S.L.P.M. values reveal this area to be unattractive with
high cost values. The optimisation system thus walks back to a region of lower wear

145
Test : Multi-Directional

0,0

Number of changes of parameters = 64

Figure 7.21 An Intennediate Plot of the Cost Surface During Minimum Cost
Optimisation Using Randomly Generated Wear Values and Simulator
Penetration Rates

146
Test : Multi-Directional

0.0

Number of changes of parameters = 64

Figure 7.22 An Intennediate Plot of the Penetration Rate S.L.P.M. During Minimum
Cost Optimisation Using Randomly Generated Wear Values and
Simulator Penetration Rates

147
values, eventually 'homing' in on the true minimum cost area. Figure 7.23 shows this
later stage.

The study of Figure 7.22 reveals some interesting observations. It can be seen that the
randomly seeded values at the start of the test (the blips in the smooth surface), are on
the periphery, and consequently, their influence on the wear S.L.P.M. is less. In this
test, the initial interpolation would have indicated low wear rates through the centre of
the wear S.L.P.M., and hence the initial search towards the maximum penetration rate
area. If however, these initial points had been in strategic positions such as the one
generated at the point of turn, then the optimisation system would have achieved its
goal much quicker.

This highlights two points :-

1) The optimisation system can achieve optimisation from very little prior knowledge,
but due to this deficiency it may take some time to do so while knowledge is learnt.
However as the system is progressively used, this knowledge will be passed from test
to test, enhancing optimisation.

2) Benefits arise if some initial knowledge is known and is of strategic importance.


Obviously we cannot dictate or know which wear values are of importance, but by
priming the S.L.P.M.'s with historical data etc, it is felt that optimisation would be
achieved much quicker.

7.3.1.2.4 Randomly Generated Wear Values and Penetration Rates


Variance

Unlike the previous test, this is beginning to represents a realistic situation with the
inclusion of a penetration variance of ~ 20% of the simulator value. The rate at which
random wear values were generated was kept at twenty, which of course is very low
but this value is used to increase the speed of the optimisation, while still illustrating the
point.

From Figure 7.24, it can be seen that the optimisation system heads towards the
minimum cost area but in a contorted fashion due to the variation in penetration rates.
Once again however, the generation of low cost values from the initially deficient wear
S.L.P.M., cause the search method to establish a minimum cost position in the
maximum penetration rate region. As randomly generated wear values entered into the

148
Test : - Mu 1t i -0 i rect i ana 1

0,0

Number of changes of parameters = 172

Figure 7.23 A Later Plot of the Cost Surface with the Minimum Cost Position
Obtained

149
Test · Multi-Directional

0.0

Number of changes of parameters = 588

Figure 7.24 The Cost Surface Generated by the Test Using Randomly Generated
Wear Values and Penetration Rate Variance

150
wear S.L.P.M., the system back tracks to the minimum cost area. In so doing, a large
number of parameter manipulations are required. Despite this, it can be seen that the
system has achieved the minimum cost drilling area. Furthermore if another test was
run using the data generated from this test, the optimisations system would home into
the minimum cost area much quicker. Figure 7.25 shows such a subsequent test.

7.3.1.2.5 Conclusion of Minimum Cost Tests

The optimisation system was tested under a series of tests for minimum cost drilling,
each with progressive severity but realism. From the initial test, it can be seen, that with
idealistic data i.e. 'on-line' and non varying, the optimisation system locates the
minimum cost position directly. Once data fluctuations etc are introduced as shown, the
path with which optimisation is achieved is more contorted. This is partially shown in
the second test with penetration rate variance but simulator wear values, where although
the path is fairly direct, the number of parameter manipulations is high. This is due to
the optimisation system requiring continuous searching of the surrounding points to
develop representative means of the surrounding points, to allow optimisation to
continue.

The idea of randomly generated wear values was also introduced, where by a wear
value from the simulator would only be entered at random intervals. The wear value
used for the cost equation would be given by the wear S.L.P.M. rather than the wear
simulator. This would mimic a more realistic situation where the generation of wear
values is more sparse. The cost optimisation scheme was tested using the randomly
generated wear values, with both simulator penetration rates and penetration rate
variance. Both tests located the minimum cost area, the penetration rate variance test,
taking much longer and with more parameter manipulations required.

Both tests highlighted the disadvantage of the system starting from a null state, i.e.
limited wear information, which consequently cause the optimisation system to wander
to establish the required knowledge. However as the system showed, that by
progressive learning, it homes to the minimum cost area quickly, rather than wandering
haphazardly. Therefore by induding historical data and the passing on of information
from test to test, the system will become more efficient at locating the minimum cost
position readily as shown by Figure 7.25.

151
Test : - Multi-Directional

0,0

Number of changes of parameters = 394

Figure 7.25 The Effect of Passing On of Knowledge From Test to Test

152
7.3.1.3 Conclusion of IBM Alone

The system was tested under both optimisation criteria, and all the results proved
satisfactory. Using the IBM on its own has allowed the system to be developed and
tested to a high degree before the complexities of data transfer etc were incorporated.
This has also allowed a demonstration system to be established, negating the need for
the use of two computers.

7.3.2 The IBM and BBC Drill Simulator

With the optimisation system successfully proven using the IDM alone, the next stage
was to test it with the BBC Drill Simulator incorporated. This would ensure that the
data transfer mechanisms and relative functions would perform properly. The tests
undertaken were similar to those when the IBM was used on its own. Consequently
plots of the maximum penetration rate, cost surfaces etc are generally not included as
reference can be made to those previously shown.

7.3.2.1 Maximisation of Penetration Rates

7.3.2.1.1 Non-Fluctuating Penetration Rate

Figure 7.26 shows the Drill Simulators response to the control provided through the
optimisation system geared for maximum penetration rates. No data variance is
included. During the testing of the Drill Simulator, the Drill Simulator was always
started with random starting conditions, to allow testing to start from unknown and
varied positions. However, the previous optimisation tests were always started at the
same position. Consequently, it can be seen that the fIrst twenty seconds of Figure 7.26
show the optimisation system attaining this position (this will be seen in all the Drill
Simulator tests). Once this position has been reached, there is a rapid increase in
penetration rate, and the maximum penetration rate is found within 100 seconds.
Continuing manipulation of the drill parameters ensures that the maximum value has
been found and is maintained, but this also causes the cyclic pattern seen towards the
end of the test.

The time taken to achieve optimisation is fairly short, but no data variance occurs, and
the response if the simulator is immediate. If the laboratory drill rig was used,
optimisation would take much longer as the drill would require several seconds to attain
the requested parameters.

153
Weight on R.P.M Pen. Rate
Bit eN) (mm/min)

2000 ~ 300

3000

~
V\
200
~
2000
1000 J In
,
,;
100
1000

o o------~----~---------, ____ r- +
100 Time!(sec)
Figure 7.26 The Response of the Drill Simulator to the Control Provided by the Optimisation System
Set for Maximisation of Penetration Rates
7.3.2.1.2 Penetration Rate Variance

Figure 7.27 shows the Drill Simulator under optimisation of penetration rates but with a
"!" 20% variance in the penetration rate readings. It can be seen that despite the data
variation (which increase with increasing penetration rate), the optimisation system has
still attained maximum penetration rate. However with data fluctuations it is difficult to
see, but it can be visualised that the average would tend towards this value. The time
taken to reach the optimum position is much greater than when no data fluctuations are
seen. However under these conditions, it is not surprising.

7.3.2.2 Minim urn Cost

7.3.2.2.1 Simulator Wear and Non Fluctuating Penetration Rates

Figure 7.28 illustrates the Drill Simulators response to minimum cost optimisation. No
data variation or randomly generated wear rates have been used. It can be seen that the
optimisation system has located the minimum cost position readily after initially
attaining the starting position. However unless calculations are performed it is difficult
from this graph to establish whether true minimum cost has been found. This highlights
one of the reasons why the optimisation system was initially tested using maximum
penetration rates, as its performance is much easier to determine.

7.3.2.2.2 Simulator Wear and Fluctuating Penetration Rates

Figure 7.29 shows the Drill Simulators response to minimum cost optimisation using
simulator wear values and penetration rate variance. From the plot, it can be seen that
the optimisation system has located the minimum cost area, despite the penetration rate
fluctuations as seen by the simulators response. Again the penetration rate fluctuations
have the effect of elongating the time taken to achieve the minimum cost position,
compared to when no variation is seen.

7.3.2.2.3 Randomly Generated Wear Values and Non Fluctuating


Penetration Rates

Figure 7.30 shows the drill simulators response to randomly generated wear values but
with no penetration rate variance. In this test the random value has been set to five,
such that after approximately five cost calculations a new wear rate value is generated
and loaded into the wear S.L.P.M. The cost optimisation surface is shown in Figure

155
Weight on R.P.M Pen. Rate
Bit (N) (mm/min)

2000 ~ 300

3000

~
V\
200
0\
2000
1000

100
1000

oJ 0 I )

100 200 Time/(sec)


Figure 7.27 The Drill's Simulator's Response to the Control Provided by the Optimisation System Set for
Maximisation of Penetration Rates but with Penetration Rate Variance
Weight. on R.P.M Pen. Rate
Bit (N) (mm/min)
i
2000 --i 200

3000
t

rJ
,.......
V\
-.l
2000
1000 -l 100
n
1000

o o o )
100 Time/(sec)

Figure 7.28 The Drill Simulator's Response to Optimisation by Minimum Cost, No Data Fluctuations Present
Weitt on R.P.M Pen. Rate
Bi (N) (mm/min)

r 2000 -1 200

3000 -4 II \~
~
V\
00
2000
1000 -i 100 ~I

1000

o 0 ------------~--------~----------._--------~~----4)
100 200 300 400 Time/(sec)

Figure 7.29 The Drill Simulator's Response to Optimisation by Minimum Cost, with Penetration Rate Variance
Weight on R.P.M Pen. Rate
Bit (N) (mm/min)

T
2000 ---J 300

3000

I--' 200
V\
\0
2000
1000

100
1000

o o )
100 200 Time/(sec)
-"<

Figure 7.30 The Response of the Drill Simulator to Optimisation by Minimum Cost, with Randomly Generated
Wear Values, the Randon Value Set at 5
7.31. As previously described, with randomly generated wear values due to the lack of
initial knowledge of the wear S.L.P.M., the optimisation system at the beginning
heads off towards greater penetration rates. In this case, improvements are gained by
increasing rotational speed, followed by weight on bit increases with intermittent
rotational speed increases. The result is the search point being located near the
maximum penetration rate area. However at this point, a wear value is entered into the
wear S.L.P.M., (at about 160 seconds on the graph, shown by the flat lines of all three
parameters due to rippling). The result is a decrease in rotational speed but with further
weight on bit increases. Again at about 190 seconds, another wear value is generated.
The resulting interpolation establishes high wear rate values are present in this area and
the optimisation system walks back to lower wear values and hence the minimum cost
area where it remains.

The same test was performed again, but with the random value increased to twenty,
Figure 7.32. Again after initially setting the starting conditions, the optimisation system
passes the minimum cost area in favour of higher penetration rates. However, after
two wear values are generated (one at 130 seconds and the other at 170 seconds), the
system returns to the minimum cost area. Further manipulations and randomly
generated wear values will progressively reduce the oscillations currently seen in the
system, (220 seconds and above).

7.3.2.2.4 Randomly Generated Wear Values and Fluctuating


Penetration Rates

The last test incorporated both randomly generated wear values and penetration rate
variance. Figure 7.33 shows the Drill Simulators response to the optimisation system
control. In this test (the cost surface shown in Figure 3.34), the optimisation system
does not head directly for the maximum penetration rate area as with previous tests.
This might be considered unusual. However, examination of Figure 7.35, a plot of the
wear S.L.P.M., indicates that a seeded wear value was present in this area, resulting in
cost values here much greater than elsewhere. As it can be seen from the cost plot, the
search method has progressively increased weight on bit until such time as a randomly
generated wear value causes the search method to back track and locate the minimum
cost area. In this test, the optimisation speed was increased due to the luck of an initial
random seed value being well placed

160
Test · Multi-Directional

0,0

Number of changes of parameters = 110

Figure 7.31 A Plot of the Cost Surface Generated by Optimisation Through


Minimum Cost with Randomly Generated Wear Values, the Random
Value Set at 5

161
"eight on R.P.M Pen. Rate
Bit eN) (mm/min)

2000 ~ 200

3000

~
0\
tv
2000
1000 l 100

1000

o o )
100 200 Time/(sec)
~~~~~~~~~~~~~~~JF~il~
~~7~.3~2~The~llS~~a~s~s~n~~~tim~~oo~~~umC~t~iliRM~~Ge~m~d
--UJ~ fJI'"-'lalQes. ilie RMdom Wear Value Set at 20
Weight on R.P.M Pen. Rate
Bit (N) (mm/min)

2000 ~ 200

3000

I--"
0\
w
2000
1000 ~ 100

1000

oJ o o--------------~------------~------------_.----------_4\
100 200 300 Time/(sec)
Figure 7.33 The Response of Drill Simulator to Optimisation by Minimum Cost, with Randomly Generated
Wear Values and Penetration Rate Variance
Test : Multi-Directional

0,0

Number of changes of parameters = 172

Figure 7.34 The Cost Surface Generated by Optimisation Through Minimum Cost
Using the Drill Simulator, with Randomly Generated Wear Values and
Penetration Rate Variance

164
Test · Multi-Directional

/f~ ,
~/9ht
Of) B'It
0,0

Number of changes of parameters = 172

Figure 7.35 The Wear Rate S.L.P.M. Generated by Minimum Cost Optimisation
Using the Drill Simulator, with RandomlyGenerated Wear Values and
Penetration Rate Variance

165
7.3.3 The IBM and the Laboratory Drill

Unfortunately due to the lack of time, machine testing was not accomplished. Several
tests using the machine and various parts of the optimisation system were undertaken
during the development, but no results were saved.

However despite this shortcoming, the BBC Drill Simulator was developed in such a
way that a direct swop with the Laboratory drill software can take place. This will
remove the possibilities of software problems in the near future. The author is confident
that the optimisation system will perform as planned during real drilling trials in the
near future.

7.4 Conclusions

The theory of the optimisation system was covered in previous chapters and gave a
good introduction to the ideas behind the optimisation scheme, how it works as well as
the likely performance. From the results of this chapter, this has been taken one step
further, indicating what the current optimisation system is capable of and its likely
future performance.

It has been previously mentioned that for the development of the initial system. the
laboratory drill rig would be used. The computer system used to monitor an control the
drill rig was not suitable (in memory capacity and processor speed) for the development
of the optimisation scheme, and therefore a IBM type machine was used. Pascal was
the chosen as the programming language, being both versatile and relatively easy to
understand due to its structured nature. The optimisation programme was split into
tasks which were developed separately, to aid debugging. With the long length of the
programme, various tasks were grouped into Units (Sub-programmes) which are
referenced by the main programme, when required. Documentation on the programme
has been omitted as its length and complexity would take a separate volume to explain.
However, it is felt that the general working of the programme can be understO<XL from
the previous theoretical chapters and the results shown in this chapter.

To aid both development and prove the system, the optimisation scheme has undergone
a large testing programme. It has been accomplished in several stages, involving
different levels of complexity, utilizing the IBM alone, the IBM and BBC drill
simulator and the laboratory drill rig.

166
Initial development test work, used the IBM alone to establish the most effective and
efficient search method with which to locate the optimum position. When using the
IBM alone, simulators have to be used to generate both the wear rates and penetration
rates required. To reduce the complexities further, initial test used maximisation of
penetration rates rather than minimisation of cost as the optimisation criteria, being both
easier to develop ( only one simulator required) and more readily visualised. Of the four
search methods tested, Figures 7.2 to 7.12, the combination of the Vector and the Uni-
Directional method proved satisfactory, but the Multi-Directional method proved by far
to be the best. It found near direct paths to the optimum point, and its efficiency was
not dependent upon starting position or direction.

With the search method selected, the testing of the optimisation system was taken a
stage further, and a variance was added to the penetration rates. In the first test, the
variance was set to ~20%, and the optimisation system attained maximum penetration
rates readily, Figure 7.14. Such was the success of the system, the variance was
increased to ~60% of the simulated value. The path taken in this case, is much more
contorted and required many more parameter manipulations, but the system once again
has attained the maximum penetration rate area, Figure 7.15.

The success of these initial tests allowed the optimisation system to be switched to
search for minimum cost. This required the use of the two simulators, i.e. the
penetration rate simulator and the wear rate simulator. The other variables in the cost
equation were assigned values, which remained the same for all subsequent tests.

The first test under the minimum cost criteria used straight simulator values i.e. no data
fluctuations were incorporated. This was accomplished with relative ease, Figure 7.18.
Penetration rate variance was added, and again the optimisation system attained the
minimum cost position, but required substantially more parameter manipulations. This
increase was due to the fluctuations in the penetration rates, altering the calculated cost
values. Only when a representative mean was develop, did the fluctuations in the
penetration rates and hence cost reduce.

In the real situation, wear rate values would only be generated at random intervals,
unlike in the previous tests which relied on a simulator value for every cost calculation.
Therefore, a randomly generated wear value system was developed. Wear values for
the cost calculations were obtained from the wear S.L.P.M. , rather than from the wear
simulator. New wear values were only entered into the wear S.L.P.M. and rippled, at

167
defined random intervals. Consequently the wear S.L.P.M. would perform its true role
as a data enhancement system, rather than just a straight data storage mechanism.

Two tests were performed using randomly generated wear data, one with simulator
penetration rates, Figure 7.23, and the other with penetration rate variance of 7 20%,
Figure 7.24. Both tests achieved the minim urn cost drilling area, the penetration rate
variance taking much longer. In both tests, the optimisation system headed towards the
maximum penetration rate area, as the wear values returned by the wear S.L.P.M. were
much lower than corresponding ones in the wear simulator, due to the wear
S.L.P.M.'s initial limited knowledge. However as wear data was generated in this
area, the ripple system revealed the extent of the high wear rates in this region, and
therefore the high costs. As a consequence, the optimisation system returned to areas of
lower wear rates and thus attained the minimum cost position.

These two tests highlight that the system can work from very little prior knowledge,
but the time taken to do so may be great. However the speed of optimisation could be
increased if some knowledge was present before hand, e.g. the priming of the wear
S.L.P.M. with historical data. Furthermore, with the passing of of knowledge from
test to test, wear predictions will progressively improve, increasing optimisation speed,
Figure 7.25.

A set of similar tests were performed using the IBM and The BBC Drill Simulator, to
incorporate the data transfer mechanisms etc. The results (Figure 7.26 - 7.35), show
both plots of the optimisation surfaces, as well as graphs of the drill simulators
response. The true effect of the penetration rate variance can be seen on some of these
tests. All the tests achieved their optimum position, their efficiency depending on the
severity of the test. The success of these tests has paved the way for Laboratory drill
test trials, as the drill simulator was designed to directly interchange with the laboratory
drill monitor and control programme.

At the time of writing, laboratory test trials had not conducted. The author though is
confident that the optimisation system will perform successfully on the laboratory rig,
with the results being presented at the forth coming SPE/IADC Drilling Conference in
Amsterdam, March 91

168
Chapter 8 - Conclusions

In the exploitation of mineral reserves, drilling is an essential part of the location and
extraction processes, so much so that drilling expertise is essential for the success of
such operations. The increases in present technology, has enable the capability limits to
be pushed further each year, and new ways are continually developed to improve or
optimise drill performance. However, the development of an automatic drill, capable of
self-optimisation is a long way into the future. Some initial attempts have been made to
produce such a system, the earliest being in 1968, where the Humble Oil and Refming
Company conducted full scale trials on a oil type drill rig. Other more recent
developments have mainly been associated with mining type drill rigs such as those
manufactured by Tamrock. However the success of these projects has been varied but
to the author's knowledge, none have been commercially successful. Therefore, scope
exists for the development of a drilling optimisation system.

A brief introduction was made into optimisation and more specifically into drill
optimisation in Chapter 1. The Chapter also described that from work conducted by
Ambrose, a trade off between penetration rates and wear rates was apparent, which
could possibly be used as the control rationale for a drill optimisation system. This
formed the basis for this research project, i.e. to develop a drill optimisation system
using the trade off between penetration rates and wear rates.

A Laboratory drill rig was already in existence, which used diamond impregnated drill
bits. It was used for both Laboratory work as well as research purposes. It was decided
to develop the optimisation system around this machine, while not being ideal, it would
serve to prove the point. Chapter 2 described, that as the optimisation system would
require control as well as monitoring capabilities, due to the constraints in the old drill
electronics, the electronic system was rebuilt to allow many additional features to be
added, such as rotational speed control, weight on bit control and stop / start control of
the drill rig. The modifications to the wear measurement jig were also described as
inaccuracies were established, as the jig was progressively used.

At an early stage, it became apparent that the capabilities of the drill monitoring
computer / control computer would be insufficient to host the optimisation scheme.
Therefore. it was decided to use the BBC as a front end processor, i.e. solely dedicated
to monitoring and control of the drill rig, and have the optimisation system running on a
separate computer. This was chosen to be an IBM type machine. Data would be passed
from machine to machine via an RS232 link. Originally this was thought to be an easy

169
process, but it proved to the contrary, and only by some unorthodox practices was a
link eventually established. While this is obviously not an ideal situation, as the system
was to be an initial development phase, with future development using a different
computer and possibly drill rig, it was though not worthwhile rebuilding the drill
electronic to be geared towards the IBM.

In Chapter 3, some initial tests were conducted to generate ideas on drill optimisation
techniques as well as highlight any problems that may occur while using the laboratory
rig. The initial test took the form of a simple optimisation scheme designed to attain
maximum penetration rates. The system worked well as shown by the results.
However, some problems did occur with both the optimisation algorithm and drill
control. As the optimisation algorithm was only for tests purposes, no modifications to
it were made. The drill control problems were of importance however, and corrective
measures were taken to alleviate them. The discovery of these problems proved
invaluable in later tests and with the lessons learnt about the laboratory rig as well as for
a future optimisation system, made these initial tests worthwhile.

These tests, also highlighted the inefficiency of initial testing of the computer software
with the laboratory rig, due to the necessity for collaring etc for each test run. This
made debugging time consuming and extremely frustrating. Therefore a Drill Simulator
was developed in which the simulator programme would exactly mimic the drill rigs
monitoring and control processes. Penetration rates were provided by a matrix type
system, such that for differing drill parameter values ( in this case Weight on bit and
rotational speed) different penetration rate values would be returned. By changing the
simulation matrices, different responses could be attained, such as would be seen in
different rock strata. This would allow testing to be accomplished away for the
laboratory rig with increased efficiency. Once the programmes had been proven on the
drill simulator, testing could return to the laboratory drill. Many other additions and
complications were added to the simulator as the research project developed to assist
development, by trying to establish more realistic scenarios. This allows quite
comprehensive testing before laboratory trials are required.

With the initial test trials of the laboratory controls etc complete and a drill simulator
established with which to aid development, attention turned to developing the main
optimisation system. Using the criteria of the trade off between penetration rates and
wear rates, ways in which this could be achieved through various drill parameters were
examined as described in Chapter 4. Several ways were looked at, such as time for
each bit run, total time to completion as well as maximising penetration rates. However

170
all these proved not suitable for various reasons, either not fitting the trade off criteria
or being to complex and inflexible to provide a good optimisation scheme. The cost of
the operation was also looked at, and a brief cost analysis was conducted, to establish
the main cost centres of the drilling operation. While optimising by minimising total
cost to completion was rejected as being to inflexible, minimising cost per metre was
developed and established as the control mechanism by which this optimisation scheme
would work. It had a number of advantages, it was a flexible system, capable of coping
with unforeseen problems, good economic benefits as well as the fact that most drilling
parameters could be related to it.

A well establish equation existed with which to calculate cost per metre and it was
adopted to be used for the main control equation (equation 4.18). However, it did not
contain any controllable parameters associated with the drilling operation, and therefore
some mathematical manipulation was required, to derive equation 4.27. This relates
cost per metre to both penetration rate and wear rate. This equation would be used as
the main decision process in the optimisation system. It was also noted in this chapter
that this was a fairly simple equation, but the optimisation system would designed to
allow a more complicated one covering more of the cost centres to be used, once the
initial system had been developed and tested to a satisfactory degree.

A sensitivity analysis was performed on this equation to understand the trends etc
associated with the various parameters involved to aid the development of the
optimisation scheme.

With the optimisation equation derived, the problem of attaining reliable drilling data for
use in the equation, and hence for optimisation was covered in Chapter 5. While
penetration rates are generally easy to measure and are often available on line, wear
rates are much harder to obtain. Some equations etc, are available which may be used to
predict wear, but these can never improve their prediction. Therefore, a data
enhancement system was developed to initially store known wear values, and estimate
unknowns. The method used a matrix based system in which wear rates could be
referenced to various parameters by the dimensions of the matrix. Theses matrices
became known as S.L.P.M.'s.

To enhance known data, an interpolation system was developed, known as the ripple
method, which would radiate the influence of a known point outwards, until no effect
was seen. An extensive testing programme was conducted, to establish the reliability of
the ripple method, and the results were shown. Generally, the results proved

171
satisfactory but with complex su~aces, the predictions began to deteriorate. However,
when considering that a development system was required, they were adequate at this
stage. Improvements could take place at a later date once the main cost optimisation
scheme was developed.

The data enhancement mechanism allowed ready available and reliable data with which
to use in the cost equation, and thus design and development of the optimisation
scheme could take place. This was elaborated in Chapter 6. The chapter described
several schemes which were tried, maxima and minima being the fIrst. By establishing
a relationship between penetration rates and wear rates, differentiation of the cost
equation could occur and hence a solution could be found. A computer programme was
developed to use this method, and while this method was abandoned for logistic
reasons, it did generate some ideas, which were useful in the development of the
current system.

With no solution being found by mathematical means, a computer based search system
was developed which from known data, would locate the minimum cost position as
shown in Figure 6.6. However as the data may only be partially known, the true
optimisation position may yet be undetected, and therefore the search method must also
be capable of 'walking' the drill to the true optimisation position by manipulation of the
drill parameters. Several method were designed to do this, their theory being covered in
Chapter 6. The design of the complete cost optimisation scheme was shown in Figure
6.9.

The current cost optimisation scheme design has not considered the effect of changes in
rock strata. This was also dealt with in Chapter 6. The optimisation system would cope
as it stands with a change in rock strata, the penetration rate and wear rate S.L.P.M.'s
learning the new process corresponding to the new rock type. However this would be a
time consuming process and irradicate all the information learnt about the previous rock
strata. Therefore it was proposed that each rock strata would have its own set of
S.L.P.M.'s. For this to be incorporated into the optimisation system though, detection
of a change in rock strata must occur in addition to its prediction. Detection of strata
boundaries is possible through the use of specific energy and other related parameters.
Rock strata prediction from the measured drilling variables, however is much more
difficult. Research in this area is relatively new, with many different ideas being
followed. At present no reliable prediction system has been developed but it is thought
that one will be developed in the foreseeable future.

172
With such a system incorporated into the scheme (Figure 6.12), once a rock boundary
has been indicated and the prediction on the rock type made, the appropriated
S.L.P.M.'s can be selected allowing the optimisation process to continue. The
interchanging of the S.L.P.M's and the progressive use of the system, will allow the
gradual build up of knowledge for each rock strata. At the time of writing the
optimisation system does not include the capability for changes in rock strata.

The testing of the optimisation system was covered in Chapter 7, which described the
various tests which were used to prove the drill optimisation system. Initial testing used
the IBM alone to eliminate the requirement for data transfer between the two
computers. It also used maximum penetration rates as its optimising criteria requiring
only a penetration rate simulator, rather than the two required by the cost system.

The first series of tests involved the selection of the search routine to locate the true
optimum position through manipulation of the drill parameters. Four methods were
proposed in Chapter 6, and their results indicated that the Multi-Directional method was
be the best, (Figures 7.2 - 7.13. This method found a near direct route to the optimum
position, and its efficiency was not effected by starting position or direction. To test the
durability of this method a variance in the penetration rates was added. Initially this was
set at -:!" 20% of the simulator value and then increased to -:!" 60%. Both these tests
acquired the maximum penetration rate region, the -:!"60% variance taking much longer
and requiring many more parameter manipulations, Figures 7.14 -7.15.

With these tests performed satisfactorily, the system was switched to optimise by
minimum cost. With no data fluctuations incorporated the system located the minimum
cost position directly, Figure 7.18. Penetration variance was added, and although the
path to minimum cost was more contorted and required a larger number of
manipulations, the minimum cost position was found once again, Figure 7.19.

The two previous tests had relied 'on line' wear measurements i.e. for every cost
calculation, a wear value was returned from the wear simulator. In reality however, this
is not possible as wear values are generated at intermittent intervals. Therefore a
randomly generated wear value mechanism was developed. In this process, the wear
values for the cost calculations were returned from the wear S.L.P.M., rather than the
wear simulator. To enable the wear S.L.P.M.to learn the wear process as in real life,
wear values were entered into the wear S.L.P.M. at a variable random interval. In this
way, the wear S.L.P.M. perform the role of a data enhancement mechanism rather than
a data storage system.

173
Two tests were perfonned, one using simulator penetration rate and the other with
penetration rate variance, Figures 7.20 -7.24. The initial limited knowledge of the wear
S.L.P.M. caused some initial wandering, but once the wear process had been learnt by
additionally generated wear values, the minimum cost position was located. The effect
of the passing on of infonnation was also demonstrated by the running of the same test,
but with the infonnation of the previous test included, and showed the minimum cost
position to be located much quicker, Figure 7.25.

With the testing of the system on the IBM alone complete, a similar series of tests were
conducted using the IBM in conjuction with the BBC Drill Simulator to incorporate the
data transfer mechanisms etc. All tests again proved satisfactory as shown by the
results, Figures 26-35. The time taken however was much longer due to the response
of the Drill Simulator. This test phase ensured that the system would work with the
BBC and that the data transfer software was free fonn errors. As the simulator
programme was designed to directly interchange with the laboratory drill monitor
Icontrol programme, and thus, these tests paved the way for laboratory drill test trials.

Unfortunately due to time constraints, full scale laboratory test trials did not take place.
However various parts were tested with the laboratory rig and worked satisfactorily.
However no results were retained at the time. The author is confident though that test
trials in the near future will prove the validity of this drill optimisation system.

In concluding, a drill optimisation system has been developed which is capable of


achieving drill optimisation either by maximising penetration rates or more importantly
by minimum cost per metre drilling. The theory of the optimisation system has been
covered and much of the development work conducted described. Results of the first
phase of testing of the optimisation system have also been included and described.
They indicate the success of this initial development minimum cost drilling optimisation
system and highlight the potential it holds for revolutionising current drilling practices.

174
Chapter 9 - Recommendations for Future Work

In this thesis, a description of the fIrst stages of the development of a drill optimisation
system have been given. It has also shown the initial testing of the system, highlighting
the potential for further development, with the ultimate aim of producing a
commercially viable drill optimisation system. However before this system becomes
commercially viable, there are many more research and development stages which must
be accomplished. Some ideas and thoughts for future work are covered in this Chapter.
It has been split into two parts, fIrstly initial improvements to the optimisation system
itself and secondly, more general recommendations.

9.1 Improvements to the Optimisation System

9.1.1 Rock Boundary Indication and Strata Prediction

This subject was briefly covered in Chapter 6, where reference was made to the
detection of strata boundaries, and the ultimately the prediction of strata type. However,
research and conclusions in this area are still vague and thus a comprehensive literature
survey and analysis would be benefIcial to establish present states of the art and
possible areas where additional strata prediction research could be undertaken. The
research conducted by Rogers and Rowsell shown in Figure 6.11 should also be
continued further as these results proved promising.

A strata boundary indicator and rock type prediction mechanism should be also
incorporated into the optimisation scheme, however elementary for the following
reasons:-

1) It would allow the development of the various S.L.P.M. switching routines which
will be required, as well as some sort of testing routine to check whether the correct
selection has been made, i.e if the drill parameters show large continual discrepancies
from the S.L.P.M. values, a wrong selection may have been made.

In addition, the problem of data corruption within the S.L.P.M.'s can be addressed.
When passing through strata boundaries bedding planes etc, are likely to corrupt the
data values held within the previous rock strata S.L.P.M.'s. Furthermore, the wrong
rock type prediction, and thus wrong S.L.P.M. selection would also cause undesired
data corruption. Therefore some mechanism, such as a buffering procedure, to store

175
and vet the data needs to be developed, to limit the possibility of corrupting the
S.L.P.M.'s.

2) If the two processes i.e. boundary indication and strata prediction are unitised, when
better recognition systems are developed, they may be directly switched with the old
methods, thus not incorporating any programme alterations as would be the case if the
prediction mechanisms were incorporated into the main optimisation programme.
Furthermore, this would allow a very basic initial predictor to be used to enable point 1
to be accomplished

3) A better understanding and possible use of the parameters measured and calculated,
may be developed when concerned with strata identification.

9.1.2 Rolling Depth

All the minimum cost test, presently formed are with the depth being held constant.
This should be changed to allow the depth to be increased at the rate of penetration as
the test progresses. As the depth increases a continual change in the cost profile will be
seen, thus requiring constant re-establishment of the minimum cost position. However,
particularly in the simulated tests, the depth increase may be required to be accelerated
say by a factor of ten, since by using real time penetration rates, the time taken to reach
depths of 1()()() metres will be great.

9.1.3 Multi-Peak Surface Prediction and Improvements to the


Ripple Method

In Chapter 6, a method of locating the minimum cost position using Maxima and
Minima theory was described. To enable differentiation of the cost equation, a
substitution had to be made. This was achieved by fitting a polynomial equation to the
data in question, to eliminate one of the unknown parameters. While this method was
rejected, it does have several important processes which could be used to improve the
performance of the optimisation system.

i ) Multi-Peak Surface Predictor.


The simulation processes etc used so far have only had one maximum or minimum
point, for simplicity reasons. Some research has suggested that drilling costs will have
only one such value (31). However, it would be advantageous if the optimisation
system could cope with multi-peak surfaces in case they arose. Such a system could be

176
developed, by using the polynomial curve fitting routine used for the maxima and
mimima system. A series of curves could be fitted to the data held within the
S.L.P.M.'s for each set of parameters. The general trends of these equations could be
established to see if a second hump or trough was present. Depending on the
optimisation required, the search algorithm could be switched to investigate this region.
If improvements were seen, then the optimisation system would remain at this new
position, otherwise it would return to the previously located optimum point

Furthermore the use of these curves could also be used to establish trends in the data to
aid the Multi-Directional search method locate the optimum point with increased speed

ii) S.L.P.M. Prediction Accuracy


From the results of the ripple method it was apparent that on non linear surfaces, the
prediction accuracy was reduced, being more pronounced on complex surfaces. This is
due to the linear averaging process employed, being unable to cope with a non-linear
process.

The process could be again enhanced by using the curve fitting routine for interpolation
purposes, instead of or in conjunction with the Ripple Method. By fitting curves to the
data contained within the S.L.P.M.'s, depending on the accuracy of the fit, the
unknown data points could be adjusted to the values attained from the resulting
polynomial equation. This would enable a much better interpolation system for non-
linear surfaces.

Only outline sketches of these two recommendations have been made but it is in the
authors opinion that it would be a valuable contribution (particularly the improvements
to the prediction accuracy) if developed and incorporated into the optimisation system.

9.1.4 Wear Rate Variance

The generation of the randomly generated wear rate mechanism went along way to
improving the realism of the optimisation tests. However, the values entered into the
wear S.L.P.M. were those of the simulator, and thus contain no variance, i.e. if the
same point was entered twice, the same value would be returned from the simulator.
Therefore it is proposed to include a variance mechanism in the wear rates values such
as those seen for the penetration rates, as it is very unlikely that in the field that
variation in wear rates will not occur.

177
9.1.5 Examination of the Optimisation Programme

During the research project, every effort was made to keep the optimisation programme
as tidy and as structured as possible. However, with progressive developments etc, this
is always difficult to maintain. Furthermore, when initially developing routines,
solutions are required and generally the fIrst found becomes the only one due to time
constraints. As a consequence, many of the routines written may be ineffIcient. It is felt
that at this stage, it would be benefIcial to initially tidy up the optimisation and support
programmes, and examine the various routines to see if improvements in efficiency can
be obtained.

9.2 General Improvements and Recommendations

9.2.1 Machine Test Trials

Of all the recommendations for further work this must be the most important and which
due to time constraints was not performed. While simulation tests etc are worthwhile
and serve to reduce awkward laboratory testing, especially when developing and
debugging software, they are no substitute for the real situation. Therefore it is
recommended that the series of initial tests performed on the IBM alone, and the IBM
and drill simulator are performed to the laboratory drill rig in the near future. This will
give creditation to the optimisation system indicating it can work in a real environment
as well as a simulated case.

9.2.2 Wear Data

Most of the minimum cost test work was conducted using totally hypothetical wear
scenarios based on the work conducted by Ambrose on diamond impregnated bits.
Despite this work, there is insuffIcient wear data to establish a comprehensive wear
scenario. Although the optimisation system can ( and is designed to) run with limited
wear knowledge and enhance its knowledge through the generation of new wear rate
values as drilling proceeds, in the laboratory situation, such a testing programme is
impracticable as it would take an exceedingly long time to accomplish just one test.

This lack of real knowledge, thus prevents the system from being tested with real data,
and hence being able to determine whether realistic solutions are being achieved.
Information such as this could be obtained from some of the wear predictor equations

178
such as Galle and Woods and it is recommended that this is undertaken to see the
simulators response.

However, historical data from industry would also be beneficial in two ways. Initially
the data could be used to prime the wear S.L.P.M. and the results of this analysed.
Secondly, with more information such as penetration rates, weight on bit, torque,
stratagraphic columns etc, the optimisation system could be run to see the correlation
between those reported by the historical data, and that by the optimisation system. If the
results were widely different, examination could determine whether further
improvements to the optimisation system were necessary or whether it could predict
lower cost holes.

9.2.3 Field Test Trials

While this is some way into the future, it is worth mentioning as some steps could be
taken to aid this process. The installation of the optimisation system on a industrial drill
rig poses many problems with both safety and reliability, compared to that of a
laboratory rig. Therefore it is proposed that a survey is conducted to establish what
electronic monitoring and control systems i.e. transducers etc, is available and
permissible for use on such a drill rig. This would enable early selection to be made,
allowing the required electronics and software to be developed.

Once the hardware side has been developed, it is envisaged that field trials would take a
two phase approach, the first using the optimisation system solely in a monitoring role,
with the optimisation system results being displayed as a suggestive action. In this
way, creditation of the optimisation system could be achieved with regards to
transducer reliability, optimisation predictions etc, without directly interfering with the
drilling operation. Secondly once the initial stage was successfully proven, the control
side could be incorporated and the optimisation system run in its entirety.

With these recommendations and further research work, it is hoped that this drill cost
optimisation system may in the future be a indispensable requirement of the rig floor.

179
-Re ferences

1) Allen, J. H.,1977: 'Detennining parameters that affect rate of penetration'. Oil


and Gas Journal, 3 October, P.94-139.

2) Ambrose, D.,1987: 'Diamond Core Bit Perfonnance Analysis'. Unpublished


PhD Thesis, University of Nottingham.

3) Arehart, R. A.,1989: 'Drill bit Diagnosis Using Neural Networks'. SPE 19558.
64th Annual Technical Conference and Exhibition, San Antonio.

~ 4) Barrett, W. L.,1984: 'Costing a Drilling Programme'. Geodrilling, February,


P.18-22.

5) Blick, E. F. and Chukwu, G. A.,1990: 'How to predict when to pull the


bit'. Petroleum Engineer International, January, P .39-41.

6) Bode, D. J., Noffke, R. B. and Nickens, H. V., 1989: 'Well Control


Methods and Practices in Small-Diameter Wellbores'. SPE 19526. 64th Annual
Technical Conference and Exhibition, San Antonio.

7) Bourdon, J. C. and McCann, D.,1989: 'Comparison of Field and


Laboratory-Simulated Drill-off Tests'. SPE 16162. SPE Drilling Engineering,
December, P.329-334.

~ 8) Brett, J. F. and Summers, M. A.,1984: 'Planning and Practical Problem


Solving Using an Engineering Simulator for Drilling'. SPE 13206. 59th Annual
Technical Conference and Exhibition, Houston.

9) Brett, J. F., Warren, T. M. and Behr, S. M.,1989: 'Bit Whirl: A New


Theory of PDC Bit Failure'. SPE 19571. 64th Annual Technical Conference and
Exhibition, San Antonio.

10) Brown, E. T., Carter, P. and Robertson, W.,1984: 'Experience with a


prototype instrumented drilling rig'. Geodrilling, February, P.10-15.

11) Bullen, G. J.,'The challenge of synthetic versus natural diamond in the field
of hard rock drilling'. Industrial Diamond Review.

12) Bullen, G. J., 'Continuing investigations into the behaviour of impregnated


mining bits in hard rock drilling'.

180
13) Burgess, T. M. and Lesso, W. G.,1985: 'Measuring the Wear of Milled
Tooth Bits Using MWD Torque and Weight-on-Bit'. SPE 13475. SPE/IADC 1985
Drilling Conference, New Orleans.

14) Cassapi, V. B., Ambrose, D. and Waller, M. D.,1987: 'Performance


and wear characteristics in diamond core drilling'. Drillex 87, P.5 -20.

,15) Chen, W. and Tian, J.,1988: 'Multiple Objective Programming in Drilling


Optimisation Technique'. Chinese Journal of Petroleum Drilling and Production, No.2,
P.1-5.

16) Collin, W. D.,'Achieving Optimum Drill Bit Perfonnance - A Practical


Method with Particular Relevance to Deep Hole Drilling'.

17) Cooper, G. A.,1986: 'Advanced Techniques for Laboratory Full-Scale


Drilling Tests'. SPE 14783. IADC/SPE 1986 Drilling Conference, Dallas.

18) Crain, I. K.,1969: 'Computer Interpolation and Contouring of Two-


Dimensional Data: A Review'. Geoexploration, Vol. 8, P.71 - 86.

19) Cunningham, R. A.,1960: 'How high rotary speed shortens bit life, increases
drilling costs'. Oil and Gas Journal, Vol. 58, No 28, 11 July, P.91-96.

20) Cunningham, R. A.,1978: 'An Empirical Approach for Relating Drilling


Parameters'. SPE 6715. Journal of Petroleum Technology, Vol. 30,No 7, July, P.987-
991.

21) Danell, R. E.,1990: 'Drill Monitoring for the Estimation of In-Situ Material
Strength'. Advances in Mining Equipment Performance Monitoring, McGill
Uni versity ,Montreal, Canada.

122) Dareing, D. W.,1984: 'Vibrations increase available power at the bit'. Oil &
Gas Journal, Vol 82, No 10, March 5, P.91-98.

23) Eckel, J. R.,1968: 'How mud and hydraulics affect drill rate'. Oil and Gas
Journal, 17 June, P.69-73.

24) Elliot, L. R., Barolak, J. G. and Coope, D. F.,1983: 'Recording


Downhole Fonnation Data While Drilling'. SPE 12360. Production Technology
Symposium, Lubbock,Texas.

181
25) Estes, J. C.,1971: 'Selecting the Proper Rotary Rock Bit'. SPE 3717. Journal
of Petroleum Technology, Vol. 23, No 11, November, P.1395-1367.

26) Galle, E. M. and Woods, H. B.,1960: 'How to calculate bit weight and
rotary speed for lowest-cost drilling'. Oil & Gas Journal, Vol. 58, No 46,November
14, P.167 -176.

27) Galle, E. M. and Woods, H. B.,196O: 'Practical ways to find proper bit
weight and rotary speed'. Oil & Gas Journal, November 21, P.160 -167.

28) Glowka, D. A.,1985: 'Implications of Thermal Wear Phenomena for PDC Bit
Design and Operation'. SPE 14222. 60th Annual Technical Conference and Exhibition,
Las Vagas.

29) Goldsmith, R.,1985: 'Engineering Opportunities and Challenges in Drilling'.


SPE14526. Journal; of Petroleum Technology, July, P.1141- 1148.

30) Gray, G. R. and Young, F. S.,1973: '25 Years of Drilling Technology:a


Review of Significant Accomplishments'. SPE 4700. Journal of Petroleum
Technology, Vol. 25, No 12, December, P.1347-1354.

31) Guo, X. z.,1988: 'A Preliminary Investigation on the Objective Function of


Penetration Cost Applied to Optimised Drilling'. SPE 14849. SPE Drilling
Engineering,

32) Hagan, T. N.,1983: 'Performance monitoring of blasthole drills - a means of


increasing blasting efficiency'. Surface Mining and Quarrying, October, P.A 173 -
A179

33) Hagan, T. N. and Aus, M.,1983: 'Some recommended features of future


drilling equipment - a blasting engineer's view'. Surface Mining and Quarrying, P.235-
244.

34) Howarth, D. F.,1986: 'Review of Rock Drillability and Borability


Assessment Methods'. Trans. Instn. Min. Metall, 95, October, P.A191-A202

35) Lacy, K • . , Burchardt, B. J. and Buras, K. F.,1983: 'A Computer


Data Base System Design for Drilling Information Analysis'. 58th Annual Technical
Conference and Exhibition, San Francisco.

182
+- 36) Langston, J. W.,1966: 'A Method of Utilizing Existing Infonnation to
Optimise Drilling Procedures'. SPE 1262. Journal of Petroleum Technology, Vol.
18,No 6, June, P.677-686.

37) Latva-Pukkila, P.,1988: 'Computerized Drilling The Latest Development in


Drilling Technology'. 21st Apcom Proceedings.

38) Li, Y. F. and Lau, C. C.,1989: 'Development of Fuzzy Algorithms for


Servo Systems'. IEEE Control System Magazine, April, P.65-72.

39) Lieghton, J. C., Brawner, C. O. and Stewart, D.,1982: 'Development


of a correlation between rotary drill performance and controlled blasting powder
factors'. August, P.67 -73

"40) Lummus, J. L.,1969: 'Factors to be Considered in Drilling Optimisation'.


Journal of Canadian Petroleum Technology, October - December, P.138-146.

41) Lummus, J. L.,1970: 'Drilling Optimisation'. SPE 27 44. Journal of


Petroleum Technology, November, P.1379 - 1388.

42) Lummus, J. L.,1971: 'Acquisition and Analysis of Data for Optimised


Drilling'. Journal of Petroleum Technology, November, P.1285 -1293.

43) Lummus, J. L. and Field, L. J.,1968: 'Non-Dispersed Polymer Mud:A


New Drilling Concept'. Journal of Petroleum Technology, March, P.59 - 65.

44) McDaniel, K. W. and Lummus, J. L.,1971: 'Here's how to apply


optimised-drilling techniques'. Oil and Gas Journal, 14 June, P.57 - 63

45) Millheim, K. K. and Huggins, R. L.,1983: 'An Engineering Simulator


for Drilling: Part 1'. SPE 12075. 58th Annual Technical Conference and Exhibition,
S an Francisco.

46) Mol, 0., Danell, R. and Leung, L., 1987; 'Studies of Rock
Fragmentation by Drilling and Blasting in Open Cut Mines'.

47) Moore, P. L.,1986: 'Drilling Practices Manual'. PennWell Books.

48) Morel, E. H.,1988: 'Optimisation of Gas Field Developments'. SPE 19033.


Journal of Petroleum Technology.

183
49) Outmans, H. D.,1960: 'The Effect of Some Drilling Variables on the
Instantaneous Rate of Penetration.'. SPE 8117. Petroleum Transactions AIME, 219,
P.137-149.

50) Pearse, G.,1989: 'Hydraulic Drills'. Mining Magazine, July, P.48-55.

51) Peck, J. and Pollitt, D.,1990: 'Lithological Recognition Based on Rotary


Blasthole Drilling Perfonnance Monitoring'. Advances in Mining Equipment
Perfonnance Monitoring, McGill University,Montreal,Canada

52) Peck, J. and Scobie, M.,1989: 'Drilling Intelligence and Automation'. Vol.
2, No.2.

53) Peck, J., Scobie, M. J. and Hendrick, C., 'Monitoring While Drilling
Production Blastholes: Applications in Surface Mining'.

54 ) Pfister, P., 1985: 'Recording drilling parameters in ground engineering'.


Geodrilling, October, P.8-14.

55) Remson, D.,1985: 'Planning Technique - Key to Drilling Efficiency'. SPE


13467. Journal of Petroleum Technology, September, P.1613-1621.

56) Rowsell, P. J.,1987: 'Drill Perfonnance Monitoring'. Unpublished B Eng.


Dissertation, University of Nottingham.

57) Samaniego, F. and Cinco-Ley, H.,1983: 'Pressure Transient Analysis for


Naturally Fractured Gas Reservoirs: Field Cases'. SPE 12010. 58th Annual Technical
Conference and Exhibition, San Francisco.

58) Scobie, M. J., Peck, J. and Kennedy, D.,1989: 'Drill Monitoring


Investigations in a Western Canadian Surlace Coal Mine'. Second Large Open Pit
Mining Conference.

JI. 59) Sheikloeslami, B. A., Miller, D. M. and Strong, R. E.,1984: 'A


Practical Method for Evaluating Rig Perfonnance'. SPE 11364. Journal of Petroleum
Technology, March, P.489-494.

60) Simpson, J. P.,1971: 'Drilling Fluids -Today and Tomorrow'. SPE 3714.
Journal of Petroleum Technology, November, P.1294 -1298.

184
61) Simpson, M. A.,1984: 'Calculator program optimises bit weight, rotary
speed, reducing drilling cost.'. Oil & Gas Journal, Vol. 82, No. 17, April 23, P.71-
78.

62) Teale, R.,1965: 'The Concept of Specific Energy in Rock Drilling'.


Int.J.Rock.Mech.Mining Sci., 2, P.57-73.

63) Thorogood, J. L.,1987: 'A Mathematical Model for Analysing Drilling


Performance and Estimating Well Times'. SPE 16524/1. Offshore Europe 87,
Aberdeen.

64) Unknown,1985: 'Computer Controlled Drilling'. International Mining,


October, P.21-25.

65) Unknown,1986: 'Advances in Exploration Drilling Technology'. International


Mining, January, P.32-35.

~ 66) Vynne, J.,1990: 'How to Improve the Drill and Blasting Efficiency in Open Pit
Mines'. Advances in Mining Equipment Perfonnance Monitoring, McGill
U ni versity ,Montreal, Canada.

¢67) Walker, S. H. and Millhiem, K. K.,1989: 'An Innovative Approach to


Exploration and Drilling: The Slim-Hole High-Speed Drilling System'. SPE 19525.
64th Annual Technical Conference and Exhibition, San Antonio.

68) Wampler, C. and Myhre, K.,1990: 'Methodology for selecting PDC bits
cuts drilling costs'. Oil & gas Journal, January 15, P.39-44.

69) Warren, T. M.,1983: 'Factors Affecting Torque for a Tri-Cone Bit'. SPE
11994. 58th Annual Technical Conference and Exhibition, San Francisco.

70) Welch, R. H.,1982: 'The Role of Synthetic Diamonds in Drilling'. Sixth


Australian Drilling Association Symposium.

71) Wijk, G.,1989: 'The Stamp Test for Rock Drillability Classification'. Int. J.
Rock Mech. Min. Sci & Geomech Abstr., Vol. 26,No 1, P.37-44.

72) Winters, W. J. and Warren, T. M.,1987: 'Roller Bit Model With Rock
Ductility and Cone Offset'. SPE 16696. 62nd Annual Technical Conference and
Exhibition, Dallas.

185
73) Wright, D. N. and De-Barra, C.,'Selection of diamond bit type for hard
rock drilling'. Industrial Diamond Review.

74) Young, F. S.,1969: 'Computerised Drilling Control'. SPE 2241. Journal of


Petroleum Technology, Vol. 246.

,,75) Zhang, H.,1988: 'New Methods for Drill Bit Evaluation and Selection'.
Chinese Iournal:Petroleum Drilling & Production Technology, VoIIO, No 3, June

,,76) Zhou, Z.,1988: 'Effect of Bottom Hole Pressure Differential on Penetration


Rate'. Chinese Iournal: Petroleum Drilling and Production Technology, No 2.

186

You might also like