Complete Final

Preface
This dissertation is submitted for the Degree of Doctor of Philosophy in Natural

Sciences at the University of Cambridge. The work reported here was carried out
between October 1999 and October 2002 under the supervisions of Pr. H. K. D. H.
Bhadeshia at Department of Materials Science and Metallurgy and Dr. P. Woollin
at TWI. To the best of my knowledge, this work is original, except where suitable
references are made to previous work. Neither this, nor any substantially similar
dissertation has been submitted for any degree, diploma or qualification at any other
university or institution. This dissertation contains less than 60,000 words.
Part of this work has been presented in the following publications :
D. Carrouge, H. K. D. H. Bhadeshia and P. Woollin, Microstructural change

in high temperature heat-affected zone of low carbon weldable 13 %Cr martensitic
stainless steels, Proceedings of the Stainless Steel World conference 2002 (Houston,
Texas, USA), 61-67.
P. Woollin and D. Carrouge, Heat-affected zone microstructures in supermarten-

sitic stainless steels, Proceedings of the Supermartensitic Stainless Steels conference
2002 (Brussels, belgium), 199-204.
Dominique Carrouge
October 2002
i
Acknowledgments
I am indebted to TWI ltd, the DTI and the EPSRC for co-funding this project
which was carried out in the frame of the Postgraduate Training Partnership (PTP)
scheme.
I would like to thanks Professors D. J. Fray and A. H. Windle for the provision
of laboratory facilities in the Department of Materials Science and Metallurgy at
the University of Cambridge. I am also very grateful to my supervisors Pr. Harry
Bhadeshia, Dr. Paul Woollin and the recently deceased Dr. Trevor Gooch for their
advice and encouragement.
Special thanks are due to Dr. Toshihiro Tsuchiyama and Dr. Yann de Carlan for
sharing with me their experiences and practical knowledge of martensitic stainless
steels. I am especially grateful to Doctors Peter Hart, Lee Smith, Andrew Leonard
and Andrea Gregori for their assistance and useful discussions at TWI.
I am also grateful to the present and past members of the Phase Transforma-
tions group; in particular, Franck Tancret, Miguel Yescas-Gonzalez, Pedro Rivera,
Dorothy Elder, Victoria Yardley, Daniel Gaude Fugarolas, Harsha Lalam, Gareth
Hopkins, Philippe Opdenacker, Carlos Garcia Mateo, Hiroshi Matsuda, Shingo Ya-
masaki, Matthew Peet and Thomas Sourmail.
Finally, I would like to thank my wife Angelique and my parents for their constant
love and support throughout the project.
ii
Abstract
Supermartensitic stainless steels have recently been introduced in the oil and
gas industries to substitute more expensive duplex stainless steels for onshore and
offshore tubing applications. Although easily joined by arc welding processes, the
service life of the supermartensitic welded joint in corrosive environments relies to a
large extent on the behaviour of the heat-affected zone (HAZ). The microstructure
of the HAZ in these new materials has, until now, received little attention.
The work presented in this thesis is concerned with the experimental study of
the microstructure developing in the HAZ of a range of supermartensitic stainless
steels, with particular attention to the retention of the potentially detrimental δ-
ferrite phase, and the development of a model to facilitate the choice of a suitable
post-weld heat-treatment temperature.
The microstructural examination of a variety of welds revealed the presence of
retained δ-ferrite in dual-phase and grain-coarsened HAZ regions. Under normal
welding conditions, δ-ferrite retention was more pronounced in dual-phase HAZ and
in molybdenum containing alloys. However, in multipass welds, δ-ferrite distribution
was non-uniform as a result of reheating effects.
A number of methods of estimating δ-ferrite retention have been employed. The
most corrosion-resistant alloys were successfully predicted to be more prone to δ-
ferrite retention.
Charpy impact toughness tests performed on simulated heat-affected zones re-
vealed a weak effect of δ-ferrite on the HAZ ductile to brittle transition temperature.
In fact, due to its larger grain size, the coarse-grained HAZ determined the impact
properties of the tested HAZ.
Finally, a neural network in a Bayesian framework has been created to estimate
the temperature of the onset of austenite formation as a function of the employed
heating rate. The model helps the determination of the appropriate tempering
temperature of welded joints to avoid partial austenitisation and can be applied to
martensitic stainless steels and low-alloy steels.
iii
Contents
1 Introduction to supermartensitic stainless steels 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.a Materials selection . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.b Constituents of supermartensitic stainless steels . . . . . . . . 3
1.2 Types of corrosion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.a Uniform attack . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.b Pitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.c Stress-corrosion cracking . . . . . . . . . . . . . . . . . . . . . 13
1.2.d Hydrogen embrittlement . . . . . . . . . . . . . . . . . . . . . 13
1.2.e Sulfide stress cracking . . . . . . . . . . . . . . . . . . . . . . 14
1.3 Phases present and mechanical properties . . . . . . . . . . . . . . . . 15
1.3.a Martensite . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.b Retained austenite . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.c Ferrite . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2 Welding and heat-affected zone microstructures and properties 24

2.1 Introduction to fusion welding . . . . . . . . . . . . . . . . . . . . . . 24
2.1.a Arc welding processes . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.b Variables associated with welding . . . . . . . . . . . . . . . . 26
2.2 Welding of supermartensitic stainless steels . . . . . . . . . . . . . . . 27
2.2.a Hydrogen-induced cold cracking . . . . . . . . . . . . . . . . . 27
2.2.b Welding consumables . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.c The welded joint . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3 HAZ properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
iv
2.3.a Hardness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3.b Charpy toughness . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.3.c Grain size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3.d Corrosion resistance . . . . . . . . . . . . . . . . . . . . . . . 34
2.4 Ferrite prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4.a Formation and decomposition of δ-ferrite . . . . . . . . . . . . 35
2.4.b Prediction of δ-ferrite retention in HAZ . . . . . . . . . . . . . 37
2.4.c Prediction of δ-ferrite retention in weld metals . . . . . . . . . 38
2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3 Experimental techniques 44
3.1 Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.1.a Optical microscopy . . . . . . . . . . . . . . . . . . . . . . . . 44
3.1.b Scanning Electron Microscopy . . . . . . . . . . . . . . . . . . 44
3.1.c Image analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.1.d Transmission Electron Microscopy . . . . . . . . . . . . . . . . 46
3.2 Chemical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.a EDX microanalysis . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.b X-ray analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.3 Mechanical properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3.a Hardness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3.b Charpy impact toughness . . . . . . . . . . . . . . . . . . . . 49
3.4 Dilatometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.4.a Thermomechanical simulator . . . . . . . . . . . . . . . . . . . 49
3.4.b Netzsch dilatometer . . . . . . . . . . . . . . . . . . . . . . . . 50
3.5 Weld heat-affected zone simulator . . . . . . . . . . . . . . . . . . . . 51
3.5.a Gleeble . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4 Characterisation and predictions 54

4.1 Studied materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.1.a Initial microstructures . . . . . . . . . . . . . . . . . . . . . . 55
4.1.b Retained austenite . . . . . . . . . . . . . . . . . . . . . . . . 55
4.1.c Grain size and hardness . . . . . . . . . . . . . . . . . . . . . 59
v
4.2 Dilatometric measurements . . . . . . . . . . . . . . . . . . . . . . . 61
4.2.a Phase transformations on heating . . . . . . . . . . . . . . . . 62
4.2.b Phase transformations on cooling . . . . . . . . . . . . . . . . 65
4.2.c Thermecmastor . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.3 Thermodynamic predictions . . . . . . . . . . . . . . . . . . . . . . . 68
4.3.a Equilibrium calculations . . . . . . . . . . . . . . . . . . . . . 68
4.3.b Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.3.c Comparison with dilatometric measurements . . . . . . . . . . 69
4.3.d Effects of Ni and Mo on phase diagrams . . . . . . . . . . . . 72
4.3.e Paraequilibrium calculations . . . . . . . . . . . . . . . . . . . 75
4.3.f Results of paraequilibrium calculations . . . . . . . . . . . . . 76
4.3.g Comparison with equilibrium temperatures . . . . . . . . . . . 77
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5 Microstructure in high temperature heat-affected zone 80

5.1 First series of single-pass welds . . . . . . . . . . . . . . . . . . . . . 80
5.1.a Experimental procedure . . . . . . . . . . . . . . . . . . . . . 80
5.1.b Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.1.c Coarse-grained heat-affected zone microstructure . . . . . . . 82
5.1.d Dual-phase heat-affected zone microstructure . . . . . . . . . . 89
5.2 Effect of cooling rate on δ-ferrite retention . . . . . . . . . . . . . . . 94
5.2.a Experimental procedure . . . . . . . . . . . . . . . . . . . . . 94
5.2.b Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.3 Effect of heat input . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.3.a Experimental Procedure . . . . . . . . . . . . . . . . . . . . . 99
5.3.b Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.4 Multipass welds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.4.a Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.4.b Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.5 Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . 110
6 Predicting δ-ferrite retention and dual-phase HAZ width 112

6.1 Assessment of existing methods of ferrite prediction . . . . . . . . . . 112
vi
6.1.a Ferrite factor and Balmforth diagram . . . . . . . . . . . . . . 112
6.1.b MAP-neural-ferrite-number . . . . . . . . . . . . . . . . . . . 113
6.2 Dual-phase HAZ width . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.2.a Experimental data . . . . . . . . . . . . . . . . . . . . . . . . 121
6.2.b Weld thermal cycle prediction . . . . . . . . . . . . . . . . . . 122
6.2.c Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
7 Effect of HAZ δ-ferrite on Charpy impact toughness 127

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
7.2 Furnace heat-treated specimens . . . . . . . . . . . . . . . . . . . . . 128
7.2.a Microstructures required . . . . . . . . . . . . . . . . . . . . . 128
7.2.b Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
7.2.c Characterisation of heat-treated specimens . . . . . . . . . . . 129
7.2.d Results of Charpy impact tests . . . . . . . . . . . . . . . . . 133
7.2.e Fractography . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
7.3 Gleeble simulated HAZ . . . . . . . . . . . . . . . . . . . . . . . . . . 138
7.3.a Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
7.3.b Microstructure . . . . . . . . . . . . . . . . . . . . . . . . . . 140
7.3.c Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
7.3.d Fractography . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
7.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
8 Neural network model for estimating the Ac1 temperature 147

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
8.2 Introduction to neural network . . . . . . . . . . . . . . . . . . . . . 148
8.2.a Regression analysis . . . . . . . . . . . . . . . . . . . . . . . . 148
8.2.b Neural networks . . . . . . . . . . . . . . . . . . . . . . . . . . 149
8.2.c Bayesian method . . . . . . . . . . . . . . . . . . . . . . . . . 149
8.3 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
8.3.a The variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
8.3.b Normalisation . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
8.3.c Overfitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
vii
8.3.d Training and testing of models . . . . . . . . . . . . . . . . . . 155
8.3.e Committee . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
8.4 Application of the model . . . . . . . . . . . . . . . . . . . . . . . . . 157
8.4.a Effect of chemical composition and heating rate . . . . . . . . 157
8.4.b Application to steel T91 . . . . . . . . . . . . . . . . . . . . . 160
8.4.c Application to supermartensitic steels . . . . . . . . . . . . . . 164
8.4.d Effect of the microstructure and discussion . . . . . . . . . . . 164
8.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
9 Summary and future work 168
APPENDIX ONE 171
viii
Nomenclature and Abbreviations
a thermal diffusivity
Ac1 and Ac3 temperature of the onset and completion (respectively) of austenite
formation on heating
Ae1 and Ae3 equilibrium temperature of the onset and completion (respectively)
of austenite formation
Ac4 and Ac5 temperature of the onset and completion (respectively) of δ-ferrite
formation on heating
Ae4 and Ae5 equilibrium temperature of the onset and completion (respectively)
of δ-ferrite formation
dhkl interplanar spacing for plans (hkl)
hj expression of the tanh function for a neural network with j hidden
units
Iγ and Iα integrated intensity from a given plane from austenite and marten-
site respectively
k the boltzman constant
MS martensite-start temperature
MF martensite-finish temperature
q intensity of the heat source
R the gas constant
r radial distance from the weld center line
T the temperature
Tp and T0 peak and preheat temperatures of a weld thermal cycle
Tq quenching temperature
t time
Vγ and Vα volume fraction of austenite and martensite
v arc velocity
wij weight attributed to the input i in a model of j hidden units
y general expression of the function modelled in a neural network
ix
α alpha-ferrite
δ delta-ferrite
γ austenite
γr retained austenite
σy yield stress
η arc efficiency
λ thermal conductivity
b.c.c. body-centred cubic
b.c.t. body-centred tetragonal
f.c.c. face-centred cubic
DBTT Ductile to Brittle Transition Temperature
EDX Energy Dispersive X-ray analysis
FF Ferrite Factor
FN Ferrite Number
FP Ferrite percentage
HAZ Heat-Affected Zone
HVN Hardness Vickers Number
LVDT Linear Velocity Displacement Transducer
MAP Materials Algorithms Project library
MT-DATA Metallurgical and Thermochemical Databank
PRE Pitting Resistance Equivalent
PWHT Post-Weld Heat-Treatment
SEM Scanning Electron Microscopy
SGTE Scientific Group Thermodata Europe
SSC Sulfide Stress Cracking
TEM Transmission Electron Microscopy
TWI The Welding Institute, Cambridge UK
UNS Unified Numbering System
WRC Welding Research Council
x
Chapter 1
Introduction to supermartensitic
stainless steels
The purpose of this chapter is to present the metallurgy and properties of recently
developed supermartensitic stainless steels.
1.1 Introduction
1.1.a Materials selection

During the past ten years, the oil and gas production industries have encouraged
steel manufacturers to develop corrosion-resistant and economically-viable alloys for
onshore and offshore pipeline applications. The materials must have [1, 2] :
1. Adequate corrosion resistance to a range of species found in water that may

be produced together with oil and gas, including CO2 , Cl− and H2 S, at tem-
peratures up to 150 ◦ C and at pH values as low as 3.
2. High yield strength (about 550 MPa) to allow the use of thinner pipes with a
saving in weight and cost.
3. Good low temperature impact toughness in case of emergency closure of the

well-head.
4. The capacity to be easily and reliably joined by welding (good weldability)

with short or no post-weld heat-treatment (PWHT).
1.1 Introduction 2
A comparison of the costs of typical materials used in the manufacture of pipelines

is reported in table 1.1. Although at the bottom of the price scale, the use of carbon
steels for transportation of corrosive media may not be a financially good choice
due to the need for corrosion protection, inspection, maintenance and early replace-
ment of corroded parts. Stainless steels on the other hand, are more expensive, due
mainly to their high nickel and chromium concentrations. However, their inherent
corrosion resistance precludes the need for frequent inspection and maintenance and
therefore are increasingly employed as pipeline materials [3, 4].
Amongst the different families of stainless steels, ferritic/austenitic (duplex)

stainless steels match all the design criteria but their high nickel and chromium
concentrations still render them uneconomic. On the other hand, cheaper marten-
sitic alloys such as AISI 410 or 420 (table 1.2) possess high yield strength but lower
corrosion resistance, toughness and weldability. However, these properties can be
enhanced by adding nickel and molybdenum in suitable amounts and by reducing the
carbon content. These latter modifications led to the introduction of the so-called
supermartensitic stainless steels.
C Cr Ni Mo Price
Carbon steel 0.26 0 0 0 1000

Duplex stainless steel 2205 ≤ 0.03 22 5 3 4300
Supermartensitic stainless steel ∼ 0.01 12 6 2.5 3200
Table 1.1: Comparison of prices (in $ per tonne) of alloys used for flowlines applications. All
compositions given in wt% [5].
Compared with their ancestors, supermartensitic stainless steels have enhanced

resistance to general and localised corrosion and improved sulfide stress-corrosion
cracking resistance at ambient temperature (figure 1.1). In addition, they possess
higher yield and tensile strength than duplex stainless steels (table 1.4) and are
about 25 % cheaper (table 1.1). This combination of factors renders the new steels
extremely attractive and further market growth is expected.
1.1 Introduction 3
Grade C Mn Si Cr Ni Mo
403 0.15 max 1.00 0.5 11.5 - 13.0 0 0

410 0.15 max 1.00 1.00 11.5 - 13.5 0 0
414 0.15 max 1.00 1.00 11.5 - 13.5 1.25 - 2.50 0
420 0.15 min 1.00 1.00 12.0 - 14.0 0 0
Table 1.2: Part of the AISI 400 series. All chemical compositions given in wt% [6].
Table 1.3 lists some typical compositions developed to cover a range of corrosion
resistance and price. The best corrosion resistance is achieved with the richly alloyed
grades (i.e. 12 Cr 6.5 Ni 2.5 Mo wt%).
Alloy C N Si Mn Cr Ni Mo Cu Ti
11Cr 2.5Ni 0.011 0.006 0.10 1.77 10.6 2.4 0.06 0.48 0.006
12Cr 4.5Ni 1.5Mo 0.017 0.011 0.20 1.77 11.7 4.9 1.39 0.51 0.007
12Cr 6.5Ni 2.5Mo 0.018 0.011 0.09 1.72 12.6 6.5 2.32 0.49 0.008
Table 1.3: Typical compositions (in wt%) of supermartensitic stainless steels developed by CLI-
FAFER [8].
1.1.b Constituents of supermartensitic stainless steels

Supermartensitic stainless steels are essentially alloys based on iron but con-
taining chromium, nickel and molybdenum. They owe their name to their room
temperature martensitic microstructure. To understand the metallurgy of this fam-
ily of steels, attention is paid to the effects that the key alloying elements have on
phase stability and properties.
1.1 Introduction 4
Solid mark : SSC

Half solid mark : Pitting
Corrosion rate / mm year −1

1
AISI 420
0.1
0.01
Supermartensitic
0.001
50 100 150 200
o
Temperature / C
Figure 1.1: Effect of temperature on corrosion rate, localised corrosion and sulfide stress-cracking
(SSC) of a conventional AISI 420 and a supermartensitic stainless steel (0.02 C 12 Cr 5.5 Ni 2 Mo
wt%) [7]. Experimental conditions : 0.001 MPa H2 S, 3 MPa CO2 , 5 % NaCl, 100 % yield stress,
immersed for 336 h.
i Chromium
Chromium is a ferrite stabiliser which has the property of being able to protect
itself with a passive oxide film and to contribute this property to other metals
with which it is alloyed. Stainless steels owe therefore their corrosion resistance to
chromium and a minimum of 10.5 wt% of this element is required to generate a thin,
adherent, continuous oxide film on the surface of the alloy. The exact composition of
such oxide films is difficult to determine, but they all contain chromium in (III), (IV)
and (VI) oxidation states in varying proportions [9]. Paradoxically, the oxide film
itself is formed by rapid corrosion of the steel surface. However, once established it
dramatically stifles the rate of further oxidation by acting as an insulating barrier
between the metal and the corrosive environment. The steel is thus said to be in a
state of passivity [10, 11].
1.1 Introduction 5
Alloy Y.S. (MPa) T.S. (MPa)
11Cr 2.5Ni 598 753

12Cr 4.5Ni 1.5Mo 769 918
12Cr 6.5Ni 2.5Mo 733 931
22Cr 5Ni 3Mo 480 680

25Cr 7Ni 3.5Mo 550 760
Table 1.4: Yield strength (Y.S.) and tensile strength (T.S.) at room temperature of steels pre-
sented in table 1.3 and two typical duplex stainless steels [8].
Figure 1.2 shows part of the Fe-Cr constitution diagram according to Bungardt
[12]. By analogy with iron and ordinary steels, the body-centered cubic (figure
1.3) ferrite precipitated from the melt is known as δ-ferrite, while ferrite resulting
from transformation of austenite during cooling is called α-ferrite. The two types of
ferrite show differing grain structures. δ-ferrite is often coarse when compared with
α-ferrite and its morphology retains features of the solidification microstructure.
Low carbon Fe-Cr stainless steels have either a ferritic or martensitic, possibly
semi-ferritic, microstructure depending on composition :
• When the chromium content is below approximately 12 wt% it is possible to

obtain a martensitic microstructure since the steel can be made fully austenitic
at elevated temperatures. Such steels solidify as δ-ferrite and are completely
transformed to austenite (γ) at high temperature, followed by relatively rapid
cooling to transformation into non-equilibrium martensite.
• A chromium content greater than approximately 14 wt% gives a completely

ferritic stainless steel over the whole temperature range corresponding to the
solid state and hence cannot be hardened on quenching.
• Between the austenite phase field and the fully ferritic domain, there is a
narrow range of compositions which defines the semi-ferritic alloys, with a
1.1 Introduction 6
1500 δ +L
Temperature / C
δ
o
1300
1100
δ
γ +
γ
900 α+γ
δ
+
α α
700
Fe 5 10 15 20
Chromium / wt. %
Figure 1.2: Range of liquid, austenite and ferrite (α and δ) phase in the iron-chromium consti-
tution diagram with a carbon content below 0.01 wt%.
microstructure consisting partly of δ-ferrite which remains unchanged following

solidification, the remainder being martensite.
Considering that a minimum of 10.5 wt% Cr is required to obtain passivity and that
a chromium content above 12 wt% produces semi-ferritic microstructures, marten-
sitic stainless steels have their chromium content in the range 10.5-12 wt%. At this
concentration, hardenability is considerably improved making the steels air harden-
able.
Figure 1.3: Schematic view of the body cen-

tered cubic (b.c.c.) structure of ferrite.
1.1 Introduction 7
ii Molybdenum
To increase corrosion resistance for transportation of oil and gas, the addition
of molybdenum is essential. Alloying with molybdenum reduces the general cor-
rosion rate, improves the resistance to localised corrosion at elevated temperatures
and also reduces sulfide stress-corrosion cracking susceptibility (figure 1.4) [12-16].
However, being a strong ferrite stabilising element (figure 1.5), it must be balanced
by the addition of austenite stabilising elements such as carbon, nitrogen or nickel.
Depending on the corrosivity of the environment the amount of molybdenum used
ranges between 0-3 wt%.
PCO = 30 atm
4 Ni 1 Mo
0.3 2 4 Ni 2 Mo
20 % NaCl
0.1 Open : No failure
o Closed : Failure 5 Ni 1 Mo
180 C
Corrosion rate / mm year −1
5 Ni 2 Mo
P H S / MPa
0.2 0.01
2
2 % Mo 1 % Mo
0.1 0.001
0.0001
0.0
3 3.5 4
4 Ni 4 Ni 5 Ni 5 Ni
(a) 1 Mo 2 Mo 1 Mo 2 Mo (b) pH
Figure 1.4: (a) Effect of molybdenum on CO2 corrosion resistance at 180 ◦ C for two different
nickel concentrations (specimens immersed for 7 days). (b) Effect of molybdenum on the resistance
to sulfide stress-corrosion cracking (applied stress: 100 % yield strength). Base composition 0.025 C
13 Cr 0.45 Mn wt% [13].
iii Carbon and nitrogen

It is necessary to expand the austenitic field to maintain the completely marten-
sitic microstructure in the presence of molybdenum. In this respect, carbon and
nitrogen are strong austenite stabilisers in Fe-Cr alloys (figure 1.6). However, in
supermartensitic stainless steels, both the carbon and nitrogen contents have to be
kept as low as possible, about 0.01 wt%. This is because martensite hardness in-
creases sharply with carbon concentration and therefore raises the probability of
sulfide stress corrosion-cracking and hydrogen-induced cold-cracking. On the other
1.1 Introduction 8
1300 γ+δ
Temperature / C
o
1100 γ
900
0 % Mo
α 1 % Mo
2 % Mo
700
Fe 5 10 15 20
Chromium / wt. %
Figure 1.5: Effect of molybdenum on the austenite phase stability field in the Fe-Cr phase
diagram [10].
hand, carbon and nitrogen in the presence of 11-13 wt% chromium and 0-3 wt%
molybdenum, can lead to the formation of a range of precipitates such as M23 C6
and M2 X, which diminish the effective chromium and molybdenum concentrations
in solid solution and therefore decrease corrosion resistance [11, 14, 15, 16].
iv Nickel
As both the carbon and nitrogen concentrations must be kept low, austenite
can be stabilised using substitutional solutes. Nickel has the strongest effect in this
respect (figure 1.7) and also a tendency to improve toughness.
v Phase balance in 0.01 C wt% Fe-Cr-Ni-Mo system

Based on microstructural observations, Kondo et al. have produced a diagram
(figure 1.8) representing the boundaries of martensite, δ-ferrite and austenite regions
obtained at room temperature in 0.01 wt% C Fe-Cr-Ni-Mo system [14]. They showed
that as the Mo concentration was increased from 0 to 2 wt%, the area corresponding
to the fully martensitic structure was considerably reduced. According to their
1.1 Introduction 9
Temperature / C
L +δ
1400 δ
o
L+ γ
δ +γ γ
1200
γ +c 2
1000
α γ +c
α +γ γ +c 1 1+ c2
800 +c 1
c
α+c 1 α+ c 1
600 2
0 0.2 0.4 0.6 0.8 1
Carbon / wt. %
Figure 1.6: Phase fields in the ternary iron-chromium-carbon at 13 wt% Cr, L, c1 and c2 refer
to liquid, M23 C6 and M7 C3 respectively [10].
1100
γ δ+ γ
Temperature / C
o
0 wt% Ni
900
2%
4%
6%
700
α
500
Fe 5 10 15 20 25
Chromium / wt. %
Figure 1.7: Influence of nickel on the range of the austenite phase field in the iron-chromium
system [10].
1.1 Introduction 10
diagram a 13 Cr, 2 Mo wt% steel, should have nickel concentration in the range
4 to 8 wt%. Above 8 wt%, increasing amount of austenite are retained in the
microstructure. This is because the martensite-start temperature (MS , equation
1.1, [17]) is depressed below room temperature and consequently the diffusionless
transformation of austenite into martensite cannot be completed unless cryogenic
treatments are employed.
MS (◦ C) = 540 − 497C − 6.3Mn − 36.3Ni − 10.8Cr − 46.6Mo wt% (1.1)
20
Martensite Austenite
18 +
Martensite
Chromium / wt%
δ −ferrite +
Austenite
0%Mo
16
1%Mo
14
2%Mo
Martensite
12
2 4 6 8 10 12 14
Nickel / wt%
Figure 1.8: Experimental diagram showing the boundaries of the austenite, ferrite and martensite
phases as a function of Cr, Ni and Mo concentration for 0.01 wt% C after austenitisation at 1050 ◦ C
and air cooling [14].
vi Other elements
Some other alloying elements are added to supermartensitic stainless steels to
facilitate their processing and modify their properties in one direction or another.
The effects are summarised below :
1.1 Introduction 11
Austenite stabilising elements
• Manganese, is added in concentrations up to 2 wt % to substitute some of

the more expensive nickel. It is also used as a deoxidiser [18].
• Copper, is believed to improve corrosion resistance in CO2 or CO2 /H2 S en-

vironments. However, different investigators disagree on the extent of its ef-
fect on corrosion resistance [13, 16]. Copper is also reported to increase yield
strength and reduce toughness if added in concentrations greater than 1-2 wt%
[16, 19, 20].
Ferrite stabilising elements
• Silicon, is a strong ferrite former, essential as a deoxidiser during refining of

the steel. However, to maintain a fully austenitic microstructure at elevated
temperatures its concentration does not generally exceed 0.7 wt% [18].
• Titanium. By forming very stable carbo-nitrides, Ti(C,N), titanium has an

effect comparable with that of a decrease in the carbon and nitrogen con-
tent. It limits the precipitation of chromium, molybdenum or vanadium carbo-
nitrides such as M2 X and MX which may reduce corrosion resistance and cause
secondary hardening during tempering [14]. The minimum titanium content
required to prevent these effects is four times the carbon plus nitrogen con-
centration (wt %) due to the stoichiometry of Ti(C,N). However, excessive
additions lead to the precipitation of intermetallic compounds such as TiNi
instead of Ti(C,N) [14, 18]. Titanium is also effective in maintaining a finer
microstructure in the weld heat-affected zones due to the pinning effect of
stable TiN precipitates [14].
Impurity elements such has sulfur and phosphorus are always kept as low as
possible, typically below 50 ppm and 200 ppm respectively, to maintain good hot
workability and reduce the susceptibility to sulfide stress-corrosion cracking [18].
1.2 Types of corrosion 12
1.2 Types of corrosion
The types of corrosion and stress-corrosion encountered in oil and gas wells can
be divided into three classes; corrosion by CO2 (so-called sweet-corrosion), cracking
in sour H2 S containing media (so-called sulfide stress-cracking or SSC, typically in
association with CO2 and Cl− ), and an intermediate regime of cracking in mod-
erately sour media (so-called mildly sour region). The tolerable concentration of
hydrogen sulfide in the corrosive fluid has yet to be clarified for safe use of super-
martensitic stainless steels but partial pressures as low as 0.1 bar (compared with
possibly 30 bar for CO2 ) lead to rapid failure. The most serious CO2 corrosion prob-
lems are encountered in deep wells (≥ 5000 m), where the increase in pressure and
temperature make the environment more aggressive. The causes and mechanism of
the most common forms of corrosion encountered in pipelines are introduced below.
1.2.a Uniform attack

Uniform attack is a form of electrochemical corrosion that occurs all over the
exposed surface and often leaves behind a scale or a deposit. In oil and gas flowlines,
CO2 is generally responsible for such attacks. Carbon dioxide is a weak acidic gas
which must hydrate to carbonic acid (H2 CO2 ) to become corrosive. The latter
process is slow but accelerates with an increase in temperature and CO2 partial
pressure [2]. General corrosion behaviour is usually measured by weight loss.
1.2.b Pitting
Pitting is an insidious type of localised corrosion, often going undetected and
with very little material loss until failure occurs. Pits may be initiated at localised
inner-surface defects such as scratches or slight variations in composition [4]. They
ordinarily appear inside the linepipes and penetrate towards the outer surface. The
mechanism for pitting is probably due to oxidation within the pit itself, with com-
plementary reduction at the surface. The solution at the pit becomes more concen-
trated, acidic and dense as pit growth progresses.
The pitting resistance equivalent number (PRE) has been introduced to estimate
the relative pitting corrosion resistance in terms of the chromium, molybdenum and
nitrogen concentrations [21]. Although, originally derived for austenitic stainless
steels, the use of the PRE formula (equation 1.2) is commonly extended to duplex
and occasionally to martensitic stainless steels. Supermartensitic and duplex stain-
less steels have PRE numbers in the range 12 to 20 and 20 to 40 respectively. Table
1.5 lists the range of chemical composition over which the formula has been derived.
P RE = Cr + 3.3Mo + 16N wt% (1.2)
Element Composition range (wt %)
Cr 18.6 - 25.3
Mo 2.2 - 4.4
N 0.048 - 0.22
Table 1.5: Composition range of confidence for the pitting resistance equivalent formula [21].
1.2.c Stress-corrosion cracking

Stress-corrosion cracking results from the combined action of an applied tensile
stress and a corrosive environment. Small cracks form and then propagate in a
direction perpendicular to the applied stress, with the result that brittle failure may
eventually occur. Cracks may form at relatively low stress levels, significantly below
the tensile strength. The stress that produces stress-corrosion cracking need not
be externally applied; it may be residual stress resulting from rapid temperature
change and uneven contraction, or difference in expansion coefficients in two-phase
alloys [4].
1.2.d Hydrogen embrittlement

Hydrogen-induced cracking and hydrogen stress-cracking are two manifestations
of hydrogen embrittlement. Macroscopically speaking, hydrogen embrittlement is
similar to stress-corrosion in that a normally ductile metal experiences brittle frac-

ture when exposed to both a tensile stress and hydrogen resulting from metal dissolu-
tion in a corrosive atmosphere. Hydrogen-induced cracks are most often transgranu-
lar, although intergranular fracture is observed for some alloy systems. Hydrogen in
its atomic form diffuses interstitially through the crystal lattice, and concentrations
as low as a few parts per million can lead to cracking [4]. A number of mechanisms
have been proposed to explain hydrogen embrittlement: most of them are based on
the interference of dislocations with dissolved hydrogen.
1.2.e Sulfide stress cracking

Sulfide stress-corrosion cracking is a special case of hydrogen embrittlement com-
monly associated with high strength steels. As shown in figure 1.9 this type of
cracking arises from corrosion reactions with subsequent absorption of hydrogen in
the metal.
Stress
+ −
Anode : Me Me + e
+
Me Embrittled
H zone
Crack +
H
H
H
+ −
Cathode : H + e H
Stress
Figure 1.9: Mechanism of hydrogen absorption in sulfide stress-corrosion cracking [22].
The presence of sulfur from H2 S accelerates embrittlement by retarding the for-

mation of molecular hydrogen and thereby increasing the residence time of atomic
hydrogen on the metal surface and hence its absorption. Kondo et al. found that
1.3 Phases present and mechanical properties 15
the most favourable solution to improve sulfide stress cracking resistance consisted
of decreasing the steel hydrogen content by reducing the general corrosion rate. In-
deed, decreasing the corrosion rate results in a decrease of hydrogen generated on
the steel surface, which in turns reduces the hydrogen permeation rate and hence
decreases the steel hydrogen content [14].
Hardness criteria are generally used for ranking of steels with regard to sulfide
stress-corrosion cracking resistance. The maximum hardness limit to avoid the po-
tential for sulfide stress cracking in martensitic stainless steels in CO2 and H2 S is
23 HRC (about 240 HV) according to the requirement of the National Association
Corrosion Engineers (NACE) standard MR0175 [23].
1.3 Phases present and mechanical properties
1.3.a Martensite
Martensite is the desired predominant phase in martensitic stainless steels. Its
presence is a consequence of the diffusionless transformation of austenite during
cooling to room temperature. The carbon which was in solid solution in the octa-
hedral interstices in the austenite phase, becomes trapped in the martensite which
contains three times as many octahedral interstices arranged in three sublattices. If
the carbon atoms order into one of these sublattices then the martensite adopts a
body-centered tetragonal crystal system.
Martensitic transformation in supermartensitic steels is athermal, meaning that

the amount of martensite formed depends only on the undercooling below the Ms
temperature and not on the time at temperature. This behaviour is expressed in
the Koistinen and Marburger equation [24] :
Vm = exp(−0.011 × (Ms − Tq )) (1.3)
where Vm is the fraction of martensite and Tq is the quenching temperature below

Ms .
The displacement associated with martensite formation is an invariant-plane

strain, in which the shear is accompanied by a volume change normal to the invariant
plane (habit plane). The shear strain is approximately 0.22 and the dilatational
strain about 0.03 [25]. These strains are large and force the martensite to adopt a
thin-plate shape in order to better accommodate the shape deformation. In alloys
containing less than about 0.6 wt% C, the martensite forms as laths which are
aligned parallel to one another. The laths are grouped into larger structural entities,
called blocks and packets (figure 1.10) [26]. The microstructure of lath martensite
is generally too fine to be revealed using optical microscopy, necessitating the use of
electron microscopy [4].
Laths
Prior austenite
grain boundary
Packet
Blocks
Figure 1.10: Schematic illustration of the microstructural fea-

tures of lath martensite.
i Martensite hardness and toughness

The tetragonal strain caused by carbon trapped in the structure renders disloca-
tion motion difficult and therefore martensite is harder than austenite. Martensite
hardness increases dramatically with the carbon content (table 1.6), but with only
0.01 wt% C, supermartensitic stainless steel martensite is relatively soft. Martensite
hardness is also claimed to be affected by the presence of nitrogen. In this respect,
nitrogen has been found to be 2.5 times less effective than carbon, but this effect is
not widely recognised [27].
Carbon content / wt% 0.068 0.133 0.206 0.450 0.600
Hardness HV10kg 364 462 480 580 620
Table 1.6: Effect of carbon content on martensite hardness in 12 Cr 0.25 Si 0.02 N (wt%) steels,
after quenching from 1050 ◦ C [11].
Richly alloyed supermartensitic stainless steels possess low Ms temperatures

(typically 200 ◦ C) and consequently little or no auto-tempering of the martensite
occurs during cooling to room temperature. The quenched and tempered condition
helps to optimise both strength and toughness. In industrial practice, the tempering
treatment is generally carried out few degrees above Ac1 to accelerate martensite
recovery and to precipitate some carbo-nitrides [28]. Due to the small carbon and
nitrogen concentrations, precipitation is limited and thus the loss of corrosion resis-
tance due to tempering is minimal.
All body-centered cubic metals, including supermartensitic stainless steels, show

a marked temperature dependence of the fracture toughness, as indicated in figure
1.11. At high temperatures fracture occurs normally by ductile rupture, whereas
cleavage is the dominating fracture mode at low temperatures. The low carbon con-
tent of supermartensitic steels and appropriate tempering treatment ensure reduced
transition temperatures so that satisfactory toughness can be achieved even at low
temperatures, as shown on figure 1.12.
1.3.b Retained austenite

Austenite (f.c.c. structure figure 1.13) might be present in supermartensitic
stainless steels as a result of two different processes. As mentioned before the addi-
tion of substitutional elements depresses MS temperature and consequently cooling
from the austenitic field might lead to uncomplete martensitic transformation. On
the other hand, austenite can be retained in the microstructure as a result of an
Absorbed Energy
Austenitic steels (f.c.c.)
Ferritic steels (b.c.c.)
Test Temperature
Figure 1.11: Effect of temperature on notch toughness for face-

centered and body-centered cubic structures [22].
0.01
0.11
300 0.01 C, 11 Cr, 1.5 Ni (wt%)
Impact Energy / J
0.01 C, 12 Cr, 6.5 Ni, 2.5 Mo (wt%)

200 0.22
0.31
100 0.43
0.53
0.63
0.67
0
−200 −100 0 100 200
o
Temperature / C
Figure 1.12: Influence of carbon content on the Charpy V-notch energy

versus temperature behaviour in steels and comparison with supermartensitic
stainless steels [29, 30].
intercritical annealing treatment i.e. carried out above the temperature of the onset
of austenite formation on heating (Ac1 ). In this case the stability of the reverted
austenite is dependent upon the heat-treatment duration and temperature. If Ac1
is exceeded by say 40 ◦ C, only a small quantity of austenite is formed. Because
the transformation is diffusional the austenite is enriched in elements such as nickel,
carbon and nitrogen. The latter enrichment strongly depresses MS temperature
and consequently austenite remains stable after cooling. At temperatures where
the equilibrium austenite content is large, the relative enrichment is reduced, and
cooling to room temperature leads to complete martensitic transformation [31].
Figure 1.13: Schematic view of the face-

centered cubic (f.c.c.) structure of austenite.
Diagrams giving quantitative estimates of the room temperature austenite con-

tents have been developed on a range of low carbon nickel-containing martensitic
stainless steels after various reheating treatments (figure 1.14). Austenite reversion
usually commences when the Larson Miller parameter, P , exceeds ∼ 16 equating to
tempering for 16 min at 550 ◦ C or 75 s at 600 ◦ C. Most of the reverted austenite
formed between P ∼ 16 and 17.5 remains stable on cooling. Above P ∼ 18 reverted
austenite becomes increasingly unstable and transforms to untempered martensite
on cooling. The maximum volume percentage of reverted austenite that remains sta-
ble after isothermal treatments and cooling to room temperature does not generally
exceed 30-35 % [1,32-37]
i Hardness and toughness in the presence of retained austenite
Figure 1.15 relates the hardness of the microstructures obtained after various
heat-treatments, to the tempering parameter P . Below a tempering parameter of ∼
18, the presence of austenite reduces hardness, but once this parameter is exceeded,
100 100
13 Cr 4 Ni 0.5 Mo 13 Cr 6 Ni 1.5 Mo
80 80
Austenite / %
Austenite / %
and 13 Cr 6 Ni
60 60
Unstable γ Unstable γ
40 40
20 20
Stable γ Stable γ
0 0
15 16 17 18 19 20 21 15 16 17 18 19 20 21
3 3
P = T (20+log t) 10 P = T (20+log t) 10
0.01 0.01
Time / h
Time / h
0.1 0.1
0 0
75
75
0
0
55
60
65
70
55
60
65
70
10 10
50
50
0
0
0
0
45
45
100 100
0
0
Figure 1.14: Effect of various tempering conditions on austenite reformation and stability in
low carbon martensitic stainless steels [1]. P is the Larson Miller parameter. Temperature (T )
and time (t) are expressed in Kelvin and hours respectively.
hardness increases sharply due to the formation of fresh, untempered martensite.
The toughness improves in the presence of stable finely dispersed films of austen-
ite in low carbon martensitic stainless steels [39, 40, 41]. It has been argued that
when retained austenite is present near a propagating crack, the concentrated strain
field at the crack tip induces transformation into martensite. This mechanically-
induced transformation would absorb energy and thus increase the toughness. The
associated volumetric expansion of the martensitic transformation would tend to
close the crack and relieve stresses at its tip. The latter mechanism absorbs strain
energy during fracture and therefore limits crack extension [42].
ii Corrosion resistance in the presence of retained austenite

In titanium free supermartensitic alloys containing 0.02 wt% C, the retention
of austenite formed during intercritical annealing was found beneficial to corrosion
resistance [33]. The latter was attributed to the dissolution of chromium and molyb-
360
13 Cr 6 Ni 1.5 Mo
340
Brinell Hardness
320
300
280 13 Cr 6 Ni (Mo)
12 Cr 4 Ni 0.5 Mo
260
240
15 16 17 18 19 20 21
3
P = T (20+log t) 10
0.01
Time / h
0.1
1
75
0
55
60
65
70
10
50
0
0
45
100
0
Figure 1.15: Effect of various tempering conditions on hardness of low carbon martensitic stain-
less steels [38].
denum carbo-nitrides in the favour of austenite. Since there was no chromium and
molybdenum loss in the matrix the corrosion resistance was improved.
Similarly, sulfide stress-corrosion cracking improvements were reported in a ti-

tanium stabilised supermartensitic steel containing austenite. It is believed that
austenite1 absorbs hydrogen from martensite and thus contributes to lower its con-
centration [32, 33]. The hardness reduction caused by austenite presence may, how-
ever, be another explanation. Consequently, austenite in supermartensitic stainless
steels does not have harmful effects on corrosion resistance.
1
Austenite has a higher solubility for hydrogen than martensite.
1.3.c Ferrite
In general the presence of δ-ferrite in the microstructure of martensitic stainless
steels is undesirable. The effects of its presence on the properties are presented
below.
i Hardness
The hardness of ferrite in low carbon semi-ferritic and martensitic stainless steels
ranges between 150 and 200 HVN [43]. These values are typically 100-150 Vickers
lower than low carbon fresh martensite and therefore low carbon semi-ferritic stain-
less steels exhibit reduced yield and tensile strength compared with fully martensitic
steels [38, 44].
ii Impact toughness
Ferrite at low temperatures is notch sensitive and fails in a brittle manner. In
martensitic stainless steels, ferrite presence in excess of 10 % can result in an ap-
proximate 50 % reduction in impact values [11, 45].
iii Corrosion resistance

In early AISI type 410 and 420 grades, ferrite was found in the microstructure
as a result of variations in balance between ferrite and austenite stabilising elements
within the specification. Sulfide stress-corrosion cracking tests showed cracking to
occur in martensite rather than in ferrite. The reduction of ferrite volume fraction
by the addition of nickel to AISI type 410 steel did not lead to any improvement of
the sulfide stress-corrosion cracking resistance. Rather than acting to initiate cracks,
ferrite was reported to obstruct crack propagation and fail in a ductile mode. This
behaviour was attributed to the lower hardness of ferrite [46].
The effect that δ-ferrite has on the sulfide stress cracking resistance of tempered
unstabilised supermartensitic stainless steel parent materials was also recently re-
ported. Cracking occurred at the δ-ferrite/martensite interface due to the chromium
depletion associated with the precipitation of chromium carbo-nitrides during tem-
pering [47].
1.4 Summary 23
1.4 Summary
The metallurgy of martensitic stainless steels has been improved to respond to

the need for more cost-effective corrosion-resistant alloys for applications in oil and
gas transportation. The constituents and phases present in the microstructure of the
new alloys as well as the mechanical and corrosion resistance properties have been
reviewed. The tempered martensitic microstructure of the parent materials is highly
desirable to ensure corrosion resistance, high strength and good low temperature
toughness.
The next chapter concentrates on the welding of supermartensitic stainless steels
and on the properties of the welded joint.
Chapter 2
Welding and heat-affected zone

microstructures and properties
The ability to weld supermartensitic stainless steels, both in land based and in
offshore constructions, is critical to their successful exploitation. However, welding
significantly affects the optimised microstructure in the region next to the weld
metal, the so-called heat-affected zone (HAZ). As a result, the HAZ is the critical
part of the weldment.
In this chapter, the welding processes employed to weld supermartensitic stainless
steel pipes are first introduced, the different HAZ regions and their properties are
then presented and finally the methods to estimate δ-ferrite retention in welded
joints are reviewed.
2.1 Introduction to fusion welding
Welding in one of the most popular joining methods for steels. Fusion welding
involves the application of a heat source to melt the edges of two surfaces, with or
without the help of a filler wire. The mixing of molten material from the workpiece
and molten drops of filler material creates a weld pool, which when solidified forms
the joint between the two components. The method by which heat is generated in
order to fuse the base metal and filler wire defines the nature of the welding process.
The fusion welds of concern to this study are formed by arc welding which is now
described in some more detail.
2.1 Introduction to fusion welding 25
2.1.a Arc welding processes

Although there are numerous arc welding systems, their general configuration
remains the same. A power source generates an arc between an electrode and the
workpiece, creating a temperature which in some regions is in excess of 10,000 ◦ C.
All arc welding processes use a shielding system to protect the molten weld metal
from the air.
Gas Tungsten Arc (GTA) Welding

This process, also called tungsten inert gas (TIG), uses non-consumable electrodes
made from tungsten alloys because of their extremely high melting temperatures (∼
3000 ◦ C). Nevertheless, the electrode tip is usually cooled to ensure it remains solid.
Filler material is supplied by a rod or wire, which melts as it is fed close to the
electrode tip. The weld pool is shielded by an inert gas, typically argon or helium,
piped in around the electrode. The TIG process usually generates clean welds with
high standards of weld metal quality in all welding positions and can be mechanised.
A schematic diagram of the arc area and components of the TIG process is shown
in figure 2.1.
Figure 2.1: Schematic view of gas tungsten arc

welding [48].
2.1 Introduction to fusion welding 26
Metal Inert Gas (MIG) Welding

This process also called gas metal arc welding (GMAW) uses a continuous wire which
is fed through the weld torch (figure 2.2). The mode of liquid metal transfer from
the electrode to the base metal can be varied by choosing different types of gases.
All metals can be welded using gas mixtures based on argon or carbon dioxide.
This process gives high metal deposition rates and can be automated, but offers less
control than TIG.
Figure 2.2: Schematic view of manual gas metal

arc welding [48].
Manual Metal Arc (MMA) Welding

A consumable coated electrode is employed (figure 2.3). The coating provides ele-
ments, which act as arc stabilizers, generate gases and a slag cover to protect the
weld pool from the environment. The coating sometimes contains alloying elements,
which help determine the weld composition. The simplicity of the process makes
it popular, but the lack of control over the weld bead and possibly high inclusion
levels might not be satisfactory for critical applications.
2.1.b Variables associated with welding

The most important variables are the process, chemical composition of the weld
deposit, heat input, the initial temperature of the parent material (pre-heat), tem-
perature of the weld deposit during multirun welding (interpass temperature), and
2.2 Welding of supermartensitic stainless steels 27
Figure 2.3: Schematic view of manual metal arc

welding [48].
heat-treatment given after welding (post-weld heat-treatment). The type of joint

and the material thickness are also important parameters.
In any welding process, the weld thermal cycle is composed of rapid heating to a
high temperature followed by rapid cooling. The heating and cooling rate are both
governed by the amount of heat input, defined as :
ηIV
Net heat input (J m−1 ) = (2.1)
υ
where I is the current in Amperes, V is the voltage applied between the power
source terminal and electrode expressed in volts, υ is the travel speed of the heat
source in m s−1 and η the arc transfer efficiency. In most of arc welding processes
the efficiency is between 0.6 and 0.99 [49].
2.2 Welding of supermartensitic stainless steels
2.2.a Hydrogen-induced cold cracking

Due to the hardness and brittleness of martensite, early martensitic steels are
considered to be the most difficult family of stainless steels to weld. During weld-
ing, the transformation of austenite into martensite is accompanied by an abrupt
decrease in the solubility of hydrogen. The latter may result in supersaturation and
cause the liberation of molecular hydrogen at micro-defects in the HAZ or the weld
metal. This liberation in molecular form introduces high stresses that are liable to
initiate cracks at room temperature in hard and brittle structures [50].
The reduced carbon concentration of supermartensitic stainless produces a soft

and ductile martensite, which when welded with a low hydrogen-generating process,
renders the risk of cold cracking virtually to nil. This represents a considerable
improvement compared with the earlier higher carbon martensitic stainless steels.
Supermartensitic alloys are therefore said to have improved weldability.
2.2.b Welding consumables

In general, supermartensitic stainless steels are welded using either matching
electrodes or superduplex stainless steel weld metal (typical composition 25 Cr,
10 Ni, 3.5 Mo, 0.25 N wt%). Matching composition consumables, now commercially
available, may be selected when the weld metal has to be stronger than the parent
materials e.g. when installation is by reeling. On the other hand, superduplex
welding alloys are employed when the weld metal must have a better corrosion
resistance and toughness [51].
2.2.c The welded joint

As illustrated in figure 2.4, a welded joint can be divided into two major regions,
the fusion zone and the heat-affected zone. The fusion zone experiences temperatures
above the melting point of the material and represents both the deposited metal
and the parts of the base metal melted during welding. The heat-affected zone on
the other hand, represents the portion of the base metal where all microstructural
changes induced by welding occur in the solid state.
i HAZ regions
Ideally, the heat-affected zone should have the same homogeneous tempered
martensitic structure as the parent material. However the gradient of temperature
in the HAZ leads to a corresponding gradient of microstructure and therefore dif-
ferent HAZ regions can be identified. The regions due to a single-pass root weld
in a supermartensitic stainless steel have recently been characterised using optical
Fusion boundary Weld metal
Heat−affected zone
Figure 2.4: A fusion weld [52].
microscopy by Enerhaug et al. as follows (figure 2.5) [53] :
(i) Partially melted region, located adjacent to the fusion boundary. This is where
incipient melting of the newly formed δ-ferrite occurs on heating.
(ii) Region which transforms completely to δ-ferrite on heating with significant grain
growth, also called the coarse-grained HAZ (CG-HAZ).
(iii) High-temperature two-phase region characterised by partial transformation of

austenite to δ-ferrite on heating.
(iv) Region which completely transforms to austenite during heating.
(v) A low peak temperature two-phase region, characterised by partial transforma-

tion of the tempered martensite to austenite during heating.
However, as mentioned before, the HAZ represents the area where all phase
transformations occur in the solid state and therefore the partially melted region
should not be regarded as belonging to the HAZ. The other regions can be artifi-
cially classified into two categories depending on the maximum temperature they
experienced before cooling, the high temperature and the low temperature HAZ.
The high temperature HAZ (HT-HAZ) includes the prior δ-ferrite region and the
high temperature two-phase region while the low temperature HAZ (LT-HAZ) refers
2.3 HAZ properties 30
13 % Cr
(i) The partly melted region

1600 L+δ
(ii) The delta−ferrite region 1400 δ

Peak temperature
Temperature / C
(iii) The high temperature 2 phase region
o
1200
(iv) The austenite region γ δ+γ
1000
(v) The low temperature
2 phase region 800
α+γ
600
α
400 δ+γ+σ
α+σ
10 20 30 40
Distance from fusion line Chromium \ wt%
2 4 6 8 10 12
Nickel \ wt%
Figure 2.5: Schematic diagram illustrating structural changes that occur within heat-affected
zone of single-pass supermartensitic stainless steel weld [53].
to the remainder of the HAZ.
The work by Enerhaug et al. was published during the time scale of the present
project and is the only study where the presence of δ-ferrite is clearly shown in
the high temperature HAZ. Although the different HAZ regions were presented,
confusions were made between the partially melted and prior fully ferritic regions in
the discussion.
2.3 HAZ properties
The welded joint service behaviour relies to a large extent on the response of
the HAZ to the external environment and stresses. Recently, catastrophic fractures
or cracks have been reported in HAZ of supermartensitic stainless steels tested in

conditions representative of field environment [54, 55].
2.3.a Hardness
There exists a hardness profile in HAZ caused by the different microstructures
that develop during welding [53]. As figure 2.6 illustrates, from the parent mate-
rial the hardness first increases to reach a maximum level of typically 350-375 HV
and then tends to slightly decrease as the high temperature regions of the HAZ
are reached. The initial hardness increase is attributed to the formation of fresh
martensite in regions heated up above Ac1 , while the reasons for the decrease are
not documented but might be caused by grain size effect or the presence of δ-ferrite.
The hardness level of 350 HV is not suitable for optimum sulfide stress cracking
resistance but can be reduced by the use of expensive and often unpractical post-
weld heat-treatment (PWHT).
400
Vickers micro-hardness (100 g)
375
350
325
300
275
250
0 0.25 0.5 0.75 1 1.25 1.5
Distance from fusion line / mm
Figure 2.6: Micro-hardness distribution in single-pass TIG bead-in-groove weld HAZ in an un-
stabilised 12 Cr 6 Ni 2 Mo (wt%) alloy. Arc energy 0.35 kJ mm−1 [56].
When employed, PWHT are usually carried out at temperatures of the order of
650 - 700 ◦ C for 5 to 10 minutes [44]. At those temperatures there is a risk of partial
austenitisation and subsequent transformation to hard martensite in some regions of
the HAZ. To avoid this problem, it is advisable to perform PWHT at temperatures
just below Ac1 .
2.3.b Charpy toughness

In general, supermartensitic steels are capable of giving high HAZ toughness
values, although in most of the published work the quoted HAZ results probably
include some parent material and/or weld metal. Typical results for a 12 Cr 5 Ni
2 Mo (wt%) alloy indicate values well in excess of 200 J at -60 ◦ C.
i Effect of δ-ferrite
Since the presence of δ-ferrite in HAZ of supermartensitic stainless steels is not
established, its effect on impact toughness has never been published. However, the
reduced impact toughness of the HAZ of martensitic stainless steels containing sub-
stantial amount of δ-ferrite (semi-ferritic stainless steels) such as the one illustrated
in figure 2.7, is well documented.
The HAZ of semi-ferritic stainless steels can either be mainly ferritic with some
martensite or conversely mainly martensitic with some ferrite. When ferrite is the
major phase in HAZ, the presence of grain boundary martensite lowers the ductile to
brittle transition temperature. This is because austenite has higher solubility for car-
bon than ferrite and transforms to hard martensite during cooling. The martensite
then acts as an internal stress raiser and thus increases the probability of cleavage
of the surrounding ferrite [57, 58]. On the other hand, predominantly martensitic
HAZ have reduced toughness when ferrite is present even in small quantities [59].
ii Effect of austenite
The effect of the presence of austenite in HAZ of supermartensitic alloys on
impact properties is not yet established. However, in supermartensitic weld metals
the presence of finely dispersed films of austenite was shown to yield higher impact
values at subzero temperatures [60].
Figure 2.7: Optical macrograph showing large amount of δ-ferrite in high temperature heat-
affected zone of a multipass weld performed on a semi-ferritic stainless steel (0.01 C 12 Cr 0.5 Ni
in wt%). Picture from this work.
A study of thermally simulated HAZ carried out on a range of supermartensitic

stainless steels revealed that austenite was no longer present in HAZ regions cooled
from temperatures exceeding 1200 ◦ C [30]. Consequently, unless a specific intercrit-
ical post-weld heat-treatment is carried out, no retained austenite is to be expected
in the high temperature HAZ of single-pass welds.
2.3.c Grain size

A small grain size in HAZ is important for strength and toughness. The de-
pendence of the yield strength on grain size is given by the Hall-Petch relationship
[61]:
σy = σi + ky d−1/2 (2.2)
where d is the grain diameter, σy is the yield stress, σi is the friction stress opposing
the movement of dislocations in the grains and ky is a constant.
The derivation of the Hall-Petch equation relies on the formation of a dislocation

pile-up at a grain boundary, one which is large enough to trigger dislocation activity
in an adjacent grain. Yield in a polycrystalline material is in this context defined as
the transfer of slip across grains. A larger grain is able to accommodate more dislo-
cations in a pile-up, enabling a larger stress concentration at the boundary, thereby
making it easier to promote slip in the nearby grain [62]. For martensitic structures,
however, dislocation sources are found at grain boundaries, which is different from
Hall-Petch approach, which considers dislocation sources within individual grains.
In this case the increase in strength due to martensite lath size is given by :
σG 115(L̄)−1 (2.3)
where L̄ is the mean linear intercept taken on random sections and at random angles
to the length of any lath section.
There is no general mechanism by, which grain refinement improves toughness.

The argument for steels is that carbide particles are finer when the grain size is
small. Fine particles are more difficult to crack and any resulting small cracks are
difficult to propagate, thus leading to an improvement in toughness.
2.3.d Corrosion resistance

The corrosion performance of supermartensitic stainless steels HAZ is claimed to
match that of the parent material both with respect to general corrosion and sulfide
stress-corrosion cracking [44]. In practice, however, the HAZ suffers preferential
pitting corrosion compared to weld metal and parent material. Enerhaug et al.
found that the HAZ microstructure was not the major factor affecting the pitting
corrosion resistance but rather the nature and the thickness of the oxide film on the
steel surface [53]. Pitting was observed at a distance of about 5 to 7 mm from the
fusion line (low temperature HAZ) and it could be concluded that the presence of
δ-ferrite in HAZ did not necessarily reduce pitting corrosion resistance. However,
it should be noted that no PWHT was applied and as mentioned in chapter 1, δ-
ferrite was found to lower sulfide stress-corrosion cracking resistance in the tempered
condition. Therefore, at present it is not clear whether the presence of δ-ferrite
affects the HAZ corrosion resistance.
2.4 Ferrite prediction 35
2.4 Ferrite prediction
Since little work has been performed on δ-ferrite retention in supermartensitic

stainless steels HAZ, and as this phase is apparently undesirable, it is interesting
to know where and when it forms and to be able to estimate its content. This
section reviews some of the work that has been performed on δ-ferrite formation
and decomposition in martensitic stainless steels and then introduces the methods
available to predict δ-ferrite retention in weld metals and HAZ of a range of stainless
steels.
2.4.a Formation and decomposition of δ-ferrite

Ferrite forms in the HAZ of martensitic stainless steels during continuous heating
after the Ac4 temperature has been exceeded. Like, Ac1 and Ac3 , Ac4 temperature
increases with the employed heating rate but to a lower extent (i.e. 30 ◦ C difference
between heating rates of 8 and 300 ◦ C s−1 ) [63]. If the alloy has the capacity to
become fully ferritic, as is the case in the micrograph shown in figure 2.7, then
the ferritisation is finished at the Ac5 temperature. Above this temperature, grain
growth occurs until melting starts at the solidus temperature.
However, not all martensitic stainless steels have the capacity to become fully
ferritic. Indeed, a detailed study of δ-ferrite formation and dissolution has been
performed by Fenn on a relatively high carbon containing martensitic stainless steel
called HT9 (table 2.1) [63]. The microstructures of quenched thermally simulated
HAZ of this alloy were never found completely ferritic. Indeed, optical micrographs
showed prior austenite grain boundaries decorated with δ-ferrite in HAZ regions
close to the fusion line. The uncomplete ferritisation was apparently not due to
δ-ferrite dissolution since cooling rates slower than 2 ◦ C s−1 were required for ferrite
to decompose. As ferritisation is expected to be fast at elevated temperatures,
steel HT9 should have been fully ferritic before cooling. To check the latter point,
thermodynamic calculations have been performed using MT-DATA.
According to MT-DATA (figure 2.8), at temperatures above 1230 ◦ C δ-ferrite
nucleates and grows at the expense of austenite. At temperatures above 1410 ◦ C,
the liquid phase appears and reacts with austenite to produce δ-ferrite (peritectic
reaction) and shortly after δ-ferrite dissolves in favour of the liquid phase. Conse-
quently, consistent with Fenn’s observations steel HT9 could never be fully ferritised.
Material C Si Mn Cr Mo Ni Nb V W
Sandvik HT9 0.21 0.37 0.48 11.20 1.00 0.52 0.01 0.30 0.02
Table 2.1: Chemical composition of steel HT9 (wt %) [63].
100
Mass percentage of equilibrium phases
Liquid
Ferrite
Austenite
80
60
40
20
0
1200 1250 1300 1350 1400 1450 1500
o
Temperature / C
Figure 2.8: Results of MT-DATA equilibrium calculations performed on steel HT9.
The calculations were performed using plus and sub − sgte databases and by allowing
austenite, ferrite and liquid phases.
According to the recent description of the different heat-affected zone regions

of supermartensitic stainless steels by Enerhaug et al. (figure 2.5), supermartensitic
alloys have the capacity to become fully ferritic before melting [53].
In stainless steels, it is well known that the decomposition of δ-ferrite is cooling

rate dependent. Mundt and Hoffmeister and Singh et al. have studied the effect of
cooling rate on the temperature of the onset of austenite reformation from δ-ferrite
(Ar4 ) on initially fully ferritised duplex and austenitic stainless steels [64, 65]. As
the cooling rate was increased from 20 to 100 ◦ C s−1 , the Ar4 temperatures were de-
pressed by about 60 ◦ C. The reverse transformation therefore starts at temperatures
where substitutional elements diffusivity is reduced and consequently contribute to
uncomplete dissolution of δ-ferrite.
2.4.b Prediction of δ-ferrite retention in HAZ

i Physical modelling
A fundamental way of predicting δ-ferrite volume fractions consists of developing
kinetic models. Hemmer compiled such a model for δ-ferrite retention in HAZ of
electron beam welded duplex stainless steels [66]. The model accounts for both the
effects of chemical composition and weld thermal cycle on δ-ferrite retention and
grain size. The model was constituted of three parts. First, dissolution of austenite
during heating at temperatures above Ac4 (in the dual-phase regime), followed by
δ-ferrite grain growth, and eventually, δ-ferrite dissolution during cooling to room
temperature.
Due to the fast heating and cooling rates experienced during electron beam
welding, paraequilibrium1 was assumed in each part of the model. Thanks to the
large amount of literature published on the studied materials, the validity of the
model could be assessed and shown to compare well with experimental data. The
following conclusions could be drawn. Increasing nitrogen concentration raised the
temperature at which the microstructure was fully ferritic (Ac5 ) and thus led to
slower ferrite coarsening behaviour in HAZ. High Ac5 temperature combined with
smaller δ-ferrite grain size favoured austenite reformation during cooling.
1
Carbon and nitrogen diffuse more rapidly than substitutional elements hence under rapid
heating conditions, the latter elements are unable to partition during the time scale related to the
welding operation.
2.4.c Prediction of δ-ferrite retention in weld metals

Most of the work dealing with prediction of δ-ferrite retention in stainless steels
has been performed on weld metals. The following sections summarise the work of
practical relevance.
i Measuring ferrite content
A problem encountered with the empirical determination of δ-ferrite is the con-
siderable scattering of δ-ferrite values measured by different methods and in different
laboratories. In austenitic and duplex stainless steels, that problem has been reme-
died by the introduction of a magnetic method based on the measurement of the
force required to tear off a defined permanent magnet from the surface of the spec-
imen. As the content of the ferro-magnetic δ-ferrite increases, the tearing force
increases, while other not ferro-magnetic constituent such as austenite, carbides,
sigma phase and inclusions have no influence at all. Thus the tearing force is an
indirect measurement of δ-ferrite content. To indicate the difference compared to
other existing methods, the measured ferrite value was designated as ferrite number
(F N ) rather than ferrite percentage (F P ) [67]. Several relationships have been pro-
posed to relate ferrite numbers to ferrite volume fractions as for example equation
2.4 which has been derived in the 0 - 15 F P range [68, 69, 70].
F N = 1.7 × F P (2.4)
Generally, the various correlations agree that the F N overstates the ferrite per-
cent but they do not agree on the amount of overstatement.
The magnetic method of measuring ferrite content is, unfortunately not applica-
ble to martensitic stainless steels since both martensite and ferrite are ferromagnetic.
Therefore ferrite measurement in martensitic stainless steels is quantified using stan-
dard point counting methods as in ASTM E562-99 or image analysis.
ii Ferrite factor
The Kaltenhauser Ferrite Factor (F F ) is a compositional relationship which is
commonly employed to assess the tendency of stainless steels base metal to retain
ferrite [71] :
F F = Cr + 6Si + 8Ti + 4Mo + 2Al + 4Nb − 2Mn − 4Ni − 40(C + N) wt%
The relationship advanced in the late sixties was initially developed to yield a
measure of the residual austenite content in ferritic stainless steel welds. The use of
this relationship has been extended to martensitic stainless steels. However, Hewitt
clearly showed the discrepancy associated with the use of the ferrite factor in the
case of low carbon martensitic stainless steels (figure 2.9) [72].
100
Martensite / %
80
60
40
20
5 10 15 20 25 30
Kaltenhauser ferrite factor

Figure 2.9: Percentage martensite versus ferrite factor [72].
The transition below the fully martensitic (F F ≤ 5) and the fully ferritic mi-
crostructure (F F ≥ 25) occur over a long ferrite factor range and show considerable
scatter. As far as heat-affected zones are concerned, it has been reported that the
transition occurs on a shorter ferrite factor range i.e. between 8 and 12 [73].
iii Diagrams
The Schaeffler diagram (Figure 2.10) developed in the fifties offers a convenient
way of representing the effect of alloy composition on room temperature weld metal
microstructure and phase volume fraction [74]. Phase boundaries of austenite, δ-
ferrite and martensite are plotted in terms of the chromium and nickel equivalents.
The chromium equivalent has been empirically determined to take into account the
effect of the common ferrite forming elements. Similarly, the nickel equivalent does
the same for the austenite formers. The ratio between the two thus describes the
proportions of each phase present in an alloy.
30
5%
eq
10%
Nickel equivalent (Ni)
A
20 20%
A+M 40%
A+F 80%
10
M A+M
+F
F+M M+F
0
0 10 20 30 40
Chromium equivalent (Cr)eq
(Cr) = Ni + 30 C + 0.5 Mn (in wt%)

eq
(Ni) = Cr + Mo + 1.5 Si + Nb (in wt%)
eq
Figure 2.10: The Schaeffler diagram. F, M and A represent ferrite, martensite and austenite,
respectively [11].
The Schaeffler diagram was improved upon by addition of a nitrogen contribution

in the Delong diagram, which has in turn been further improved for modern steels
by the WRC-1988 and WRC-1992 diagrams [67, 76, 77]. Those diagrams made
predictions in terms of ferrite number (F N ) instead of ferrite percent (F P ). Further
research devoted to the ferritic/martensitic region led more recently to the Balmforth
et al. diagram [43, 75]. The latter diagram (figure 2.11) gives iso-ferrite contour maps
in volume-percent ferrite and represents a significant improvement for prediction in
the ferrite plus martensite region. Table 2.2 lists the modification brought to the Cr
and Ni equivalent formulae and table 2.3 gives the composition range of confidence
6 A+M+F
Ni + 35 C + 20 N (in wt%)
5 M 10% 30% 50% 70% 90%
20%40% 60% 80%
4
3
F+M
2
F
1
0
8 10 12 14 16 18 20 22 24
Cr + 2 Mo + 1.5 Si + 10 (Al + Ti) (in wt%)
Figure 2.11: The new ferritic-martensitic diagram with iso ferrite lines in volume % [75].
for the Balmforth et al. diagram.
All diagrams offer a practical way of quantitatively estimating the room temper-
ature microstructure of weld metals, however the effect that cooling rates can have
on the phase balance are not accounted for.
iv Neural networks
Neural network models developed using both the WRC92 database and more
recent data are available since 2001. These models can be used to predict more
accurately the ferrite number of austenitic weld metals as a function of chemical
composition [78, 79, 80]. Although the databases contain information mostly on
austenitic steels, the use of the models can in principle be extended to martensitic
stainless steels. Chapter 6 will investigate the capacity of one of those models to
make predictions and comparisons between different supermartensitic stainless steels
alloys.
Source Year Cr-Equivalent (wt%) Ni-Equivalent (wt%)
Schaeffler 1949 Cr+Mo+1.5Si+0.5Nb Ni+0.5Mn+30C

Delong et al. 1956 Cr+Mo+1.5Si+0.5Nb Ni+0.5Mn+30(C+N)
Kaltenhauser 1971 Cr+6Si+8Ti+4Mo+2Al 4Ni+2Mn+40(C+N)
WRC-1992 1992 Cr+Mo+0.7Nb Ni+0.25Cu+35C+20N
Balmforth et al. 2000 Cr+2Mo+10(Al+Ti) Ni+35C+20N
Table 2.2: Evolution of the chromium and nickel equivalent formulae [43].
Element Composition range
Cr 11 - 30
Ni 0.1 - 3.0
Si 0.3 - 1.0
C 0.07 - 0.2
Mn 0.3 - 1.8
Mo 0 - 2.0
Al 0 - 0.3
Ti 0 - 0.5
N 0 - 0.25
Table 2.3: Composition range of confidence for the new ferrite plus martensite diagram [75].
2.5 Summary 43
2.5 Summary
Due to their low carbon concentrations, supermartensitic stainless steels can

reliably be arc welded. The service behaviour of the welded joint relies to a large
extent on the microstructure developing in the HAZ. In this regard, the properties
are likely to be altered by the presence of δ-ferrite in any marked extent in the
high temperature HAZ. However, limited information is available on the presence of
δ-ferrite and on the effects of the welding variables on its stability in the new steels.
On the other hand, fundamental modelling of δ-ferrite retention in HAZ is de-
pendent upon the availability of reliable thermodynamic data and the capacity to
predict weld thermal cycles.
Chapter 3
Experimental techniques
In this chapter the different techniques and equipment that have been employed in
experiments are introduced.
3.1 Imaging
3.1.a Optical microscopy

To reveal the microstructures, specimens were first mounted in conductive bake-
lite then the newly formed surfaces were ground on SiC paper to grit 1200 and
polished with a 1 µm cloth coated with diamond paste. The samples were finally
etched using one of the three etchants listed in table 3.1. The use of different etchants
was necessary to highlight specific features and phases and thus obtain complemen-
tary information on microstructures. Digital photographs were taken using a Leica
DMR digital camera.
3.1.b Scanning Electron Microscopy

The fracture surfaces of Charpy specimens and the HAZ of a variety of welds
were observed using a JEOL 5800 scanning electron microscope (SEM). Secondary
electron mode was employed in each case and the machine was operated at an
accelerating voltage of 15 kV.
3.1 Imaging 45
Name Composition Etching procedure Specificity
Villela’s 1 g C6 H3 OH(NO2)3 Immersion of sample Reveals martensite

[81] 5 ml HCl for 20 to 30 s
100 ml C2 H6 OH
Sulfuric 20 ml H2 SO4 Immersion of sample Reveals δ-ferrite

[81] 0.01 g NH4 CNS and electrolytic etching but not martensite
80 ml H2 O at 4 V for 20 s
Stock 1 g K2 S2 O5 Immersion of sample Reveals δ-ferrite

[82] 20 ml HCl until specimens in blue and martensite
80 ml H2 O becomes red yellow/brown
Table 3.1: Compositions and conditions of use of etchants.
3.1.c Image analysis

The volume fractions of δ-ferrite in the various specimens have been determined
using SigmaScan Pro 5.0 software using the following procedure. A series of digital
micrographs obtained using either electron or light microscopy were taken from
representative areas of the specimens. The images were first enhanced to use the
full intensity spectrum and the total number of pixels in the image was determined.
To highlight δ-ferrite, the intensity threshold method was employed and the sum of
pixels constituting the selected features was performed. The ratio of the latter sum
with the original number of pixels led to the determination of the volume fraction.
For the image analysis technique to be effective, pictures have to be taken under
the same conditions of brightness and contrast and the δ-ferrite should have the same
range of intensity throughout the image. Every effort was made to keep the lighting
condition constant but due to the different etching responses of the materials this
was not always achievable.
3.2 Chemical analysis 46
To estimate the uncertainty associated with this method, three measurements

were performed on each image. The most likely intensity range corresponding to
δ-ferrite was chosen in the first measurement, while in the two other the apparent
minimum and maximum intensity ranges were chosen. This method was applied to
two or three pictures of the area of interest and allowed measurement of the volume
fraction of δ-ferrite and estimation of the associated error.
3.1.d Transmission Electron Microscopy

Some specimens were observed using transmission electron microscopy (TEM)
to identify retained austenite in the parent materials and δ-ferrite in the HAZ. Thin
foils were examined using a JEOL 200 CX TEM operated at 200 kV.
Sections were cut to create discs of 3 mm in diameter and approximately 50 µm

thickness. The discs were electropolished in a solution of 5 % perchloric acid, 15 %
glycerol and 80 % ethanol at an applied voltage of 50 V at room temperature.
Selected area diffraction could be used to distinguish austenite (f.c.c.) from

low carbon martensite (b.c.c.). However, δ-ferrite and martensite could not be
distinguished using diffraction since they both have the b.c.c. structure. This was
however not a problem because δ-ferrite had a characteristic morphology that can
be easily identified.
3.2 Chemical analysis
3.2.a EDX microanalysis

To examine potential partitioning of alloying elements some micro-analytical ex-
periments were carried out using an energy dispersive X-ray (EDX) analysis system
on a CAMSCAN CS144 scanning electron microscope. ZAF (atomic number, ab-
sorption, fluorescence) corrections were employed, the acquisition time was set to
100 s and the dead time was kept below 20 %.
3.2.b X-ray analysis

The volume fractions of retained austenite (Vγ ) in the different specimens were
measured using X-ray diffraction with Cu Kα radiation at 40 kV and 40 mA. A
Phillips diffractometer was used with step scan mode to cover the angular 2θ range
from 47.1 to 103 ◦ . The 2θ step size was 0.05 ◦ with a dwell time of 30 s. Samples
were ground and polished using the normal metallographic preparation procedure
with two cycles of polishing and etching before the X-ray diffraction test to remove
any deformed layer caused by polishing.
The volume fraction of austenite could be estimated from measurements of the

integrated intensities of martensite1 and austenite phases assuming they are the only
phases present. The ratio of the intensities of diffraction peaks from two phases of
a polycrystalline sample is given by equation 3.1 [83]:
Iγ (hkl) Rγ (hkl) Vγ
= × (3.1)
Iα (hkl) Rα(hkl) Vα
where Iγ (hkl) and Iα (hkl) are the integrated intensity from a given plane (hkl) from
the austenite and martensite (or ferrite) phase respectively. Vγ and Vα are the
volume fractions of austenite and martensite respectively, and Rγ (hkl) and Rα (hkl)
are given for a specific peak by :
1
R= [|F |2 (p)(Lp )]e−2m (3.2)
v2
where :
v = volume of unit cell,
F = structure factor,
p = multiplicity factor,
Lp = Lorentz-polarization factor,
e−2m = temperature factor.
Considering that all materials in reality exhibit crystallographic texture, the

average integrated intensity for at least three specific reflections for austenite and
1
Low carbon martensite and ferrite have similar structure.
martensite (table 3.2) were taken into account (equation 3.3 and 3.4).
Phase i Diffracting plane (hkl )
Martensite 1 200
Martensite 2 211
Martensite 3 220
Austenite 1 200
Austenite 2 220
Austenite 3 311
Table 3.2: Diffracting hkl planes used for martensite and austenite.
1 Iiγ
3
Iγ
= (3.3)
Rγ 3 i=1 Rγi
1 Iiα
3
Iα
= (3.4)
Rα 3 i=1 Rαi
The value of Vγ /Vα in equation 3.1 can be obtained from the measurement of
Iγ /Iα and the calculation of Rγ /Rα . Once Vγ /Vα is found, the value of Vγ can be
obtained from the additional relationship :
Vγ + Vα = 1 (3.5)
The 2θ values for three austenite peaks were used to calculate the d spacings with
Bragg’s law and then the lattice parameter a. These values were plotted against
sin2 θ. An accurate value of aγ was found by extrapolation to sin2 θ =1 [83].
3.3 Mechanical properties 49
3.3 Mechanical properties
3.3.a Hardness
Hardness measurements were made using a Vickers pyramidal hardness-testing
machine using a load of 10 kg. The indent had diagonals of typically 200 µm. When
the width of the observed region was smaller than the latter figure, micro hardness
was employed using a load of 100 g. Indents this time had diagonals of about 25 µm.
Ten measurements were taken over the metallographic specimen area regardless of
the technique employed to measure the hardness.
3.3.b Charpy impact toughness

The Charpy impact toughness test was used to characterise the ability of different
microstructures to absorb energy during the process of fracture. A pre-notched
standard sample of 10 mm square section × 55 mm with a 2 mm deep V-notch
positioned through-thickness at the middle of the specimen is fractured using a
weighted pendulum hammer [62]. To reveal any ductile-brittle transition the tests
were conducted in the range 20 ◦ C to -200 ◦ C and were performed according to
British Standard 10045-1 [84].
3.4 Dilatometry
Two dilatometers have been used during this project to attempt to measure
phase transformation temperatures over a range of heating rates.
3.4.a Thermomechanical simulator

The “Thermec-mastor” thermomechanical simulator allows the computer-controlled
application of heat-treatments and deformation to a specimen of material. The tem-
perature and diametrical dilation of the specimen are measured and thus the progress
of phase transformations within the material can be followed.
The experiments were performed in a sealed chamber (figure 3.1) which is evac-
3.4 Dilatometry 50
uated (to a vacuum of ∼ 102 Pa) to prevent oxidation effects. The solid cylindrical
specimen with an 8 mm diameter and a length of 12 mm seats centrally between
two silicon nitride dies. The die holders are insulated from the dies by two mica
discs. The temperature is recorded using a platinum / platinum-10 % rhodium ther-
mocouple spot-welded in a central position to the specimen. The accuracy of the
temperature reading is ± 3 ◦ C whilst the variation along the specimen length is no
more than ± 10 ◦ C.
The heating of the sample is by a water cooled, high frequency induction coil
which surrounds the sample. The maximum heating rate that can be achieved is
around 50 ◦ C s−1 but is also dependent upon the material. The cooling of the sample
is controlled via helium jets.
The diametrical dilation of the sample is monitored using a He-Ne laser beam.
The beam is positioned at the same height as a thermocouple spot-welded onto the
sample to ensure that the temperature and dilation measurements are from the same
location. The accuracy of the dilation reading is ± 1 µm. Time, temperature and
dilation are recorded and are stored on a computer for subsequent analysis.
3.4.b Netzsch dilatometer

Dilatometric measurements were also carried out using a Netzsch 402C dilatome-
ter capable of operation between 25 and 1600 ◦ C. Details of the instrument are shown
in figure 3.2. The heating and cooling of the sample are ensured by a sealed furnace,
which surrounds the specimen holder. The temperature of the specimen is measured
by a Pt/Pt-Rh thermocouple located about 1 mm above its surface. To limit the
temperature difference between the furnace and the specimen, it is essential to use
slow heating rates (1 ◦ C s−1 maximum). During the experiment, the longitudinal
dilation is measured using a pushrod connected to a linear velocity displacement
transducer (LVDT). Time, temperature and dilation are recorded and are stored on
a computer for subsequent analysis.
To allow rapid transmission of the heat, 12 mm long hollow cylinders of 8 mm

3.5 Weld heat-affected zone simulator 51
Hydraulic
ram
Si 3 N 4
die
Induction
heating coil
Scanned
laser Laser
beam detector
Test Si 3 N 4
specimen
die Thermocouple
Figure 3.1: The arrangement of equipment within the Thermec-mastor thermomechanical sim-
ulator.
external diameter and 5 mm internal diameter have been used. Tests were carried
out using a static argon atmosphere to prevent oxidation of the specimens.
3.5 Weld heat-affected zone simulator
3.5.a Gleeble
The Gleeble is a thermomechanical testing device that is used in a wide range of
applications including simulation and testing of weld heat-affected zones. A major
advantage of the Gleeble is that it generates large volumes of microstructure that
simulate small, hard to study region in actual weldments. A range of specimen ge-
ometries can be used, but by using bars, Charpy specimens can directly be machined
after simulation, thus enabling the toughness of any region of the heat-affected zone
to be measured.
The machine (figure 3.3), is interfaced to a computer that is readily programmed

Pushrod Thermocouple Specimen
Argon atmosphere
To LVDT
Furnace Sample holder
Figure 3.2: The arrangement of equipment within the Netzsch 402C dilatometer. LVDT : linear
velocity displacement transducer.
to provide reference signals for close loop control of the applied thermal cycle. Heat-
ing is accomplished by the flow of low-frequency alternating current in the specimen.
The current distribution is controlled by the cross section of the specimen and the
jaws used to mount it. The feedback signal necessary for closed-loop control is ob-
tained from a thermocouple welded to the specimen surface [85].
Thermal and electrical

contact to jaws Pt/Rh Thermocouple Water cooled jaws
Heating
current
Bar shaped specimen
Figure 3.3: The arrangement of equipment within the Gleeble simulator.

Chapter 4
Characterisation and predictions
The aim of the work presented in this chapter was first to characterise the as-received
microstructure of the supermartensitic stainless steels studied in this thesis, then
determine their phase transformation temperatures and finally compare the latter
with equilibrium and paraequilibrium predictions made using MT-DATA software.
4.1 Studied materials
Three supermartensitic stainless steels covering the contemporary range of com-

position have been studied in this work. Steels A, B and C whose chemical com-
positions are reported in table 4.1 consisted of seamless pipes of different diameters
and wall thicknesses. Steel A is a highly alloyed supermartensitic stainless steels
stabilised by the addition of titanium. Steels B and C are lean and medium alloyed
compared to steel A and also contain less titanium. The chromium concentration
varies between 10.9 and 12.5 wt%, but the most significant differences between the
three alloys was their nickel and molybdenum concentrations.
The details of manufacturers, designation and dimensions are listed in table 4.2.
No information could be gathered about the production route and the few data
obtained about heat-treatments are reported in table 4.3. Unfortunately, in most
cases the solution treatment times and temperatures were not specified.
4.1 Studied materials 55
Steel name C Si Mn Ni Cr Mo Cu Ti N
A 0.01 0.26 0.46 6.46 12.20 2.48 0.03 0.09 0.0070

B 0.01 0.18 1.14 1.55 10.90 ≤ 0.01 0.49 0.01 0.0060
C 0.01 0.19 0.24 3.12 12.50 0.02 0.06 0.01 0.0120
Table 4.1: Chemical compositions of steels A, B and C in wt%.
Steel name Manufacturer Designation Dimension (mm)
A Sumitomo SM 13CRS 80 273.1 (OD) × 12.0 (WT)

B Kawasaki KL 12CR 168.3 (OD) × 11.0 (WT)
C NKK NK 13CRW 273.0 (OD) × 15.9 (WT)
Table 4.2: Details of steels A, B and C. OD: outer diameter, WT: wall thickness.
4.1.a Initial microstructures

As all materials received different heat and mechanical treatments, it was im-
portant to characterise their as-received state before producing any welds. The
initial microstructures of the materials, electrolytically etched, are shown in figure
4.1. They all consisted of tempered martensite with no evidence of δ-ferrite and
some cube shaped titanium carbo-nitride particles in steel A. The absence of ferrite
indicated that all steels could be made completely austenitic at the austenitising
temperature. Steels B and C (molybdenum free alloys) showed some corrosion pits
on their surfaces after etching. The presence of the latter reflected their reduced
corrosion resistance compared with steel A.
4.1.b Retained austenite

The presence of any retained austenite was also investigated. X-ray diffraction
tests were carried out according to the procedure described in chapter 3. Those
Figure 4.1: Optical micrographs of the initial microstructures of steels A (a), B (b) and C (c),
electrolytically etched.
Steel name Heat-treatment
A Quenched from 900 ◦ C and tempered at 640 ◦ C

B Not reported
C Air cooled and tempered
Table 4.3: Information about heat-treatments provided by steels supplier.
tests revealed that unlike steels B and C (figures 4.2 and 4.3), steel A contained a
significant amount of retained austenite (figure 4.4). The volume percentage of the
latter was 13 % within an error of ± 0.5 %.
140
α (211)
120
Square root of intensity
α (200)
100
80
α (220)
60
40
20
0
45 60 75 90 105
2θ
Figure 4.2: X-ray diffraction pattern of steel B in as-received condition.
Figure 4.5 shows that this austenite remained stable after immersion in liquid
nitrogen for 30 minutes, but after reaustenitisation at 1000 ◦ C for 30 minutes and
water quenching no austenite could be detected. Those results indicated that the
presence of retained austenite in the as-received condition of steel A was the conse-
quence of the prior tempering treatment. In fact tempering must have been carried
140
α (211)
Square root of intensity 120
100 α (200)
80
α (220)
60
40
20
0
45 60 75 90 105
2θ
Figure 4.3: X-ray diffraction pattern of steel C in as-received condition.
140
120
α (211)
100
80 α (200)
60 α (220)
γ (200)
40
γ (220) γ (310)
20 γ (222)
0
45 60 75 90 105
2θ
Figure 4.4: X-ray diffraction patterns of steel A in as-received condition.

140 30 min in liq N. α (211)

o
1000 C 30 min + WQ.
120

α (200)
100
80 α (220)
60 γ (200)
γ (220) γ (310)
40
γ (222)
20
0
45 60 75 90 105
2θ
Figure 4.5: X-ray diffraction patterns of steel A after indicated thermal treatment (WQ: water
quench).
out few degrees above the steel Ac1 temperature. In this case, as introduced in chap-
ter 2, the new austenite is significantly enriched in carbon and nickel and remains
stable after cooling.
A TEM examination of a thin foil prepared on steel A showed this retained

austenite to be present as small elongated grains located at martensite lath bound-
aries as illustrated on figure 4.6.
Within the limits of experimental error, neither carbides nor nitrides were de-
tected in the three alloys over the angular range investigated in the X-ray diffraction
experiments. A part from Ti(C,N) particles no other carbo-nitrides could be ob-
served during the TEM examination of steels A, B and C.
4.1.c Grain size and hardness

The materials had similar initial grain sizes but significantly different hardnesses
(table 4.4).
Figure 4.6: (a) Bright field image of a martensite plus retained austenite region. (b) Dark field
image of the same region as (a). (c) Selected area diffraction pattern of austenite corresponding
to the dark field image.
4.2 Dilatometric measurements 61
To measure the materials maximum hardness, samples were austenitised at

1000 ◦ C for 30 minutes and water quenched to obtain fully martensitic microstruc-
tures. After heat-treatment all materials had similar hardness levels but the hardness
increase of steel A was significantly smaller than that of steels B and C (table 4.5).
The latter was due to the virtually carbon free martensite of steel A as a result of
the stabilisation by titanium.
Steel name Grain size / µm Hardness HV10kg
A 25 ± 4 296 ± 3
B 13 ± 3 219 ± 1
C 17 ± 4 252 ± 5
Table 4.4: Initial hardnesses and grain sizes of steels A, B and C.
Steel name Fresh martensite hardness HV10kg Hardness change
A 317 ± 8 + 21
B 323 ± 3 + 104
C 322 ± 6 + 70
Table 4.5: Hardness of steels A, B and C after 1000 ◦ C 30 minutes and water quench, and
hardness change compared with as-received condition (table 4.4).
4.2 Dilatometric measurements
Very few data have been published on the phase transformation temperatures
of supermartensitic stainless steels. In this regard, the temperatures of δ-ferrite
formation and completion as well as dissolution have never been measured experi-
mentally. However to understand the microstructures developing in HAZ it is useful
to determine those temperatures.
The phase transformation temperatures presented in this section have been ob-
tained using the Netzsch DIL 402 dilatometer presented in chapter 3, employing
slow heating and cooling rates of 10 ◦ C min−1 .
4.2.a Phase transformations on heating

The detection of phase transformation within the steels using dilatometry relies
upon the fact that the phases have different densities. Austenite formation is associ-
ated with an overall contraction of the specimen. Conversely δ-ferrite appearance is
associated with an overall expansion of the specimen corresponding to the transfor-
mation of the f.c.c. lattice into the b.c.c. lattice. In practice, the Ac1 temperature
is the one at which the dilation curves starts to deviate from the straight line in the
temperature range 500 - 800 ◦ C while the deviation corresponding to Ac4 is expected
at a higher temperature.
The heating part of the three dilation curves plotted versus temperature is re-
ported in figure 4.7 and the measured Ac1 and Ac3 temperatures are listed in table
4.6.
Alloy Ac1 Ac3 Ac1 -Ac3
A 554 719 165

B 678 792 114
C 650 798 148
Table 4.6: Measured Ac1 , Ac3 temperatures and Ac1 -Ac3 temperature intervals in (◦ C) of steels
A, B and C, using a heating rate of 10 ◦ C min−1.
Steel A (6.4 wt% Ni) had the lowest Ac1 temperature followed by steels C (3.1
wt% Ni) and B (1.5 wt% Ni). Nickel is well known to depress the Ac1 tempera-
ture and this was confirmed experimentally. Steel A also had the largest Ac1 -Ac3
temperature interval.
An unusual and reproducible contraction could be observed at around 900 ◦ C on

the dilation curve of steel A. This phase transformation is believed to be associated
with the dissolution of titanium carbides. However, considering the small volume
fraction of these particles it would be expected to register anything on the dilation
curve. MT-DATA calculations on steel A (showed later in this chapter) are however
consistent with this assumption since it predicted TiC dissolution at ∼ 900 ◦ C under
equilibrium conditions.
A magnified representation of the high temperature part of the dilation curve is

shown in figure 4.8. The measured phase transformation temperatures and temper-
ature ranges are reported in table 4.7. The Ac4 temperatures all exceeded 1300 ◦ C
and the temperature intervals of δ-formation (Ac4 -Ac5 ) were shorter than 50 ◦ C. On
the other hand, steel A combined the highest Ac4 and Ac5 temperatures and had
its Ac4 -Ac5 temperature range about twice as long as steels B and C. Finally, as
expected, all steels had large fully austenitic range (Ac3 -Ac4 ) of about 650-700 ◦ C.
Alloy Ac4 Ac5 Ac4 -Ac5 Ac3-Ac4
A 1372 1417 45 653

B 1330 1352 22 652
C 1347 1373 26 697
Table 4.7: Measured phase transformation temperatures and temperature intervals in (◦ C) of

steels A, B and C, using a heating rate of 10 ◦ C min−1 .
As Fenn reported, the Ac4 temperature is less sensitive to heating rate than Ac1
and Ac3 , therefore the temperatures determined under slow heating condition can
be expected to be close to the actual temperatures during welding [63]. As interest
is placed on the presence of δ-ferrite in HAZ, the short Ac4 -Ac5 temperature range
indicates that the HAZ region over which δ-ferrite forms is narrow compared with
the other HAZ regions.
2.5
Steel A Heating -->
Steel B
Steel C
1.5
dl/lo / %
0.5
0
100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400
Temperature / o C
Figure 4.7: Heating part of the dilation curves obtained on steels A, B and C for a heating rate
of 10 ◦ C min−1 .
Steel A Heating -->

Steel B
2.4 Steel C
2.2
dl/lo / %
1.8
1.6
1.4
1200 1250 1300 1350 1400 1450
o
Temperature / C
Figure 4.8: Magnified representation of the high temperature part of figure 4.7.
4.2.b Phase transformations on cooling

Figure 4.9 shows the high temperature cooling part of the three dilation curves
plotted versus temperature. Under slow cooling condition all fully ferritised steels
retransformed to austenite. The temperatures of the onset and completion of δ-
ferrite dissolution on cooling, Ar4 and Ar5 , which correspond to the change from
the b.c.c. lattice to the f.c.c. lattice are reported in table 4.8.
Steel name Ar4 Ar5 Ar4 -Ar5
A 1388 1350 38
B 1265 1205 60
C 1285 1225 60
Table 4.8: Measured phase transformation temperatures and temperature intervals in (◦ C) of

steels A, B and C using a cooling rate of 10 ◦ C min−1 .
Steel A <-- Cooling

Steel B
2.4 Steel C
2.2
dl/lo / %
1.8
1.6
1.4
1200 1250 1300 1350 1400 1450
o
Temperature / C
Figure 4.9: Cooling part of the dilation curves obtained on steel A, B and C for a cooling rate
of 10 ◦ C min−1 .
The reverse transformation temperatures of steel A were higher than those of

steels B and C. It was also noted that steels B and C had similar Ar4 temperatures
and transformed to austenite over a longer temperature range (60 ◦ C compared with
38 ◦ C for steel A).
Finally the MS and MF temperatures of the three alloys (figure 4.10) are reported
in table 4.9. As could be expected from the X-ray results, all MF temperatures
were above room temperature. Consistent with its relatively large alloying content,
steel A, had the lowest MS and MF temperatures. Once again, molybdenum free,
steels B and C, had similar phase transformation temperatures. A summary of all
experimental phase transformation temperatures is presented in figure 4.11.
Steel name MS MF
A 205 105
B 410 230
C 420 250
Table 4.9: Measured MS and MF temperatures (in ◦ C) of steels A, B and C using a cooling rate
of 10 ◦ C min−1 .
4.2.c Thermecmastor
The thermecmastor was also employed to attempt to determine the Ac4 and Ac5
temperatures in the 5 - 50 ◦ C s−1 heating rate range. However due to unstability
problems of the machine at elevated temperatures many experiments failed, or when
the desired temperatures were reached the data were too noisy to allow the detection
of the expansion associated with δ-ferrite formation. Therefore, this dilatometer was
not suitable for our purpose.
Steel A <-- Cooling

Steel B
0.4 Steel C
0.2
dl/lo / %
-0.2
-0.4
50 100 150 200 250 300 350 400 450 500 550
Temperature / o C
Figure 4.10: Low temperature cooling part of the dilation curves obtained on steels A, B and C
for a cooling rate of 10 ◦ C min−1.
Figure 4.11: Summary of experimental phase transformation temperatures measured on steels

A, B and C. Heating and cooling rate of 10 ◦ C min−1.
4.3 Thermodynamic predictions 68
4.3 Thermodynamic predictions
Having measured the phase transformation temperatures of the three steels un-
der condition close to equilibrium, it is now possible to assess the predictions of
thermodynamic software such as MT-DATA (National Physical Laboratory) on the
system of concern.
MT-DATA uses experimentally determined parameters to model thermodynamic

variables. For this reason, the reliability of the calculations it performs depends on
the parameters present in the databases. Mass conservation and a minimisation of
Gibbs free energy allow calculations of the phase transformation temperatures, the
volume fractions of phases present and their chemical compositions. As MT-DATA
uses thermodynamics, equilibrium or paraequilibrium1 are always assumed and no
kinetic information is produced.
4.3.a Equilibrium calculations

Calculations of the phase transformation temperatures and phase mass fractions
have been performed on the three steels over the temperature range 400 - 1600 ◦ C
at intervals of 25 ◦ C using the plus and sub-sgte databases.
Simpler systems are more accurately predicted by MT-DATA hence only the
elements in major quantities are taken into account in calculations. The latter were
made without silicon since MT-DATA has been found to give silicon an excessively
strong ferrite-forming tendency in the systems of concern. The most important
elements and phases and the most common carbides and nitrides have been allowed
and are listed in table 4.10.
1
Carbon and nitrogen diffuse more rapidly than substitutional elements hence under rapid
heating condition, the latter elements are unable to partition during the time scale related to the
welding operation.
Alloy Allowed elements Allowed phases
A Fe, C, Mn, Ni, Cr, Mo, Ti, N M23C6 , γ, δ, Fe3 C, M7 C3 , TiC, TiN, Liq
B Fe, C, Mn, Ni, Cr, Cu, Ti, N M23 C6 , γ, δ, Fe3 C, M7 C3 , TiC, TiN, Cu, Liq
C Fe, C, Mn, Ni, Cr, Cu, Mo, Ti, N M23 C6 , γ, δ, Fe3 C, M7 C3 , TiC, TiN, Cu, Liq
Table 4.10: List of elements and phases used for steels A, B and C in equilibrium calculations.
4.3.b Results
The mass fractions of the equilibrium phases of steels A, B and C are plotted
against temperature in figures 4.12 to 4.14 and the equilibrium phase transformation
temperatures are summarised in table 4.11.
Consistent with microstructural observations, all steels are predicted to have the
capacity to become fully austenitic. The materials are also expected to undergo
ferritisation at temperatures lower than 1300 ◦ C and become entirely ferritic before
melting starts (∼ 1500 ◦ C). On the other hand, it should be noted that Cu, M23 C6 ,
TiC and TiN were the only stable phases predicted by MT-DATA. Steel A was the
only alloy predicted to contain austenite down to 400 ◦ C, and to be free from M23 C6 .
The Ae4 -Ae5 and Ae5 -solidus temperature ranges corresponding to the regions
over which δ-ferrite forms and grows are listed in table 4.12. Steel A is predicted
to possess the largest Ae4 -Ae5 (159 ◦ C) and the smallest Ae5 -solidus temperature
interval (65 ◦ C). On the other hand, steels B and C are predicted to have comparable
Ae4 -Ae5 and also Ae5 -solidus temperature range.
4.3.c Comparison with dilatometric measurements

Predicted versus measured Ac3 , Ac4 and Ac5 temperatures are reported in fig-
ure 4.15. As could be expected, predicted phase transformation temperatures were
systematically lower than the measured ones. The Ae3 temperatures were between
14 and 45 ◦ C lower than the experimental values, but the Ae5 temperatures were
100

10
Liquid
Ferrite
Austenite
1 TiC
TiN
0.1
0.01
400 600 800 1000 1200 1400 1600
o
Temperature / C
Figure 4.12: Results of MT-DATA equilibrium calculations performed on steel A.
100
10
Liquid
Ferrite
Austenite
1 M23C6
Cu
TiN
0.1
0.01
400 600 800 1000 1200 1400 1600
o
Temperature / C
Figure 4.13: Results of MT-DATA equilibrium calculations performed on steel B.

Mass fraction of equilibrium phases 100
10 Liquid
Ferrite
Austenite
M23C6
1 Cu
TiN
0.1
0.01
400 600 800 1000 1200 1400 1600
o
Temperature / C
Figure 4.14: Results of MT-DATA equilibrium calculations performed on steel C.
Alloy Ae1 (5%) Ae3 (95%) Ae4 (5%) Ae5 (95%) Solidus (5%)
A ≺ 400 705 1237 1396 1461

B 692 750 1261 1350 1490
C 675 753 1242 1360 1492
Table 4.11: Equilibrium phase transformation temperatures (in ◦C) obtained after calculations
performed on steels A, B and C. The Ae1 , Ae4 and solidus temperatures have been obtained by
taking the temperature at which 5 % of the growing phase have formed. Similarly, the Ae3 and
Ae5 temperatures have been obtained by taking the temperature at which 95 % of the growing
phase have formed.
Alloy Ae4 - Ae5 Ae5 - Solidus
A 159 65
B 89 140
C 118 132
Table 4.12: Equilibrium interval temperatures (in ◦ C) between Ae4 -Ae5 and Ae5 -solidus, deduced
from table 4.11.
relatively well predicted by MT-DATA. However, poor agreement was found regard-
ing the Ae4 temperatures. The latter were between 70 to 135 ◦ C lower than the
actual temperatures. Although, there were significant differences between measured
and predicted temperatures, steel A was successfully predicted to have the longest
dual-phase field.
Unfortunately, the solidus temperatures could not be measured and consequently
no comparison can be made for the Ae5 -solidus temperature range.
4.3.d Effects of Ni and Mo on phase diagrams

Nickel and molybdenum play a major role in the metallurgy of supermartensitic
stainless steels, it is therefore interesting to study their effects on the extent of
phase stability fields. Based on a simplified composition (0.01 C, 12.2 Cr, 0.5 Mn
in wt%), two equilibrium pseudo-binary phase diagrams have been calculated using
MT-DATA. The latter calculations were performed in the 0-10 wt% nickel range for
0 and 2 wt% molybdenum.
In the molybdenum free diagram (figure 4.16), the Ni concentration range over
which the alloy has the capacity to become fully austenitic while keeping the ferritic
solidification mode is large i.e. between 0.5 and ∼ 6 wt%. However, the addition
of 2.5 wt% Mo significantly shortens this range and shifts it to higher nickel con-
centrations i.e. between 3.5 and ∼ 7.5 wt%. Therefore, consistent with Kondo et
al. observations, the addition of molybdenum reduces the size of the fully austenitic
domain [14].
As far as δ-ferrite is concerned, the austenite/δ-ferrite phase field is significantly

850
a)
Predicted Ae3 temperatures / C

o
800
Steel A
750 Steel B
Steel C
700
700 750 800 850
Measured Ac3 temperatures / oC
1400
b)
o
1350
1300
Steel B
1250 Steel A
Steel C
1200
1200 1250 1300 1350 1400
Measured Ac4 temperatures / oC
1450
c)
o
1400 Steel A
Steel C
1350 Steel B
1300
1300 1350 1400 1450
o
Measured Ac5 temperatures / C
Figure 4.15: Comparison between MT-DATA predicted Ae3 , Ae4 and Ae5 temperatures and
experimental temperatures measured by dilatometry.
Figure 4.16: Calculated pseudo-binary phase diagram for a 0.01 C, 12.2 Cr, 0.5 Mn (wt%) with
two different levels of molybdenum. Black curve corresponds to 0 Mo (wt%) and red curve 2.5 Mo
(wt %). For clarity, the liquid + ferrite and liquid + ferrite + austenite fields have not been
represented.
300
0 wt% Mo
1.5 wt% Mo
Ae4-Ae5 temperature interval / C
2.5 wt% Mo
o
250
200
150
100
50
0
0 1 2 3 4 5 6 7 8
Nickel / wt%
Figure 4.17: Equilibrium Ae4 -Ae5 temperature intervals as a function of nickel and molybdenum
concentrations. Calculations based on a simplified composition : 0.01 C, 12.2 Cr, 0.5 Mn (wt%).
enlarged by addition of molybdenum. For example a Mo free alloy with 4 wt% Ni is

predicted to have an Ae4 -Ae5 interval of 116 ◦ C while the same temperature interval
for a 2.5 wt% Mo alloys is 244 ◦ C (figure 4.17). The enlarging effect of Mo on the
dual-phase field could be noticed in the dilatometric tests. Indeed, steel A had its
Ac4 -Ac5 temperature interval twice as long as those of steels B and C, although the
measured value did not exceed 45 ◦ C.
Figure 4.18 shows the effect of nickel and molybdenum on the Ae5 -solidus tem-
perature interval. The addition of nickel strongly reduces the extent of the fully
ferritic range. Molybdenum, is however seen to expand this range, but its effect is
less strongly marked compared with figure 4.17.
300
0 wt% Mo
Ae5-Solidus temperature interval / C
1.5 wt% Mo
o
250 2.5 wt% Mo
200
150
100
50
0
0 1 2 3 4 5 6 7 8
Nickel / wt%
Figure 4.18: Calculated equilibrium Ae5 -solidus temperature intervals as a function of nickel and
molybdenum concentrations. Calculations based on a simplified composition : 0.01 C, 12.2 Cr,
0.5 Mn (wt%).
Consequently, molybdenum free supermartensitic stainless steels, like steels B

and C, tend to have short dual-phase field and long fully ferritic range, while the
opposite applies to molybdenum containing alloys like steel A.
4.3.e Paraequilibrium calculations

Under the fast heating rates encountered during welding, substitutional elements
might not have time to partition. Consequently, the concentration of substitutional
elements remain unchanged and the phase transformation temperatures and mass
fractions are affected. Calculations under paraequilibrium conditions are therefore
required. In practice, those calculations are made by specifying that the ratio of iron
to substitutional atoms to be constant at all temperatures. Carbon and nitrogen,
on the other hand, are allowed to redistribute.
In this part, the components and phases allowed for are listed in table 4.13.
Under paraequilibrium condition, phases such as TiN, TiC or Cu cannot be allowed.
However at elevated temperatures, TiN and TiC can be found to be stable under
equilibrium condition. To account for the loss of carbon and nitrogen tied up in
Ti(C,N) when these phases were predicted to be present under equilibrium condition,
the stoichiometric amounts of C and N were removed from the bulk composition and
then paraequilibrium calculations were performed.
Alloy Allowed elements Allowed phases
A Fe, C, Mn, Ni, Cr, Mo, N γ, δ-ferrite, Liq

B Fe, C, Mn, Ni, Cr, Cu, N γ, δ-ferrite, Liq
C Fe, C, Mn, Ni, Cr, Cu, Mo, N γ, δ-ferrite, Liq
Table 4.13: List of elements and phases used for steels A, B and C for paraequilibrium calcula-
tions.
4.3.f Results of paraequilibrium calculations

In the following, the temperature of the onset and completion of δ-ferrite for-
mation under paraequilibrium conditions are referred to as Ap4 and Ap5 . The Ap4 ,
Ap5 and solidus temperatures are reported in table 4.14.
For each steel, δ-ferrite formation is predicted to start at about 1300 ◦ C and
the temperature range of ferrite formation is not predicted to exceed 30 ◦ C. Steel A
is predicted to possess the shortest Ap4 -Ap5 and Ap5 -solidus temperature intervals
(table 4.15).
Alloy Ap4 (5%) Ap5 (95%) solidus (5%)
A 1322 1323 1477

B 1302 1310 1500
C 1295 1322 1496
Table 4.14: Paraequilibrium phase transformation temperatures (in ◦ C) obtained after calcula-
tion performed on steels A, B and C.
Alloy Ap4 - Ap5 Ap5 - solidus
A 1 154
B 8 190
C 27 174
Table 4.15: Paraequilibrium temperature ranges (in ◦ C) between Ap4 -Ap5 and Ap5 -solidus,
deduced from table 4.14.
Although the predicted temperature ranges of δ-ferrite formation (Ap4 -Ap5 ) were
closer to those determined using dilatometry, the comparisons are not valid. This is
because the dilatometric experiments were carried out under slow heating conditions
(10 ◦ C min−1 ) and therefore partitioning of substitutional elements had time to occur
at elevated temperatures, whereas under paraequilibrium condition redistribution of
such solutes is forbidden.
4.3.g Comparison with equilibrium temperatures

A comparison between tables 4.11 and 4.14 indicates that the Ap4 temperatures
were between 27 and 58 ◦ C higher than the equilibrium temperatures and the Ap5
temperatures were between 40 and 73 ◦ C lower than the corresponding Ae5 temper-
atures. Also, under both conditions, the solidus temperatures were almost identical.
As a result, under paraequilibrium conditions the temperature ranges of δ-ferrite
formation were shorter, but the length of the fully ferritic domains were longer as
shown on figure 4.19.
Figure 4.19: Comparison of the A4 -A5 and A5 -solidus temperature intervals between equilibrium
and paraequilibrium conditions for steels A, B and C.
4.4 Summary 79
4.4 Summary
In the three studied steels, austenite was only retained in the as-received mi-
crostructure of the richly alloyed steel. The presence of this phase was shown to be
caused by a tempering treatment carried out in the intercritical temperature range.
In the same alloy, the presence of titanium in concentration sufficient to bind all
available carbon and nitrogen, was seen to limit martensite hardness change be-
tween the tempered and as-quenched conditions.
The dilatometric tests showed that δ-ferrite starts forming in supermartensitic

stainless steels at temperatures above 1300 ◦ C and the temperature range of forma-
tion does not exceed 50 ◦ C. As a result, the HAZ region over which δ-ferrite forms
can be expected to be narrow compared with the other HAZ regions.
MT-DATA correctly predicted all steels to have the capacity to become fully
austenitic and completely ferritic before melting. However, the agreement between
the equilibrium and experimental Ac4 temperatures was particularly poor. The
effects of nickel and molybdenum on equilibrium phase diagrams were also studied
under equilibrium condition. Molybdenum restricts the nickel range over which the
alloy can be made fully austenitic. It also strongly enlarges the Ae4 -Ae5 temperature
range but has a weaker effect on the fully ferritic range.
Finally, MT-DATA calculations performed under paraequilibrium conditions pre-
dicted shorter dual-phase fields but longer fully ferritic ranges compared with equi-
librium predictions.
Chapter 5
Microstructure in high
temperature heat-affected zone
Little work has been performed on the characterisation of the microstructure in

the HAZ of welded supermartensitic stainless steels. In particular, the presence of
δ-ferrite in the high temperature heat-affected zone1 (HT-HAZ) has not been thor-
oughly investigated. The aim of the work presented in this chapter was therefore
to provide a detailed study of the HT-HAZ microstructure, focussing on δ-ferrite
location, morphology, volume fraction and to examine the effects that welding pa-
rameters have on its retention.
5.1 First series of single-pass welds
To examine the different HAZ regions and compare the high temperature HAZ
microstructure developments between the three studied materials, a series of single-
pass autogenous bead-on-plates have been produced.
5.1.a Experimental procedure

Plates of dimension 100×50×5 mm were machined out of the pipes and were
taken so that the welds could be performed along the tube axis. To avoid reheating
effects and to study how the heat input affects the microstructure, four autogenous
1
The high temperature heat-affected zone refers to the HAZ regions next to the fusion line.
5.1 First series of single-pass welds 81
welds were produced using an automated T.I.G. (Tungsten Inert Gas) welding pro-
cess and two heat inputs. To allow for direct microstructural comparison, steels A
and B, and steels A and C were welded together as illustrated on figure 5.1.
The welding conditions are summarised in table 5.1. Argon was used as the
shielding gas during the welding operation. After welding, a transverse section was
taken approximately in the middle of each weld for optical examination. In the
following, the welds produced with the higher heat input (∼ 0.7 kJ mm−1 ) are
referred to as WAB1 and WAC1, and WAB2 and WAC2 for the lower energy welds
(∼ 0.3 kJ mm−1 ).
T.I.G Torch
Plate B
100 mm
Plate A
5 mm
100 mm
Figure 5.1: Schematic illustration of the welding configuration for plates A and B.
Condition Current Voltage Travel speed Arc efficiency Heat input

(A) (V) (mm s−1 ) [22] (kJ mm−1 )
1 208 9.3 1.66 0.6 ∼ 0.7
2 101 9.2 1.66 0.6 ∼ 0.3
Table 5.1: Welding parameters for autogenous dissimilar welds produced.

5.1.b Results
The microstructures of weld WAB1 (side A and B) and WAC1 (side C) revealed
with Villela’s reagent and the electrolytic etchant are shown on figure 5.2. The
microstructure of weld WAC1 side A is not shown because it is identical to that of
WAB1 side A. The microstructures of welds performed with 0.3 kJ mm−1 (WAB2
and WAC2) are shown in figure 5.3.
The welds performed with the highest heat input showed the finger tip mor-
phology, whereas welds carried out with the lower heat input showed hemispherical
morphology. The microstructures of the HT-HAZ apparently consisted, from the
fusion line of a coarse-grained and a two-phase (ferrite and martensite) region. The
electrolytic etchant did not reveal martensite and thus proved very effective in show-
ing ferrite in both the grain coarsened and dual-phase regions. However, the fusion
lines and weld pools could not easily be observed. Specimens etched with Villela’s
reagent showed the martensite and outlined the fusion line so that the use of both
etchants was complementary.
Contrarily to results presented by Enerhaug et al. the partially melted region

could not be observed [53]. In fact in their work the partially melted region was
most likely confused with the coarse-grained region. On the other hand, unlike steel
HT9 (studied in chapter 2), supermartensitic stainless steels have the capacity to
become fully ferritic before melting [63]. In this respect, it should be noted that MT-
DATA calculations made under equilibrium and paraequilibrium conditions correctly
predicted the existence of the fully ferritic domain (chapter 4).
5.1.c Coarse-grained heat-affected zone microstructure

The coarse-grained (CG) regions clearly had larger grain size than the dual-phase
(DP) HAZs, thus indicating that the materials had been completely ferritised before
cooling to room temperature.
In steel A, welds WAB1 and WAB2, the different morphologies of the austen-
ite that had formed during cooling from δ-ferrite could clearly be observed (figure
5.4). In addition to allotriomorphic (localised at grain boundaries) and intragranular
Figure 5.2: Optical micrographs of the same area of weld WAB1 side A ((a) and (b)). Optical
micrographs of the same area of weld WAB1 side B ((c) and (d)). Optical micrographs of the same
area of weld WAC1 side C ((e) and (f)). Villela’s etchant was used for (a), (c) and (e). Electrolytic
etchant was employed for (b), (d) and (f). Plate thickness 5 mm, heat input 0.7 kJ mm−1 .
Figure 5.3: Optical micrographs of weld WAB2 and WAC2 electrolytically etched. The fusion
line has been underlined and regions labelled. Plate thickness 5 mm, heat input 0.3 kJ mm−1.
austenite, an unusual austenite morphology was found. A survey of the literature

revealed this structure to be Widmanstätten austenite. Small amounts of “lathy” δ-
ferrite, whose width was about 400 nm, could be found within former Widmanstätten
austenite laths as seen on figure 5.5.
Mainly intragranular but also some Widmanstätten austenite could be found in

coarse-grained HAZ of steels B and C (figures 5.6 and 5.7). Those steels apparently
had a smaller volume fraction of Widmanstätten austenite compared with steel A.
From these observations, it could be concluded that the reverse transformation from
the fully ferritic region was virtually complete in all coarse-grained HAZ despite the
presence of tiny δ-ferrite films between Widmanstätten austenite laths.
i Widmanstätten austenite
The Widmanstätten austenite structure is more common in austenitic stainless
steels solidifying as δ-ferrite [65, 86]. Like Widmanstätten ferrite found in car-
bon steels, the appearance of Widmanstätten austenite is one of narrow wedges
emanating either directly from a grain boundary, or from allotriomorphic ferrite (al-
lotriomorphic austenite in the case of Widmanstätten austenite). In carbon steels,
Widmanstätten ferrite forms on cooling from austenite in the temperature range ∼
750-650 ◦ C. The structure forms by a displacive mechanism in which the rate of
transformation is controlled by the rate at which the carbon can diffuse to the side
of the growing plate.
In austenitic stainless steels, the Widmanstätten austenite structure is believed

to originate by reconstructive transformation [86, 87, 88]. In supermartensitic stain-
less steels, the morphology and coarseness of the Widmanstätten austenite (figure
5.5) provides evidence to support that it also forms through a reconstructive trans-
formation mode. Indeed, displacive transformations in steels are accompanied by a
shape deformation, which is an invariant-plane strain with a large shear component.
To minimise the strain energy associated with the shape change, the transformation
products adopt a morphology, which is a thin plate that tapers towards an ever
decreasing thickness as the plate-edge is approached. In two-dimensional sections,
Figure 5.4: Details of the CG-HAZ of steel A in weld WAB1, as seen at

the bottom of the plate (electrolytically etched). Plate thickness 5 mm,
heat input 0.7 kJ mm−1 .
Figure 5.5: Ferrite (light blue color) trapped within the former
Widmanstätten austenite in the CG-HAZ of weld WAB1 side A. Sam-
ple etched with color etchant. Plate thickness 5 mm, heat input 0.7 kJ
mm−1 .
Figure 5.6: Widmanstätten austenite in the CG-HAZ of steel B in

weld WAB1 (electrolytically etched). Plate thickness 5 mm, heat input
0.7 kJ mm−1.
Figure 5.7: Widmanstätten austenite in the CG-HAZ of steel C in

weld WAC1 (electrolytically etched). Plate thickness 5 mm, heat input
0.7 kJ mm−1.
such a plate appears lenticular with sharply pointed ends. The rounded tips of the
Widmanstätten austenite are therefore inconsistent with a displacive transformation
mechanism.
ii Phase transformation sequence in coarse-grained HAZ

Figure 5.8, shows the expected microstructural changes occurring in the coarse-
grained HAZ of supermartensitic stainless steels.
1400 A δ δ
c5
A c4 γ δ
1200
Temperature / C
δ γ
o
1000 Aγ
γ
Wγ
800 A
c3
Iγ
600
400
200
Ms
Time /s
Figure 5.8: Schematic showing microstructure evolution in coarse-grained HAZ during welding.
Aγ : allotriomorphic austenite, Wγ : Widmanstätten austenite, Iγ : intragranular austenite.
Upon heating, austenite starts to transform into δ-ferrite at the Ac4 temperature,
and the reaction is completed at the Ac5 temperature. Once the δ-ferrite single-phase
field has been reached, grain growth occurs. At high temperatures (≥ 1300 ◦ C),
equilibrium establishes rapidly and therefore it is reasonable to assume that the
ferrite had a homogeneous composition. Upon cooling δ-ferrite decomposed into
allotriomorphic, intragranular and Widmanstätten austenite followed by austenite
transformation to martensite on cooling below MS .
The difference in Widmanstätten austenite content between the three steels is

probably due to the temperature at which Widmanstätten austenite forms from
δ-ferrite since the dilatometric results (chapter 4) showed steel A to have its Ar4
temperature about 100-120 ◦ C higher than steels B and C.
5.1.d Dual-phase heat-affected zone microstructure

Micrographs of the δ-ferrite found in dual-phase HAZ of steels A, B and C
for both heat inputs are reported in figure 5.9. According to its smaller grain
size compared with the coarse-grained HAZ, the dual-phase region must have been
partially ferritised before cooling. Unlike steels B and C, steel A showed two different
δ-ferrite morphologies at the top and at the bottom surface of the plate of weld
WAB1. At the cap of the dual-phase HAZ, δ-ferrite was present both at prior
austenite grain boundaries and as parallel films within the grains (figures 5.10 and
5.9 (a)). At the root of the dual-phase HAZ, intragranular δ-ferrite was dissolved and
only grain boundaries δ-ferrite remained (figure 5.9 (b)). The different morphologies
can be attributed to a cooling rate difference between the top and bottom surface
of the plate as indicated by the larger grain size of the coarse-grained region (figure
5.2 (b)). In fact as will be shown later, the cooling rate at the bottom of the plate
was about 2.5 times slower than at the cap.
The ferrite content was also found to be different between the three alloys and
dependent upon the heat input. For a given heat input, the volume percentage of δ-
ferrite in dual-phase HAZ was apparently larger for steel A, however, quantification
of δ-ferrite as a function of heat input is reported later in this chapter.
i Elemental partitioning
Tables 5.2 and 5.3 summarise the results of fifteen EDX microanalyses performed
on the martensitic matrix and δ-ferrite in dual-phase HAZ of steels A, B and C.
Ferrite in steel A was significantly Cr and Mo enriched as well as Ni depleted

compared to martensite. Similarly, δ-ferrite in steels B and C was enriched in Cr
but depleted in Ni. This showed that despite the fast heating and cooling rates
Figure 5.9: Optical micrographs of : weld WAB1 side A ((a) and (b)), weld WAB1 side B ((c)
and (d)) and weld WAC1 side C ((d) and (e)). Villela’s etchant was used for (a), (c) and (e).
Electrolytic etchant was employed for (b), (d) and (e). Note the different DP-HAZ widths in (b),
(d) and (f). Plate thickness 5 mm, heat input 0.3 kJ mm−1 .
Figure 5.10: TEM micrograph showing grain boundary and

intragranular δ-ferrite found in the dual-phase HAZ of weld WAB1
side A.
Phase Steel A
Cr Ni Mo
δ-ferrite 13.5 ± 0.9 4.8 ± 1.0 2.6 ± 0.4
Matrix 11.7 ± 0.3 7.0 ± 0.4 1.7 ± 0.2
Table 5.2: EDX microanalysis (wt%) of ferrite and martensite in dual-phase of steel A (weld
WAB1).
Phase Steel B Steel C
Cr Ni Cr Ni
δ-ferrite 13.2 ± 0.7 0.8 ± 0.3 14.1 ± 0.8 2.1 ± 0.4
Matrix 10.6 ± 0.3 1.5 ± 0.2 12.1 ± 0.3 3.0 ± 0.2
Table 5.3: EDX microanalysis (wt%) of ferrite and martensite in dual-phase HAZ of steels B
and C (weld WAB1 and WAC1).
experienced by the HAZ during welding, partitioning of substitutional elements had

time to occur. An X-ray map taken in dual-phase HAZ of weld WAB1, side A,
allowed direct observation of elemental distribution (figure 5.11).
Figure 5.11: Secondary electron SEM image and X-ray mapping of Cr, Mo and Ni, in dual-
phase HAZ of weld WAB1 side A. In each square the X-ray signal emitted by the element present
in the square is acquired. The intensity (brightness) of the signal is a function of the element
concentration at the location considered.
ii Microstructure evolution in dual-phase HAZ

In the dual-phase region, the microstructural development can be summarised
as follows (figure 5.12). During heating, uncomplete growth of the δ-ferrite phase
nucleated from prior austenite grain boundaries and intragranularly occurred in the
Ac4 -Ac5 temperature range. As EDX results showed, despite the fast heating and
cooling rates experienced by the dual-phase HAZ, redistribution of substitutional
elements and therefore also interstitial elements took place. The mechanism of fer-
rite formation is thus diffusional. Consequently, in the partially ferritised region, the
extent of δ-ferrite dissolution is determined by the level of chromium and molybde-
num enrichment of the δ-ferrite and the cooling rate. Fast cooling rates lead to the
retention of both types of ferrite, while slow cooling rates allow dissolution of the
intragranular δ-ferrite.
1400 A
c5
γ
A c4 γ + δ
1200 δ
Temperature / C
o
1000
γ
800 A
c3 γ
600 δ
400
200
Ms
Time /s
Figure 5.12: Schematic of the dual-phase HAZ microstructure evolution during welding.
iii Weld thermal cycle measurement

An attempt has been made to measure the cooling rate at the cap and at the
root of the high temperature HAZ. The idea was to reproduce weld WAB1 and
attach thermocouples as close as possible to the fusion line. Three holes of 0.5 mm
diameter were consequently drilled in plate A at specific positions to accommodate
R-type thermocouples of 0.25 mm diameter. Considering the small size of the high
temperature HAZ, the main difficulty was to locate the thermocouples at the right
place. Thermocouple T1 was located at the top of the plate and T2 and T3 at the
bottom. The temperature was recorded using a data logger connected to a laptop.
Figure 5.13 shows the actual positions of the thermocouples after the experiment.
Unfortunately during cutting the thermocouples moved or fell off the plate so that
it was not possible to confirm their actual positions in the holes.
5.2 Effect of cooling rate on δ-ferrite retention 94
The temperature versus time plots are reported in figure 5.14. The times required
to reach peak temperatures were surprisingly long i.e. about 9 s for each thermo-
couple. T1 and T2 reached almost the same peak temperature of about 1400 ◦ C
but T3 went only up to ∼ 1250 ◦ C. The most interesting comparison was therefore
between thermocouples 1 and 2. To determine the cooling rates, the cooling time
between 1400 and 1200 ◦ C (∆t14−12 ) was chosen. The ∆t14−12 was 1.7 s at the cap
of the HAZ and 4.4 s at the bottom of the plate. Those times correspond to cooling
rates of about 117 and 45 ◦ C s−1 respectively. This shows that the cap of the HAZ
cooled about 2.5 times faster than the bottom of the HAZ. It is the slower cooling
rate of the bottom of the HAZ that allowed the intragranular δ-ferrite to dissolve.
5.2 Effect of cooling rate on δ-ferrite retention
The aim of this section was to study the effect of a wide range of cooling rates
on the δ-ferrite retention in specimens quenched from the δ-ferrite single-phase field
and the δ-ferrite plus austenite region. As microstructures were best revealed in
steel A, only this steel has been selected for this part.
5.2.a Experimental procedure

Small coupons of dimension 10×15×3 mm were machined in steel A. Specimens
were placed in a pre-heated furnace (at 1375 ◦ C for the first test and then at 1350 ◦ C)
at the center of the bottom surface and were let in 5 minutes. Specimens were sub-
sequently either water quenched, air cooled, or were left in the furnace switched
off with the door open until the temperature reached about 700 ◦ C and then wa-
ter quenched. To get rid of the oxidation layer after heat-treatment, coupons were
mounted and roughly ground to remove approximately 1 mm of matter. For optical
examination specimens were finely ground, polished and electrolytically etched in
sulfuric acid solution. The volume fractions of δ-ferrite were measured at a magni-
fication of 100 using image analysis and the procedure described in chapter 3.
Figure 5.13: Macrograph showing the position of the three thermocouples in plate A. Plate
thickness 5 mm, heat input 0.7 kJ mm−1.
1600
Thermocouple 1
Thermocouple 2
Thermocouple 3
1400
1200
C
1000
o
Temperature /
800
600
400
200
0
0 5 10 15 20 25 30 35 40 45 50
Time / s
Figure 5.14: Thermal cycles measured at three different positions in HAZ of steel A.
5.2.b Results
Figure 5.15 shows the microstructures obtained after water quenching, air cooling
and furnace cooling from 1375 ◦ C. The specimen was clearly cooled from the δ-ferrite
single-phase region as indicated by the large grain size. The volume fraction and
coarseness of Widmansttäten austenite were seen to increase with decreasing cooling
rate, as a result, increasingly thicker films of δ-ferrite remained trapped between
Widmansttäten austenite subunits (figure 5.15 (b), (d) and (f)).
Figure 5.16 illustrates the microstructures obtained after cooling from 1350 ◦ C.
δ-ferrite was observed intragranularly and also at grain boundaries, the material was
therefore quenched from the δ-ferrite + austenite phase field. Reducing the cooling
rate clearly led to the retention of less ferrite.
The volume fraction of δ-ferrite measured for each heat-treatment and cooling
method is plotted in figure 5.17. It can be seen that the quantity of retained δ-ferrite
is dependent upon the cooling rate and heat-treatment temperature. When cooling
occurs from the δ-ferrite single-phase field, fast cooling rates limit the formation
of Widmansttäten austenite but not that of the allotriomorphic and intragranular
austenite and consequently only small quantities of δ-ferrite are trapped. On the
other hand, when cooling occurs from the δ-ferrite + austenite phase field, the faster
the cooling rate the larger the δ-ferrite volume fraction. This is because solute
redistribution is limited during fast cooling. Consequently in weld heat-affected
zones fast cooling rates prevent δ-ferrite dissolution in dual-phase HAZ but limit its
volume fraction in coarse-grained HAZ.
According to the dilatometric results of chapter 4, ferritisation should not have
occurred at 1350 ◦ C (i.e. Ac4 was 1372 ◦ C). However, inside the furnace chamber
the temperature is not necessarily uniform and differences between the temperature
set on the furnace and actual temperature are not surprising. It should also be men-
tioned that since the dilatometer used in chapter 4 did not allow the thermocouple to
be attached to the specimen surface, a small error in the phase transformation tem-
peratures can be expected. However considering the slow heating rate used in the
dilatometric experiments, the extent of the error should be limited to few degrees.
Figure 5.15: Optical micrographs illustrating the microstructures of steel A, obtained after
isothermal transformation at 1375 ◦ C for 5 minutes and water quenched ((a) and (b)), air cooled
((c) and (d)) and furnace cooled ((e) and (f)). Micrographs (b), (d) and (f) are magnified pictures
of (a), (c) and (e) respectively. Electrolytically etched.
Figure 5.16: Optical micrographs illustrating the microstructures of steel A, obtained after
isothermal transformation at 1350 ◦ C for 5 minutes and water quenched ((a) and (b)), air cooled
((c) and (d)) and furnace cooled ((e) and (f)). Micrographs (b), (d) and (f) are magnified pictures
of (a), (c) and (e) respectively. Electrolytically etched.
5.3 Effect of heat input 99
Figure 5.17: Volume fraction of retained δ-ferrite as a function of heat-treatment temperature

and cooling mode.
5.3 Effect of heat input
There is no published information on the effect of heat input on δ-ferrite content

in supermartensitic stainless steels. However, this information might be useful if the
presence of δ-ferrite in HAZ is to be minimised. To quantify the effect of weld heat
input a second series of bead-on-plates has been performed.
5.3.a Experimental Procedure

Three autogenous single-pass welds have been produced for each material us-
ing three heat inputs. Thicker plates of dimension (100×100×10 mm) have been
machined to allow the use of a relatively high heat input and to keep the high tem-
perature isotherms parallel to the weld pool (3 dimensional heat flow). Table 5.4
lists the welding parameters employed. The volume percentage of δ-ferrite in dual-
phase HAZ was measured by image analysis according to the procedure described
in chapter 3.
5.4 Multipass welds 100
Current Voltage Travel speed Heat input

−1
(A) (V) (mm s ) (kJ mm−1 )
WX1.0 275 10.0 1.66 1.0

WX0.7 208 9.3 1.66 0.7
WX0.3 101 9.2 1.66 0.3
Table 5.4: Welding parameters for the three single-pass welds carried out on steels A, B and C.
X is used to substitute A, B or C.
5.3.b Results
The microstructures of the welds performed on the three steels are reported in
figures 5.18 to 5.20. As expected, the volume of the weld pool and the size of the
HAZ increased with increasing heat input.
The volume percentages of retained δ-ferrite in dual-phase HAZ of the three steels
are plotted as a function of heat input in figure 5.21. Ferrite presence never exceeded
15 % and had the tendency to decrease with increasing heat input. Consistent with
the results of the observations made in the previous section, steel A contained larger
amount of δ-ferrite in its dual-phase HAZ compared to steels B and C.
The volume fraction of δ-ferrite trapped in the coarse-grained regions could not
be easily measured, because the distribution of Widmansttäten austenite was non
uniform in those regions. However, being closer to the fusion line, the cooling rate
of the coarse-grained HAZ is always faster than that of the dual-phase HAZ and
therefore according to figure 5.15 less δ-ferrite is to be expected in this region.
5.4 Multipass welds
The work described in the previous sections established the effect of chemical
composition and heat input on δ-ferrite retention in single-pass welds, but neglected
reheating effects that occur during the deposition of subsequent passes. However,
Figure 5.18: Optical micrographs of single-pass welds performed on steels A ((a) and (b)), B
((c) and (d)) and C ((e) and (f)) using a heat input of 0.3 kJ mm−1 (electrolytically etched).
Micrographs (b), (d) and (f) are magnified pictures of micrographs (a), (c) and (e) respectively.
Plate thickness 10 mm.
Figure 5.19: Optical micrographs of single-pass welds performed on steels A (a), B (b) and C
(c) using a heat input of 0.7 kJ mm−1 (electrolytically etched). Plate thickness 10 mm.
Figure 5.20: Optical micrographs of single-pass welds performed on steels A (a), B (b) and C
(c) using a heat input of 1 kJ mm−1 (electrolytically etched). Plate thickness 10 mm.
Volume percentage retained delta-ferrite / %

20
Steel A
Steel B
Steel C
15
10
0
0 0.2 0.4 0.6 0.8 1
-1
Heat input / kJ mm
Figure 5.21: Volume fraction of retained δ-ferrite in dual-phase HAZ as a function of heat input
for steels A, B and C. Plate thickness 10 mm.
actual welds are multipass and therefore the primary HAZ microstructure is sub-
ject to reheating. To check the effect of reheating on the high temperature HAZ
microstructure, actual multipass welds have been observed.
The multipass welds studied were provided by TWI. They all had similar joint
design i.e. 60◦ angle with no face root, and were welded using a duplex stainless
steel filler wire (Zeron 100, table 5.5) employing up to 14 passes with an interpass
temperature of 150 ◦ C. Tables 5.6 to 5.8 summarise the welding parameters em-
ployed. The first two passes are usually deposited using a higher heat input due to
heat loss through the gap between the workpieces.
5.4.a Observations
Details of the microstructure of multipass welds performed on steels A, B and
C are reported in figures 5.22 to 5.27. The features of the high temperature HAZ
were more difficult to observe in multipass welds due to various level of tempering.
C Si Mn Ni Cr Mo Cu W N
Zeron 100 0.018 0.39 0.69 9.30 24.80 3.80 0.60 0.61 0.22
Table 5.5: Chemical composition of Zeron 100 in wt%.
Run Process Filler size Current Voltage Heat input

number (mm) (A) (V) (kJ mm−1 )
1 TIG 2.4 60 9.0 1.10
2 TIG 2.4 88 9.5 1.24
3-4 MMA 2.5 62-66 22.1-23.9 0.44-0.70
5-7 MMA 3.2 82-96 22.2-24.3 0.70-1.06
Table 5.6: Welding parameters for multipass weld performed on steel A.
In each weld, attention was focussed to the cap and the root of the HAZ since it is
those regions that are in contact with the external and internal environments.
No microstructure could be observed in the coarse-grained HAZ of the multipass

weld performed on steel A (figure 5.22). The morphology and volume fraction of
the δ-ferrite found in dual-phase HAZ were different at the cap of the weld HAZ
compared to the root (figure 5.23). Similar to the observations made on the low
heat input bead-on-plates on steel A, ferrite at the cap of the HT-HAZ was present
at prior austenite grain boundaries and as intragranular films. Whilst at the HAZ
root, δ-ferrite was only present at prior austenite grain boundaries.
The origin of the different ferrite morphologies is multiple. First a cooling rate
difference it to be expected in HAZ between the cap and the root. Indeed due to
the presence of a larger heat sink, the cap of the HAZ cools down faster than the
root, and as seen before, this prevents the intragranular δ-ferrite from dissolving. In
addition, due to the deposition of several layers, the HAZ root is reheated several
times and this encourages the remaining ferrite to dissolve. Consequently, δ-ferrite
Figure 5.22: Optical macrograph showing continuous film of δ-ferrite (arrowed) in the high
temperature heat-affected zone of the multipass weld performed on steel A (electrolytically
etched). 7 passes were employed to complete the weld
Figure 5.23: Optical photographs illustrating the different morphology of δ-ferrite in heat-affected
zone (a) at the cap and (b) at the root (electrolytically etched).

1 TIG 2.4 80 10.0 1.36
2 TIG 2.4 110 11.0 1.43
3-11 MMA 2.5 69-77 22.0-26.4 0.57-1.23
Table 5.7: Welding parameters for multipass weld performed on steel B.

1 TIG 2.4 95 9.8 1.48
2 TIG 2.4 122 10.5 1.50
3-14 MMA 3.2 86-91 23.1-25.9 0.59-1.50
Table 5.8: Welding parameters for multipass weld performed on steel C.
distribution in the whole dual-phase HAZ is far from homogenous and quantification
is therefore difficult. However, based on the results of the previous section, the
volume fraction of δ-ferrite at the cap of the HAZ can be estimated to be of the
order of 10 to 15 %.
Ferrite in dual-phase HAZ of steels B and C could not be revealed using the
electrolytic etchant, instead, Villela’s reagent was employed (figure 5.24 and 5.26).
The coarse-grained HAZs at the cap of steels B and C were clearly larger than
that of steel A, and showed the presence of some Widmansttäten austenite. Also,
consistent with the results obtained on the bead-on-plates, steels B and C showed
little or no evidence of δ-ferrite retention in dual-phase HAZ. Contrarily to steel A,
grain refinement was observed at the root of the HAZ of steels B and C. The latter
difference might be explained by the larger number of passes employed to perform
those welds.
Figure 5.24: Optical macrograph showing the high temperature heat-affected zone (ar-
rowed) of the multipass weld performed on steel B (etched with Villela’s reagent). 11 passes
were employed to complete the weld.
Figure 5.25: Close-up of (a) the cap of the HT-HAZ and (b) the root of the HT-HAZ (etched
with Villela’s reagent).
Figure 5.26: Optical macrograph showing the high temperature heat-affected zone (ar-
rowed) of the multipass weld performed on steel C (etched with Villela’s reagent). 14 passes
were employed to complete the weld.
Figure 5.27: Close-up of (a) the cap of the HT-HAZ and (b) the root of the HT-HAZ (etched
with Villela’s reagent).
5.5 Summary and conclusions 110
5.4.b Remarks
Compared with the high temperature HAZ of the semi-ferritic stainless steel
showed in figure 2.7, supermartensitic stainless steels retain significantly less δ-ferrite
and possess a more complex microstructure. Consequently, establishing the effect of
δ-ferrite on the properties is more difficult. However, since steels similar to steel A
are more likely to retain detectable amount of δ-ferrite in HAZ, it is advisable to use
those materials to establish the effect of δ-ferrite on impact and corrosion resistance
properties. On the other hand, assuming δ-ferrite to be undesirable, attention should
be drawn to the top surface of the weld HAZ, since the previous section revealed
δ-ferrite to be mainly concentrated at the cap of the HAZ.
The effect of the nature of the filler wire on δ-ferrite retention has not been
investigated. Nevertheless considering that the dual-phase HAZ is usually located
at around 100 µm from the fusion line, the migration of carbon or nitrogen from or
towards the weld metal is unlikely.
5.5 Summary and conclusions
The use of autogenous bead-on-plates has permitted the identification of the

different high temperature HAZ regions and sequence of phase transformations.
The high temperature HAZ consists of a coarse-grained and a narrow dual-phase
HAZ. Thorough examination of the microstructure revealed the presence of small
amounts of δ-ferrite in both regions. In coarse-grained HAZ, δ-ferrite was retained
as thin films between Widmanstätten austenite laths. On the other hand, δ-ferrite
was found at prior austenite grain boundaries and also within the austenite grains
in dual-phase HAZ. Strong partitioning of ferrite stabilising elements into δ-ferrite
and fast cooling rates explained its retention.
Using small heat-treated specimens and different cooling methods, it could be

demonstrated that fast cooling rates favour ferrite retention in specimens cooled from
the austenite plus δ-ferrite region but minimise its presence in specimens cooled from
the fully ferritic field.
On the other hand, the volume fraction of δ-ferrite retained in dual-phase HAZ
5.5 Summary and conclusions 111
was quantified as a function of heat input. Larger heat inputs led to wider dual-
phase HAZ regions and lower δ-ferrite volume fractions. In all experiments, the most
corrosion resistant alloy (containing molybdenum) retained larger quantities of δ-
ferrite, and had a higher tendency to form Widmanstätten austenite when cooled
from the fully ferritic phase field.
Actual weldments, performed using arc welding techniques, show a non-uniform
distribution of δ-ferrite in dual-phase HAZ. Due to reheating effects δ-ferrite formed
during the deposition of the first layer, tend to dissolve and consequently δ-ferrite
is mainly located at the cap of the weld HAZ. This is, however, mainly noticeable
on molybdenum containing alloys.
Chapter 6
Predicting δ-ferrite retention and

dual-phase HAZ width
As described in chapter 5, examination of single-pass bead-on-plate weld deposits,

revealed that the richly alloyed supermartensitic steel was more prone to ferrite re-
tention and possessed a larger dual-phase HAZ than any of the other alloys studied.
Since the presence of δ-ferrite might be harmful to the properties, it would be inter-
esting to be able to predict the susceptibility to δ-ferrite retention and to be able to
estimate the extent of the region where it proliferates.
6.1 Assessment of existing methods of ferrite pre-

diction
6.1.a Ferrite factor and Balmforth diagram

The chemical compositions of the three steels listed in table 6.1 have been used
to calculate their ferrite factor (F F , equation 6.1 in wt%) and locate their position
on the recently developed Balmforth diagram (chapter 2).
F F = Cr + 6Si + 8Ti + 4Mo + 2Al + 4Nb − 2Mn − 4Ni − 40(C + N) (6.1)

Steel A had the lowest ferrite factor (-3.04) followed by steel C (0.68) and B
(2.94). This order suggests that steel B should retain the largest amount of ferrite,
6.1 Assessment of existing methods of ferrite prediction 113
but as figure 2.9 shows, F F values below 5 lead to relatively small amounts of
retained δ-ferrite, typically less than 10 %. Experimentally, although the volume
fractions of HAZ retained δ-ferrite are of the same order of magnitude, steel A
contained more ferrite than the other alloys. Consequently, it is concluded that the
ferrite factor is not very helpful for making comparisons.
Steel name C Si Mn Ni Cr Mo Cu Ti N
A 0.01 0.26 0.46 6.46 12.20 2.48 0.03 0.09 0.0070

B 0.01 0.18 1.14 1.55 10.90 ≤ 0.01 0.49 0.01 0.0060
C 0.01 0.19 0.24 3.12 12.50 0.02 0.06 0.01 0.0120
Table 6.1: Chemical composition of steels A, B and C in wt%.
On the other hand, the Balmforth diagram (figure 6.1) predicts steels B and C
to retain similar levels of ferrite (about 20 %), but less than steel A (50 %). In fact,
the larger volume percentage predicted for steel A is mainly due to the presence of
molybdenum. The coefficient 2 attributed to molybdenum significantly affects the
value of the Creq and consequently raises the expected level of retained ferrite. On
the other hand, the low levels of C and N in supermartensitic steels means that the
Nieq virtually equals the nickel concentration. Considering that Mo and Ni are the
elements whose concentrations vary the most in supermartensitic steels, it is the
balance between those elements that strongly influences ferrite retention.
Although the absolute levels of ferrite retention do not match with the experi-
mental values of chapter 5, Balmforth diagram correctly predicts that steel A has
the highest tendency to retain ferrite.
6.1.b MAP-neural-ferrite-number
There exist more advanced methods for estimating δ-ferrite retention in welds.
The methods based on empirical modelling such as neural networks1 can be very
useful when complex interactions between solutes are expected to affect the retained
1
An introduction to neural networks is given in chapter 8.
Steel A
7
11
00
00
11
6 A+M+F
Ni + 35 C + 20 N (in wt%) M 10% 30% 50% 70% 90%

5
Steel C 20% 40% 60% 80%
4
1
0
0
1
3
Steel B F+M
2 11
00
00
11 F
1
0
8 10 12 14 16 18 20 22 24
Cr + 2 Mo + 1.5 Si + 10 (Al + Ti) (in wt%)
Figure 6.1: Balmforth diagram and the position of steels A, B and C [43].
2
ferrite content. Some models designed to predict the ferrite number (F N as
opposed to ferrite percentage F P ) of austenitic weld metals have recently been
developed.
Because magnetic methods offer a convenient way of measuring the ferrite content
in an austenitic matrix, a large number of data of the kind essential to empirical
modelling can be produced. As mentioned in chapter 2, such magnetic methods
cannot be employed for martensitic steels because ferrite and martensite are both
ferromagnetic. However, neural networks are in principle capable of making extrap-
olations for compositions that are not covered in the data range used to create them,
assuming that relationships are recognised and that each predictions is accompanied
by an estimate of uncertainty.
Although the main interest here is on heat-affected zones and not weld metals,
2
It is possible to relate F N to F P by relationships such as F N = 1.7 × F P as introduced in
chapter 2
the aim of the following calculations is to compare the capacity of an alloy to retain
δ-ferrite regardless of its location within the weld.
The recent neural network model proposed by Murugananth et al., MAP-NEURAL-

FERRITE-NUMBER3 was chosen to conduct calculations [80]. The model uses
chemical composition as the input and was trained on a database containing 925
experimental data. The composition range of the elements considered in the analysis
is reported in table 6.2.
Input element Minimum Maximum Mean Standard deviation
C 0.00 0.20 0.04 0.02

Mn 0.35 12.67 1.85 1.73
Si 0.03 6.46 0.53 0.34
Cr 1.05 32.00 20.47 2.70
Ni 4.61 33.50 11.26 2.60
Mo 0.01 10.70 1.48 1.55
N 0.01 2.13 0.08 0.10
Nb 0.01 0.88 0.04 0.09
Ti 0.01 0.54 0.03 0.04
Cu 0.02 6.18 0.15 0.41
V 0.04 0.27 0.06 0.02
Co 0.01 0.69 0.05 0.04
Fe 45.60 72.51 63.98 4.21
FN 0.00 117.00 12.63 18.40
Table 6.2: Input variables used in the model. Composition in wt% [80].
3
www.msm.cam.ac.uk/map/neural/programs/ferrite-number.html
To check the capacity of the neural network model to make predictions on su-
permartensitic stainless steels, the effect of the major solute elements on the F N
has been investigated using steel A as a base composition. The results of these pre-
dictions are reported in figure 6.2 for the austenite stabilising elements and figure
6.3 for the ferrite stabilising elements.
The neural network model identified clear trends for the effects of nickel, chromium,
molybdenum and titanium on the ferrite number. Consistent with the nature of each
of these elements the F N values were predicted to decrease with Ni and to increase
with the addition of Cr, Mo and Ti.
i Effects of Cr, Ni and Mo

Contour plots have been drawn to visualise the effects of Cr, Ni and Mo on the
predicted ferrite number. The calculations were performed on a Cr-Ni-Mo ternary
alloy with Cr and Ni concentrations varying between 9-14 wt% and 0-8 wt% respec-
tively, and with two different levels of molybdenum (0 and 2 wt%).
Most of the F N values do not exceed 10 in the molybdenum-free diagram (figure

6.4). However consistent with Balmforth diagram, a sensible F N increase is felt for
high chromium and low nickel containing alloys, the maximum value being about
40. On the other hand, at 2 wt% molybdenum (figure 6.5), the F N values are
significantly higher than in the previous case. Also, similar to the molybdenum-free
diagram the effect of molybdenum on ferrite retention is more strongly felt at low
Ni concentrations. Finally in both cases, the errors become larger as the nickel
concentration is reduced. The latter was to be expected since the database did not
contain experimental data on low nickel containing steels.
ii Application to steels A, B and C
Figure 6.6 (a) shows the predicted F N values for steels A, B and C. Due to the
size of the error bars it cannot be concluded that steel A is more likely to retain
δ-ferrite than steels B and C. However the maximum predicted levels were typically
20 F N (equivalent to about 12 %) and were consistent with those measured in
chapter 5.
Predicted Ferrite Number

40 40
a) b)
35 35
30 30
25 25
20 20
15 15
10 10
5 5
0 0
0 0.01 0.02 0.03 0.04 0.05 0 0.01 0.02 0.03 0.04 0.05
Carbon / wt% Nitrogen / wt%

40 80
c)
35 70
30 60
25 50
20 40
15 30
10 20
5 10
d)
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 0 1 2 3 4 5 6 7 8 9 10
Manganese / wt% Nickel / wt%
40
e)
35
30
25
20
15
10
5
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Copper / wt%
Figure 6.2: Predictions of ferrite number (F N ) as a function of chemical composition for austenite
stabilising elements (base composition steel A : 0.01 C, 0.46 Mn, 0.26 Si, 12.2 Cr, 6.46 Ni, 2.48
Mo, 0.007 N, 0.09 Ti wt%). The calculations were done using a published neural network model
[80].

40 40
f) g)
35 35
30 30
25 25
20 20
15 15
10 10
5 5
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 10 10.5 11 11.5 12 12.5 13 13.5 14
Silicon / wt% Chromium / wt%
40

40
h) i)
35 35
30 30
25 25
20 20
15 15
10 10
5 5
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.05 0.1 0.15 0.2
Molybdenum wt% Titanium / wt%
Figure 6.3: Predictions of ferrite number (F N ) as a function of chemical composition for ferrite
stabilising elements (base composition steel A : 0.01 C, 0.46 Mn, 0.26 Si, 12.2 Cr, 6.46 Ni, 2.48
Mo, 0.007 N, 0.09 Ti wt%). The calculations were done using a published neural network model
[80].
The evolution of F N as a function of nickel concentration was also investigated

for each steel (figure 6.6 (b) and (c)). The level of nickel required to suppress δ-
ferrite retention was significantly greater for steel A (≥ 10 wt%) than for steels B
and C. This showed that although the presence of Mo must be balanced by the
appropriate amount of Ni to allow the steel to become fully austenitic at elevated
temperature (6.5 wt% in the case of steel A), the nickel concentration required to
suppress δ-ferrite retention might be larger. It should however be noted that the
addition of an extra 4 wt% Ni to steel A would strongly depress the MS temperature
and possibly change the solidification mode and hence would not necessarily lead to
a completely martensitic alloys after cooling to room temperature.
Figure 6.4: Contour plots showing predicted F N and associated errors as a function of Cr and
Ni concentration for a ternary alloy with 0 wt% Mo.
Figure 6.5: Contour plots showing predicted F N and associated errors as a function of Cr and
Ni concentration for a ternary alloy with 2 wt% Mo.
40
a)

30
20
10
0
Steel A Steel B Steel C
100
b)
80
60
40
Steel A
Steel B ||
20 ||| v|
||
v|
0
0 2 4 6 8 10 12 14
Nickel / wt%
100
c)
80
60
Steel B
40 |
|
|| Steel C
||
| |
20 | |
| |
| v
V
0
0 2 4 6 8 10
Nickel / wt%
Figure 6.6: Predicted ferrite number for steels A, B and C (a). Effect of nickel on F N in steels
A and B (b) and B and C (c). The arrows indicate the nickel content and corresponding predicted
F N positions of steels A, B and C.
6.2 Dual-phase HAZ width 121
6.2 Dual-phase HAZ width
The aim of the work presented in this section was to compare the extent of the
dual-phase HAZs of the welds performed in chapter 5 with values estimated using
Rosenthal’s heat-flow equations.
6.2.a Experimental data

The nine welds performed on the 10 mm thick plates of steels A, B and C in
chapter 4 have been chosen for providing experimental data. The widths of all dual-
phase HAZs have been measured according to the procedure described in chapter 3,
and the results are plotted as a function of the employed heat input in figure 6.7.
As expected, the extent of the dual-phase regions were small (maximum value
of about 150 µm) and increased with increasing heat input. For any given heat
input, steel A had the largest range of δ-ferrite formation followed by steels C and
B respectively.
200
Steel A
Dual-phase HAZ width / micrometers
Steel B
175 Steel C
150
125
100
75
50
25
0
0 0.2 0.4 0.6 0.8 1
-1
Heat input / kJ mm
Figure 6.7: Measured widths of the dual-phase heat-affected zones as a function of heat input
for steels A, B and C. Plate thickness 10 mm.
6.2.b Weld thermal cycle prediction

During welding the temperature profile in HAZ can be estimated by using Rosen-
thal’s equations [22]. There are well known assumptions associated with this analyt-
ical method but experience has shown that they can be a useful approximation. In
arc welding processes, the flow of heat from the source depends on the thickness of
the plate. For thick plates, the arc is assimilated to a point source of constant power
moving along the surface of an isotropic semi-infinite body, at initial temperature
T0 , limited in one direction by a plane that is impermeable to heat. The thermal
cycle experienced by the heat-affected zones (assuming instantaneous application of
heat) at a certain fixed position from the heat source as defined by a radial distance
r gives [49] :
q
v r2
T = T0 + exp − (6.2)
2πλt 4at
T0 is the temperature prior to welding or the pre-heat temperature in K,
q is the arc power (J s−1 ),
v is the arc velocity (m s−1 ),
t is the time expressed in second,
r is the radial distance from the weld centre line (m),
a is the thermal diffusivity (m2 s−1 ) and is defined as follows :
λ
a= (6.3)
ρc
where λ is the thermal conductivity in J s−1 m−1 K−1 and ρc specific heat per unit
volume in J m−3 K−1 .
Equation 6.2 gives a reasonable description of the thermal cycle as experienced

by the heat-affected base metal and is to be applied only at temperatures outside
the fusion zone. The peak temperature at a given distance in HAZ is then obtained
using the following equation:
q
2 v
Tp = T0 + (6.4)
πe ρcr2
where e= 2.718.
It should be noted that the latter equation does not depend upon thermal con-
ductivity.
6.2.c Application
The equilibrium and paraequilibrium phase transformation temperatures deter-
mined in chapter 4 (summarised in table 6.3) were used together with the expression
of the peak temperature distribution (equation 6.4) to determine the extent of the
dual-phase HAZ for a given heat input. To allow for more precise comparisons with
experimental data, calculations were only performed for a level of heat input of
0.3 kJ mm−1 , for which the high temperature isotherms were found to be almost
semi-circular (figure 5.18). The value of ρc was taken to be 4.60 106 J m−3 K−1
according to reference [89].
Ap4 (5%) Ap5 (95%) Ae4 (5%) Ae5 (95%)
Steel A 1322 1323 1237 1396

Steel B 1302 1310 1261 1350
Steel C 1295 1322 1242 1360
Table 6.3: Reminder of equilibrium and paraequilibrium phase transformation temperatures (in
◦
C) calculated in chapter 4.
The results are presented in figure 6.8. The dual-phase HAZ widths calculated
using the paraequilibrium temperatures were, considering the error bars, in relatively
good agreement with the measured values except for steel A, where the Ap4 -Ap5
temperature interval is only 1 ◦ C. On the other hand, the dual-phase HAZ widths
calculated using the equilibrium temperatures overestimated the actual width by a
factor varying between 3 and 4.
For steels B and C, the satisfactory agreement between actual and calculated
Figure 6.8: Comparisons between measured and calculated widths of the dual-phase HAZ of
steels A, B and C for a heat input of 0.3 kJ mm−1 .
6.3 Summary 125
widths using paraequilibrium temperatures suggests that ferritisation is governed by

carbon and nitrogen diffusion. It should however be kept in mind that partitioning
was observed between δ-ferrite and austenite and therefore the transformation is
certainly governed by an intermediate state.
As far as equilibrium temperatures are concerned, the poor agreement with ex-
perimental values might be explained by the fact that kinetics have been neglected.
Indeed under rapid heating condition the ferritisation reaction might start at a higher
temperature, but the completion of the transformation is also likely to be delayed.
However, the dilatometric tests carried out under slow heating conditions in chapter
4 clearly showed MT-DATA systematic underestimation of the Ae4 temperature.
MT-DATA unsatisfactory predictions are related to the relatively large number of
elements allowed in the calculations (often 8 or 9), and the lack of experimental
thermodynamic data relevant to supermartensitic stainless steels.
It is therefore clear that neither paraequilibrium nor equilibrium conditions are
satisfactory to predict the width of the dual-phase HAZ of supermartensitic stainless
steels. Consequently improvement of MT-DATA databases are required to solve the
problem.
6.3 Summary
While the ferrite factor equation did not provide a useful means for making
comparisons between different supermartensitic alloys, the diagram proposed by
Balmforth correctly predicted the higher capacity of steel A to retain δ-ferrite. Be-
sides the expressions of the Creq and Nieq indicated that Ni and Mo play a major
role on the extent of δ-ferrite retention.
According to the neural network results, it is difficult to balance the ferrite
stabilising effect of molybdenum by only adding the minimum amount of nickel
required to obtain a fully austenitic steel at high temperature. Unfortunately, due
to the size of the error bars the predictions made using the neural network model did
not show steel A’s larger capacity to retain ferrite. Consequently, only Balmforth
diagram seems suitable for supermartensitic stainless steels.
6.3 Summary 126
Due to a lack of accurate thermodynamic data on the systems of concern, it

was not possible to predict the extent of the HAZ region where δ-ferrite is mainly
concentrated.
Chapter 7
Effect of HAZ δ-ferrite on Charpy

impact toughness
Supermartensitic stainless steel components in the oil and gas industry usually op-
erate at temperatures of the order of 100 ◦ C. However, during the rapid closure of
a well-head in an emergency, the loss of pressure can result in a rapid decrease of
temperature to about -40 ◦ C. In this case, good low temperature impact toughness
of the welded joint is necessary. The work presented in this chapter had the aim
of determining whether δ-ferrite found in dual-phase heat-affected zones of super-
martensitic stainless steels significantly affects the impact properties.
7.1 Introduction
It is well established that coarse δ-ferrite grains reduce the Charpy impact tough-
ness of low carbon semi-ferritic and martensitic stainless steels [57, 58, 59]. However,
in supermartensitic stainless steel heat-affected zones (HAZ) there is as yet no evi-
dence to support the notion that δ-ferrite is a threat to impact properties.
In the heat-affected zones of supermartensitic alloys, δ-ferrite is present mainly
in the dual-phase region of the high temperature HAZ. As this region is narrow,
typically 50 to 100 µm wide, sampling in isolation is impossible even using sub-size
specimens. To avoid this problem, attempts have been made to reproduce on a
larger scale, the microstructure of the dual-phase HAZ using a furnace and a weld
HAZ simulator.
7.2 Furnace heat-treated specimens 128
Since steel A (table 7.1) was particularly prone to δ-ferrite retention in HAZ
(chapter 5) the work presented in this chapter focuses on this material.
C Si Mn Ni Cr Mo Cu Ti N
Steel A 0.01 0.26 0.46 6.46 12.2 2.48 0.03 0.09 0.0070
Table 7.1: Chemical composition of steel A in wt%.
7.2 Furnace heat-treated specimens
7.2.a Microstructures required

To compare the effect of δ-ferrite on impact properties, a fully martensitic and
a ferritic-martensitic microstructures were required.
A furnace heat-treatment at a temperature of 1100 ◦ C for 15 minutes was cho-
sen to generate the fully martensitic microstructure. On the other hand, following
results presented in chapter 4, the temperature of 1350 ◦ C was chosen to produce
the mixed ferrite-martensite microstructure. However, due to the relatively slow
heating rate of the furnace, a difference in prior austenite grain size between the two
microstructures was expected. To remedy, that problem, another fully martensitic
microstructure (consisting of martensite and dissolved δ-ferrite) was also studied.
The latter microstructure was produced using a supercritical annealing treatment.
The heat-treatment which allowed dissolution of most of the δ-ferrite without chang-
ing the austenite grain size was carried out at 1100 ◦ C for an hour.
Besides, Charpy impact tests were also carried out on the material in its as-
received condition. In this case the microstructure consisted of tempered martensite
with about 13 % retained austenite (chapter 4).
7.2.b Procedure
Specimens of 12 mm square section and 60 mm long were machined from the
pipe of steel A in the longitudinal direction as shown in figure 7.1. The samples
were then placed in a pre-heated furnace at the center of the bottom surface and
were let in for 15 minutes before water quenching.
Figure 7.1: Orientation of the Charpy specimen within the pipe.
After optical verification of the desired microstructures, specimens were ma-

chined to the standard dimension of 10 mm square section × 55 mm with a 2 mm
deep V-notch positioned through-thickness. Charpy impact tests were finally carried
out between 20 and -200 ◦ C to reveal the ductile brittle transition and ten specimens
were used to generate each curve.
A hyperbolic tangent function was chosen to fit the transition curves :

T − T0
Impact energy (J) = A0 + B0 tanh (7.1)
C0
Where A0 , B0 , C0 and T0 are fitting constants and T is the temperature in ◦ C. This
representation of the transition curve facilitates comparison and the determination
of the ductile to brittle transition temperature (DBTT). The latter was taken as the
temperature coresponding to half the height of the transition curve.
7.2.c Characterisation of heat-treated specimens

To facilitate the interpretation of the results, heat-treated specimens were char-
acterised before impact-testing in terms of phase presence, volume fraction, hardness
and grain size.
i Microstructure
Figure 7.2 shows the microstructures of the fully martensitic, martensitic/ferritic
and annealed martensitic/ferritic specimens observed on the actual-size Charpy
specimens. Typical cuboid Ti(C,N) precipitates were the only observed non-metallic
particles. The latter were up to 10 µm long and occurred in random dispersion or
as small distinct clusters. There were no obvious differences in volume fractions and
distribution of those particles between the different microstructures.
In the martensitic/ferritic microstructure, δ-ferrite was located both at grain
boundaries and within the grains and its volume percentage measured by image
analysis was 14 % ± 2 %. This figure was comparable to the results obtained in
chapter 4.
Annealing of the mixed microstructure led to a reduction of the δ-ferrite content
to 2 % ± 0.5 % and spheroidisation of the remaining ferrite. The dissolution of
δ-ferrite took a long time despite the relatively elevated annealing temperature.
This is because significant partitioning of substitutional solutes occurred at elevated
temperature (chapter 4) and back diffusion takes time even at 1100 ◦ C.
ii Hardness and grain size

The hardness and grain size measured on the heat-treated materials are reported
in table 7.2. There was little hardness change between the different microstructures
even in the presence of δ-ferrite up to 14 %. As hardness is proportional to tensile
strength, all specimens could be assumed to have a similar strength [4].
There was a small difference in grain size between the various microstructures.
In fact, the prior austenite grain size of the specimens heat-treated at 1350 ◦ C
was about 20 µm larger than the one heat-treated at 1100 ◦ C. Due to the grain
boundary pinning effect of δ-ferrite there was no grain size difference between the
martensitic/ferritic and annealed martensitic/ferritic microstructures.
iii Retained austenite

The presence of retained austenite in the furnace heat-treated specimens has
been checked using X-ray diffraction. No austenite could be detected in any of the
specimens (figure 7.3).
Figure 7.2: Microstructures obtained after isothermal treatment at (a) 1100 ◦ C and (b) 1350 ◦ C
for 15 minutes and water quenched. (c) Microstructure obtained after treatment at 1350 ◦ C for
15 minutes and water quenched followed by annealing at 1100 ◦ C for 60 min and water quenched.
12000
Martensite
α (211)
10000
8000
Intensity
α (200)
6000
4000
α (220)
2000
0
45 60 75 90 105
2θ
12000
Martensite + ferrite
α (211)
10000
8000
Intensity
α (200) α (220)
6000
4000
2000
0
45 60 75 90 105
2θ
6000
Annealed martensite + ferrite
5000
α (211)
4000
Intensity
α (200)
3000
2000
α (220)
1000
0
45 60 75 90 105
2θ
Figure 7.3: X-ray diffraction patterns of the fully martensitic, martensitic/ferritic and annealed
ferritic/martensitic microstructures.
Microstructure Hardness Grain size / µm
Tempered martensite + retained austenite 296 ± 3 25 ± 4

Martensite 317 ± 8 60 ± 9
Martensite + δ-ferrite 311 ± 8 80 ± 12
Annealed Martensite + δ-ferrite 304 ± 7 80 ± 12
Table 7.2: Macro-hardness (HV10kg ) and grain size of the studied microstructures.
7.2.d Results of Charpy impact tests

Table 7.3 summarises the measured ductile to brittle transition temperatures
(DBTT) for each tested microstructure.
Microstructure DBTT (◦ C)
Tempered martensite + retained austenite -98

Martensite -98
Martensite + ferrite -46
Annealed martensite + ferrite -86
Table 7.3: Ductile to brittle transition temperatures obtained for each studied microstructure.
Figure 7.4 shows the impact transition curves obtained from the as-received, fully
martensitic and martensitic/ferritic microstructures. The fitted transition curves of
the as-received and fully martensitic microstructures were similar. Their DBTT
were also identical and far below 0 ◦ C. However, the as-received microstructure had
slightly higher failure energies at low temperatures. This is because austenite that
remains stable down to -196 ◦ C (chapter 4) tends to transform to martensite during
impact and thus absorbs some energy and raise the toughness as proposed by Bilmes
et al. [42]. At room temperature there was more scatter in the energy required to
break the specimens but there was apparently no significant difference between the
two microstructures.
Similarly, there was no difference between the high temperature failure energies of
the fully martensitic and martensitic/ferritic microstructures (figure 7.4). However,
a significant shift in ductile to brittle transition temperature was observed. In fact,
the DBTT of the martensitic/ferritic microstructure was raised by about 52 ◦ C to
about -46 ◦ C. The origin of this increase could not be attributed directly to the
presence of δ-ferrite since there was a 20 µm difference in prior austenite grain size
between the two microstructures.
The observation of the transition curves of the martensitic/ferritic and annealed
martensitic/ferritic microstructures of identical grain size removed this uncertainty
(figure 7.5). In fact the microstructure containing a reduced amount of δ-ferrite had
significantly better DBTT (-86 ◦ C compared to -46 ◦ C). This consequently showed
that for a given prior austenite grain size the presence of δ-ferrite significantly affects
the transition temperature.
Figure 7.6 compares the transition curves of the annealed martensitic/ferritic

(grain size 80 µm) with the tempered microstructure (grain size 25 µm). Annealing
clearly improved the notch toughness to a level comparable to that of the as-received
condition. This indicated that the effect of the prior austenite grain size in the 25 -
80 µm range on the ductile to brittle transition temperature was not significant.
7.2.e Fractography
Observation of the fracture surfaces in the SEM revealed a variety of fracture
modes. In particular, ductile rupture, cleavage and quasi-cleavage could be identi-
fied. Mc Donald et al. summarise these fracture modes as follows [90, 91] :
1. Ductile rupture. Dimples resulting from microvoid formation and coalescence,

are characteristic of ductile rupture. Microvoids are generally initiated at
inclusions which can sometimes be observed within the dimples.
2. Cleavage. Stepped facets are caused by fracture through regions of differ-

ent crystallographic orientation. Cleavage occurs by rapid propagation of a
250
Tempered martensite
Martensite
Impact energy / J 200
150
100
50
0
-200 -180 -160 -140 -120 -100 -80 -60 -40 -20 0 20
o
Temperature / C
Figure 7.4: Charpy impact energy curves obtained on the tempered martensite + retained
austenite (as-received condition), fully martensitic and martensitic/ferritic microstructures.
250
200
Impact energy / J
150
100
50
0
-200 -180 -160 -140 -120 -100 -80 -60 -40 -20 0 20
o
Temperature / C
Figure 7.5: Charpy impact energy curves obtained on the martensitic/ferritic and annealed
martensitic/ferritic microstructures. Note that both microstructures have identical prior austenite
grain size of 80 µm.
250
Tempered martensite
200
Impact energy / J
150
100
50
0
-200 -180 -160 -140 -120 -100 -80 -60 -40 -20 0 20
o
Temperature / C
Figure 7.6: Comparison of the Charpy impact energy curves between the as-received and annealed
martensitic/ferritic microstructures.
crack along crystallographic planes, and may be initiated by inclusion/matrix

decohesion. This is a very low energy-absorbing mode of fracture.
3. Quasi-cleavage. A less well-defined fracture mode, known as quasi-cleavage, is

associated with tear ridges which represent limited plastic deformation sepa-
rating cleavage facets. Unlike pure cleavage, energy is absorbed as a result of
the limited plastic deformation.
Examples of ductile rupture obtained from the fully martensitic and marten-
sitic/ferritic microstructures tested at 0 ◦ C are shown in figure 7.7. Most particles
associated with initiation were cracked. These particles were easily identified as
titanium carbo-nitride by EDX and also due to their typical cuboid shape.
Specimens tested at low temperatures fractured in a transgranular manner (fig-
ure 7.8). This indicated that the presence of grain boundaries δ-ferrite did not
embrittle the latter. In general, the cleavage facet sizes appeared similar but the
martensitic/ferritic microstructure showed more secondary cracks.
Figure 7.7: Dimpled fracture surfaces of the fully martensitic (a), and martensitic/ferritic (b)
microstructures tested at 0 ◦ C.
Figure 7.8: Fracture surfaces of the fully martensitic (a) and martensitic/ferritic (b) microstruc-
tures tested at -150 and -100 ◦ C respectively.
7.3 Gleeble simulated HAZ 138
Titanium carbo-nitride particles were often seen to initiate cracks in all studied
microstructures. River markings on the cleavage facets of the specimens showed
evidence of both cleavage (figure 7.9) and quasi-cleavage fracture (figure 7.10) ini-
tiated by Ti(C,N) inclusions. The capacity of these particles to reduce toughness is
well known. Brittle fracture induced by the cleavage of TiN inclusions has recently
been investigated by Fairchild et al. [92, 93]. TiN particles are well bonded to the
matrix and thus allow high stresses in a notch tip plastic zone to act on inclusions
without debonding the interface. Once an inclusion cleaves, the strong bond allows
for transfer of the crack into the martensitic matrix.
In addition to the SEM observation, cross sections perpendicular to the fracture

surfaces, as shown in figure 7.11, have been observed to reveal the fracture path.
Figures 7.12 and 7.13 show example of such cross sections for specimens of the
martensitic/ferritic and annealed martensitic/ferritic microstructures that failed by
cleavage. In the martensitic/ferritic specimens, secondary intragranular cracks prop-
agated through martensite and δ-ferrite stringers, while shorter cracks were seen to
initiate both in martensite and at the δ-ferrite martensite interface in the annealed
microstructure. In the latter case, cracks could not easily link up when δ-ferrite was
more dispersed and possessed a spheroidised morphology, thus explaining the better
impact properties.
7.3 Gleeble simulated HAZ
The furnace heat-treated specimens allowed the impact properties of the marten-
sitic/ferritic microstructure to be studied. To investigate the effect of δ-ferrite on
HAZ toughness in a situation closer to actual welds, a Gleeble HAZ simulator (de-
scribed in chapter 3) has been used.
7.3.a Procedure
Ten specimens of dimension 11×11×100 mm were sent to Corus Swinden Tech-
nology Centre for HAZ simulation. The operation was performed at the maximum
temperature that the equipment could achieved i.e. 1350 ◦ C, using a fast cooling
Figure 7.9: Fracture surfaces of the ferritic/martensitic microstructure tested at -100 ◦ C showing
Ti(C,N) initiated cleavage.
Figure 7.10: Fracture surfaces of the ferritic/martensitic microstructure tested at -100 ◦ C show-
ing Ti(C,N) initiated quasi-cleavage.
Fracture surface
Observation plane
Figure 7.11: Location of transverse section in the Charpy specimens.
rate (cooling time between 800-500 ◦ C of 10 s). After simulation, all specimens were
polished and etched to reveal the microstructure and to decide the notch location
for the final machining to standard size Charpy specimens.
7.3.b Microstructure
The typical microstructure of the simulated specimens is shown in figure 7.14
(a). The central part of the specimens showed three high temperature HAZ regions
and therefore indicated the existence of a peak temperature gradient in the middle
of the specimens. The coarse grained region, figure 7.14 (b), had semi-circular shape
and depth of 2.5 mm. The martensite + δ-ferrite region, figure 7.14 (c), surrounded
the previous region and in turn was surrounded by the prior fully austenitic area.
The simulated specimens showed a range of grain size. The coarse-grained and
dual-phase regions had prior δ-ferrite grain size of about 160 µm and 30 µm, re-
spectively. Similar to the microstructure observed in the dual-phase HAZ of the low
heat input single-pass welds of chapter 5, δ-ferrite was present at grain boundaries
and as films within the prior austenite grains. It could be noted that the width of
the dual-phase region was significantly larger compared to actual weld HAZs. In
addition, the volume percentage of δ-ferrite measured by image analysis at a mag-
nification of 200 was also larger than that of single-pass weld HAZs i.e. 40 % ±
5 %.
Figure 7.12: Cross section of the fracture surface of the martensitic/ferritic microstructure tested at
-100 ◦ C. Electrolytically etched to reveal δ-ferrite.
Figure 7.13: Cross section of the fracture surface of the annealed martensitic/ferritic microstructure
tested at -150 ◦ C. Electrolytically etched to reveal δ-ferrite.
Figure 7.14: Microstructure of Gleeble simulated specimens. (a) Low magnification of the
central region, (b) close-up of the coarse-grained region and (c) close-up of the dual-phase region.
All specimens were electrolytically etched to reveal δ-ferrite.
7.3.c Results
The transition curve obtained on the Gleeble simulated specimens is presented
on figure 7.15. For comparison the transition curves of the as-received and marten-
sitic/ferritic microstructures have been reported. The Gleeble specimens had the
lowest upper shelf energy of the three curves. The latter was probably due to the
combined action of large grain size and presence of δ-ferrite in the tested microstruc-
ture. More importantly, the simulated HAZ had a ductile to brittle transition tem-
perature comparable to that of the as-received material (-94 ◦ C and -98 ◦ C respec-
tively) although like the ferritic/martensitic microstructure the transition occurred
over a shorter temperature range. The behaviour of the Gleeble simulated specimens
at low temperatures was therefore significantly better than the ferritic/martensitic
microstructure studied in the first part.
250
Tempered martensite
Gleeble
200
Impact energy / J
150
100
50
0
-200 -180 -160 -140 -120 -100 -80 -60 -40 -20 0 20
o
Temperature / C
Figure 7.15: Charpy impact energy curves obtained using the Gleeble and comparison with real
weld HAZ.
7.4 Summary 144
7.3.d Fractography
All specimens tested at temperatures above -100 ◦ C showed ductile failure mode
while those broken at temperatures below -100 ◦ C showed completely brittle be-
haviour. Interestingly the Gleeble specimen tested at -100 ◦ C showed two fracture
modes (figure 7.16). Observation of a cross section revealed the coarse-grained re-
gion to fail by cleavage whilst the dual-phase region showed evidence of ductility
below the transition temperature (figure 7.17). The ductility of the dual-phase HAZ
indicates that the DBTT of this region is lower than -100 ◦ C. These results provide
evidence to support that the presence of δ-ferrite in dual-phase HAZ up to 40 %
does not significantly affect the impact properties of the whole high temperature
HAZ. Clearly the specimen transition temperature was strongly influenced by the
grain size of the coarse-grained HAZ.
7.4 Summary
Part I of the study revealed that for a similar strength, particle volume-fraction
and distribution, and a given prior austenite grain size, the ductile to brittle tran-
sition temperature was very sensitive to the volume fraction of δ-ferrite stringers in
the microstructure. On the other hand, the effect of prior austenite grain size in the
25 - 80 µm range on the transition temperature was small. It was therefore clear
that the presence of δ-ferrite up to 14 % had a greater influence on toughness than
austenite grain size variation in the 25 - 80 µm range.
On the other hand, the tests carried out on the Gleeble simulated specimens re-
vealed that the dual-phase HAZ does not play a major role in determining the DBTT
even when δ-ferrite is present up to 40 %. This is apparently because the small grain
size of the dual-phase region offers suitable obstacles to cracks propagation. With
a prior δ-ferrite grain size of 160 µm, the coarse-grained region strongly influences
the impact properties of the high temperature HAZ but the transition-temperature
was raised by only few degrees compared with the as-received microstructure. As
multipass welding contributes to the reduction of δ-ferrite content in HAZ the effect
7.4 Summary 145
Figure 7.16: Fracture surfaces of the coarse-grained (a) and dual-phase region (b) of the Gleeble
simulated specimen tested at -100 ◦ C.
Figure 7.17: Cross section of the fracture surface showed in figure 7.16. Coarse-grained (a) and dual-
phase region (b). This cross section was taken perpendicular to the examination plane shown in figure
7.11.
7.4 Summary 146
of this phase on the transition temperature of actual welds should be even smaller.
It should be noted that no attempts have been made to evaluate the effect of
Widmanstätten austenite in coarse-grained region on the impact properties. How-
ever, due to its morphology and coarseness, Widmanstätten austenite might be
detrimental. As fast cooling rates minimise its volume fraction (chapter 4) it could
be advised to weld supermartensitic stainless steels using low heat inputs to generate
fast cooling rates and minimise grain growth.
The results presented in this work have been obtained on a highly alloyed
titanium-stabilised supermartensitic stainless steel. Medium and lean alloys are
less prone to δ-ferrite formation in dual-phase HAZ but might have larger coarse-
grained HAZ and harder martensite for a given C and N content (chapter 4). In
this case the high temperature HAZ impact properties are likely to be governed by
the coarse-grained HAZ grain size and the potential presence of Ti(C,N) particles.
Chapter 8
Neural network model for

estimating the Ac1 temperature
8.1 Introduction
Although efforts are being devoted to avoid the need for post-weld heat-treatments
(PWHT) in supermartensitic stainless steels, this operation is still often employed
to reduce martensite hardness and thus improve the heat-affected zone sulfide stress-
corrosion cracking resistance. The tempering operation should be carried out few
degrees below the steel Ac1 temperature in order to avoid the formation of austenite
which may subsequently transform into untempered martensite on cooling to room
temperature.
In practice, when an estimation of the Ac1 temperature is required empirical
formulae are often employed [38]. However, such relationships often neglect poten-
tial interactions between elements and do not have a mechanism to indicate the
uncertainties of extrapolation.
A more general method of regression analysis involves neural networks within
a Bayesian framework. Gavard et al. employed neural networks to model Ac1 and
Ac3 temperatures of a wide range of steels as a function of their chemical compo-
sition and of the heating rate used to measure the temperatures [94]. Using this
approach, it was possible to satisfactorily predict phase transformation tempera-
tures and reproduce known metallurgical trends, although Ac1 temperatures were
harder to model than Ac3 . This is because of the lack of good data and also the
8.2 Introduction to neural network 148
neglect of the initial microstructure as an input to the model.

While the number of experimental Ac1 data can easily be increased, the mi-
crostructure is rarely reported in the published literature. Because of the latter,
predicting the Ac1 temperature of a weld HAZ is difficult but as a first step it is
still interesting to improve the original neural network model and thus obtain better
guidance than with linear regression equations.
8.2 Introduction to neural network
A neural network is a powerful non-linear regression tool which when imple-

mented within a Bayesian framework allows the calculation of error bars represent-
ing the uncertainty in the fitting parameters. To explain how a neural network works
it is interesting to first introduce conventional regression analysis.
8.2.a Regression analysis

In conventional regression analysis the data are best-fitted to a specified rela-
tionship which is usually linear. The result is an equation in which each of the input
xi is multiplied by a weight wi ; the sum of all such products and a constant θ then
gives an estimate of the output :

y= wi xi + θ (8.1)
i
Equation 8.2 is an example of such linear regression used to estimate the Ac1
temperature of low carbon 13 wt% Cr steels as a function of chemical composition
(in wt%) [38].
Ac1 (◦ C) = −1500(C + N) − 50Ni − 25Mn + 25Si + 25Mo + 20Cr + 650 (8.2)

8.2 Introduction to neural network 149
8.2.b Neural networks

In neural network analysis, an expression similar to that of a linear regression

(i.e. i wi xi + θ) forms the argument of, in the present work, a hyperbolic tangent
function and gives the output of a so-called hidden unit. The output of a hidden unit
is therefore a non-linear function of the inputs and the hyperbolic tangent function
is chosen because of its flexibility. To capture more complex relationships, neural
networks are often composed of several hidden units and each of them possesses its
own set of adjustable weigths and constant as :

(2) (2)
hj = tanh wij xi + θj (8.3)
i
The output y of the neural network model is then simply obtained by a linear
combination of the output of each hidden unit and by adding another constant θ(1)
as :
(1)
y= wij hj + θ(1) (8.4)
j
Figure 8.1 illustrates the structure of a three layers neural network consisting of
the inputs, the hidden units and the output. The existence of the hidden layer allows
the model to identify interaction between variables. Notice that the complexity of
the model is related to the number of hidden units. The weights and constants
used in the hidden units are adjusted during the training of the model. The latter
operation is made using an algorithm which implements a learning method based
on Bayesian statistics.
8.2.c Bayesian method

In regression analysis, the most probable values of the weigths are used for a
given model. This method results in an overall error obtained by comparing the
predictions against experimental values, with no indication of the uncertainty as a
function of position in the input space.
MacKay has developed a useful treatment of neural networks using Bayesian
statistics that recognises that there are many functions which can be fitted or ex-
trapolated into uncertain regions of the input space, without compromising the fit
8.3 Analysis 150
Inputs Hidden layer Outputs

x1
h1
x2
h2 y
hj
xn
Figure 8.1: Schematic illustration of the input, hidden and output layers of a neural network
model of n inputs and j hidden units. Each arrow represents a weight.
in adjacent regions which are rich in accurate data [95]. Instead of calculating the
best set of weights, the algorithm infers a probability distribution for the weights
from the data presented. When making predictions, the variety of model solutions
corresponding to different possible set of weights are averaged using the probability
of these sets of weights, a process called marginalising. It results that it is possible
to quantify the uncertainty of fitting. If the inferred distribution is sharply peaked
in the weight space, the most probable set will give by far the largest contribution to
the prediction and alternative solutions will have little importance. Consequently,
the prediction will be associated with a small uncertainty. If on the contrary, the
data are such that different sets of weights are equally probable, alternatives will
contribute in similar proportions and the error bars will be large. This is typically
the case in regions of the input space where data are scarce or exceptionally noisy
as illustrated in figure 8.2.
8.3 Analysis
8.3.a The variables

The dataset employed in this work is from earlier work by Gavard et al. compiled
from published literature to include some 788 experiments [94]. In addition to the
8.3 Analysis 151
Output
A B
Input
Figure 8.2: Schematic illustration of the uncertainty in defining a fitting function in regions
where data are sparce (B) or where they are noisy (A). The thinner lines represent error bounds
due to uncertainties in determining the weigths [96].
heating rate, the original input variables consisted of the elements C, Si, Mn, Cu,
Ni, Cr, Mo, Nb, V, W and Co together with the trace elements S, P, Ti, Al, B, As,
Sn, Zr, N and O. A Gaussian process analysis by Bailer-Jones et al. revealed that
the dataset could be reduced to the 11 major elements without significantly affecting
the accuracy of the predictions [97]. Following these results, the set of inputs for the
present work (table 8.1) included the chemical composition of the first 11 elements
(in wt%), the heating rate (◦ C s−1 ) and measured Ac1 temperatures (◦ C). It is
emphasised that unlike linear regression analysis, the ranges stated in table 8.1
cannot be used to define the range of applicability of the neural network model.
This is because the inputs are in general expected to interact.
A total of 88 sets of experimental data were added to the original database

including data generated during this project and some collected from published
literature on 9 - 13 wt% Cr steels [35, 59, 72, 98, 99, 100]. The experimental Ac1
temperatures that have been added to the original database have been measured on
steels as-received microstructures (i.e. tempered martensite and possibly retained
austenite). The final database contained 876 experiments and a total of 113 data
8.3 Analysis 152
were from 9 - 13 wt% Cr steels. A visual impression of the spread of the data is
shown in figure 8.3.
Input element Minimum Maximum Mean Std. deviation
C 0.0 0.96 0.28 0.17

Si 0.0 2.83 0.39 0.45
Mn 0.0 3.06 0.81 0.39
Cu 0.0 2.01 0.05 0.14
Ni 0.0 9.12 1.08 1.61
Cr 0.0 17.97 2.03 3.54
Mo 0.0 4.80 0.35 0.43
Nb 0.0 0.52 0.00 0.04
V 0.0 2.45 0.06 0.13
W 0.0 8.59 0.08 0.49
Co 0.0 4.07 0.06 0.40
Heating rate 0.01 100.0 2.37 8.36
Ac1 520.0 921.0 729.6 57.4
Table 8.1: Input variables used to develop the new model. Composition in wt%, heating rate in
◦
C s−1 and Ac1 in ◦ C.
8.3.b Normalisation
All the variables were normalised within a range ± 0.5 as follows :
(x − xmin )
xN = − 0.5 (8.5)
(xmax − xmin )
where xN is the normalised value of x, which has the minimum and maximum values
given by xmin and xmax , respectively. The normalisation operation facilitates the
subsequent comparison of the relative significance of the variables.
8.3 Analysis 153
1000 1000 1000

a) b) c)
C
Measured Ac1 / oC
C
900 900 900
o
o
Measured Ac1 /
Measured Ac1 /
800 800 800
700 700 700
600 600 600
500 500 500
400 400 400

0 0.10.20.30.40.50.60.70.80.9 1 0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3 3.5
Carbon / wt% Silicon / wt% Manganese / wt%
1000 1000 1000

d) e) f)
C
C
900 900
Measured Ac1 / oC
900
o
o
Measured Ac1 /
Measured Ac1 /
800 800 800
700 700 700
600 600
600
500 500
500
400 400
0 0.5 1 1.5 2 2.5 400 0 2 4 6 8 10 12 14 16 18
0 1 2 3 4 5 6 7 8 9 10
Copper / wt% Chromium / wt%
Nickel / wt%
1000 1000 1000

g) h) i)
C
Measured Ac1 / oC
C
900 900 900
o
o
Measured Ac1 /
Measured Ac1 /
800 800 800
700 700 700
600 600 600
500 500 500
400 400 400

0 1 2 3 4 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.5 1 1.5 2 2.5 3
Molybdenum / wt% Niobium / wt% Vanadium / wt%
1000 1000 1000

j) k) l)
C
900 900 900

o
o
Measured Ac1 /
Measured Ac1 /
Measured Ac1 /
800 800 800
700 700 700
600 600 600
500 500 500
400 400 400

0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 0 20 40 60 80 100
Tungsten / wt% Cobalt / wt% Heating rate / oC s-1
Figure 8.3: The database values of each variable versus the Ac1 temperature.
8.3 Analysis 154
8.3.c Overfitting
Because of the great flexibility of the functions used in the neural network, there
is a possibility of overfitting data. This can occur when the generated models become
too complex for predicting a simple trend (figure 8.4). To avoid overfitting two
solutions are implemented. The first is contained in the algorithm due to MacKay
: complexity parameters α and β (see equation 8.6) are inferred from the data and
allow automatic control of the model complexity. The second consists of equally
dividing the experimental database into a training and a test dataset. The training
set is used to create a large number of models with different number of hidden units
(1 to 22) and different sets of initial weights (6 in this case). The test data set is used
to see how the trained models generalise on unseen data. The error on the testing
set decreases at first, as the fitting improves, but increases again when overfitting
occur (figure 8.5 (b)).
3
y
0
0 1 2 3 4 5 6
x
Figure 8.4: A complicated model may overfit the data. In this case, a linear relationship is all
that is justified by the noise in the data.
8.3 Analysis 155
8.3.d Training and testing of models

The training of a model involves the derivation of the weights which minimise
an objective function, typically :
M (w) = βED + αEW

1 (m) 2
ED = t − y (m)
2 m
1 2
EW = w (8.6)
2 m m
where w is a vector describing the set of weights, ED is the overall error, and EW
the regulariser, used to force the network to use small weights. α and β are control
parameters which influence the complexity of the model. t(m) is the target in the
database for the set of inputs x(m) , while y (m) is the corresponding predicted output.
The perceived noise level during training tends to decrease as the model becomes
more complex figure 8.5 (a). The models are then tested on the unseen test data
and are subsequently ranked using either the test error (TE) as in figure 8.5 (b), or
the log predictive error (LPE) (figure 8.5 (c)). The TE and LPE are calculated as :
1 2
TE ∝ t(m) − y (m) (8.7)
m
2

√
1 (m)
(m) 2 (m) 2 (m)
LPE = t −y /σy + log 2π σy (8.8)
m
2
(m) 1
where σy (= β 1/2
) is related to the uncertainty of fitting for the set of inputs x(m) .
As predictions with error bars are preferred to simple scalar predictions, it is more
judicious to compare models in term of their predictive performance as measured by
the LPE. Under the LPE, as contrasted with the TE, the penalty of making a wild
prediction is much less if that wild prediction is accompanied by appropriately large
error bars [101]. The best models should have high LPE values. Figure 8.5 (e) and
(f) show normalised predicted values versus experimental values for the best model
possessing five hidden units.
8.3 Analysis 156
8.3.e Committee
Combining the output of several models, by building a committee often makes
optimum predictions. The predictions, ȳ, of a committee is the average of the
prediction of its members, and the associated error bar is calculated as :
1 (l)
ȳ = y
L l
1 (l) 2 1 (l) 2
σ2 = σy + y − ȳ (8.9)
L L
l l
Where L is the number of networks in the committee. It should be noted that

equation 8.9 considers the prediction for a given, single set of inputs and that the
exponent (l) refers to the model used to make prediction.
The optimum size of a committee is determined by including an increasing num-

ber of models in the committee and by comparing their performance on the testing
set. A plot of the test error of the committee versus its size L gives a minimum error
when the committee contains more than one model as shown on figure 8.5 (d). The
committee with two models possessing five hidden units each was found to have an
optimum membership with the smallest combined test error.
After selection, the models in the committee are retrained (weights adjusted) on
the entire database without changing the complexity of each model. The committee
of models is generally referred to as the “neural network model”. The predictions
of the latter on the complete database show few outliers as illustrated in figure 8.7.
Because the algorithm used to train the model includes an automatic relevance
determination, variables that are found not to influence the output are assigned small
weights. This therefore helps identify the most important variables in the analysis.
Figure 8.8 illustrates the significance of each of the input variables in the two best
models, as perceived by the neural network, in influencing the Ac1 temperature.
The magnitude of the significance reflects the extent to which a particular input
explains the variation of the output. As expected, the nickel and chromium content
8.4 Application of the model 157
as well as the heating rate feature prominently, on the other hand, niobium, cobalt
and tungsten are not seen to be influencial.
8.4 Application of the model
The neural network can capture interactions between the inputs because the
functions involved are non-linear. The nature of these interactions is implicit in the
values of the weights, but the weights are not always easy to interpret. For example,
there may exist more than just a pairwise interaction, in which case the problem
becomes difficult to perceive from an examination of the weights. A better method
is to actually use the network to make predictions and to see how these interactions
depend on various combinations of inputs.
8.4.a Effect of chemical composition and heating rate

The effects of element additions on Ac1 have been examined on steel C, a medium
alloyed supermartensitic steel of the following chemical composition : 0.01 C, 0.19
Si, 0.24 Mn, 0.06 Cu, 3.12 Ni, 12.5 Cr, 0.02 Mo, 0.009 Nb, 0.03 V, 0.04 Co wt%.
For these predictions calculations were made at a heating rate of 1 ◦ C s−1 which was
close to the database mean heating rate and were made on the maximum range of
data present in the database.
It is emphasised that Ti and N are not taken into account in this analysis, but
as these elements combine to form very stable TiN particles the error associated
with their neglection is expected to be small. However, if Ti is present in quantities
greater than 4×N (stoichiometric ratio in wt%) then the appropriate amount of
carbon should be removed to allow TiC to form. Conversely, if nitrogen is in excess
it will combined with nitride forming elements such as Cr, Mo and V but considering
that the excess will be smaller than 0.01 wt% (in supermartensitic stainless steels),
its effect on Ac1 is probably negligible.
Figures 8.9 and 8.10 presents the results of all calculations. Some predictions
0.1 5
a) b)
0.08 4
Noise level
Test Error
0.06 3
0.04 2
0.02 1
0 0
0 5 10 15 20 25 0 5 10 15 20 25
Number of hidden units Number of hidden units
700 1.05
c) d)
Log Predictive Error
Combined Test Error

650 0.95
0.9
600 0.85
0.8
550 0.75
0.7
500 0.65
0 5 10 15 20 25 0 5 10 15 20 25 30 35 40
Number of hidden units Number of models in committee
Figure 8.5: The perceived noise level (a), the test error (b), the log predictive error (c) of the
models with increasing number of hidden units, the combined test error (d) for an increasing
number of models in committee.
Training set Testing set

0.6 0.6
Predicted values, normalised
Predicted values, normalised
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 -0.6 -0.4 -0.2 0 0.2 0.4 0.6
Database values, normalised Database values, normalised
Figure 8.6: Plots of the predicted Ac1 temperature against experimental values (normalised)
for the training and testing of the best model. The plotted error bars correspond to the fitting
uncertainties.
1000
Predicted Ac1 / oC
900
800
700
600
500
500 600 700 800 900 1000
Measured Ac1 / oC
Figure 8.7: Predicted against measured Ac1 temperatures performed on the complete database
using the optimum committee. The plotted error bars correspond to ± 1 σ as defined in equation
8.9.
Figure 8.8: Perceived significance of input parameters in the best two models for each input
variable. H.R. stands for heating rate.
showed clear trends in the evolution of Ac1 as a function of the input data considered.
This was namely the case for manganese and nickel (figure 8.9 (c) and (e)). Those
solutes are both well-known austenite stabilising elements, and consistent with the
model predictions, their addition decrease Ac1 .
No clear trends could be observed on the remaining elements. However, the
model provides predictions with error bars that are dependent upon the input space
region and this represents a significant improvement compared with traditional lin-
ear regression equations. For example, there is little information in the database
for vanadium in concentration greater than 0.5 wt% (figure 8.10 (i)) and the model
recognises this by yielding increasingly large error bars. On the other hand, exam-
ination of the model inferred significance (figure 8.8) reveals that Nb, W and Co
have little influence on Ac1 and thus could be removed in a future analysis.
The evolution of Ac1 as a function of heating rate is reported in figure 8.11. A

clear increase of Ac1 temperature is identified by the model at heating rates up to
∼ 20 ◦ C s−1 . The initial temperature increase is explained by the diffusive nature of
the transformation. In fact, the number of successful atomic jumps per unit of time
of substitutional elements across the martensite/austenite interface is reduced at
faster heating rates so that the transformation starts at a higher temperature. The
model successfully recognised the effect of heating rate on Ac1 temperature while
linear regression equations usually disregard this effect.
8.4.b Application to steel T91

To show the performance of the new model on martensitic stainless steels, com-
parison has been made using steel T91 (table 8.2), an alloy studied in Gavard et al.
work [94]. It should be noted that the data from this alloy were not included in the
database.
As can be seen from figure 8.12, the new neural network model shows small error
bars and good agreement with the experimental data except at 20 ◦ C s−1 . However
it should be noted that no experimental error bars were provided but assuming those
to be ± 10 ◦ C, all predictions become satisfactory.
900 900
a) b)
C
C
850 850
o
o
Predicted Ac1 /
Predicted Ac1 /
800 800
750 750
700 700
650 650
600 600
0 0.05 0.1 0.15 0.2 0 0.5 1 1.5 2 2.5 3
Carbon / wt% Silicon / wt%
900 900
c) d)
C
C
800 850
o
o
Predicted Ac1 /
700 Predicted Ac1 / 800
600 750
500 700
400 650
300 600
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2
Manganese / wt% Copper / wt%
900 900
e) f)
C
850 850
o
o
Predicted Ac1 /
Predicted Ac1 /
800 800
750 750
700 700
650 650
600 600
0 1 2 3 4 5 6 7 8 9 0 2 4 6 8 10 12 14
Nickel / wt% Chromium / wt%
Figure 8.9: Predicted variation in Ac1 temperature as a function of chemical composition for C,
Si, Mn, Cu, Ni and Cr. Prediction made on steel C : 0.01 C, 0.19 Si, 0.24 Mn, 0.06 Cu, 3.12 Ni,
12.5 Cr, 0.02 Mo, 0.009 Nb, 0.03 V, 0.04 Co wt%. The neural network error bars correspond to ±
1 σ as defined in equation 8.9.
900 900
g) h)
C
C
850 850
o
o
Predicted Ac1 /
Predicted Ac1 /
800 800
750 750
700 700
650 650
600 600
0 0.5 1 1.5 2 2.5 3 0 0.1 0.2 0.3 0.4 0.5
Molybdenum / wt% Niobium / wt%
1000 900
i) j)
C
C
950 850
o
o
Predicted Ac1 /
900 Predicted Ac1 / 800
850 750
800 700
750 650
700 600
0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5 3 3.5 4
Vanadium / wt% Tungsten / wt%
900
k)
C
850
o
Predicted Ac1 /
800
750
700
650
600
0 0.5 1 1.5 2 2.5 3 3.5 4
Cobalt / wt%
Figure 8.10: Predicted variation in Ac1 temperature as a function of chemical composition for
Mo, Nb, V, W and Co. Prediction made on steel C : 0.01 C, 0.19 Si, 0.24 Mn, 0.06 Cu, 3.12 Ni,
12.5 Cr, 0.02 Mo, 0.009 Nb, 0.03 V, 0.04 Co wt%. The neural network error bars correspond to ±
1 σ as defined in equation 8.9.
900
C
850
o
Predicted Ac1 /
800
750
700
650
600
0 10 20 30 40 50
o -1
Heating rate / C s
Figure 8.11: Effect of heating rate on Ac1 temperature for a medium alloyed supermartensitic
stainless steel of compositions (wt%): 0.01 C, 0.19 Si, 0.24 Mn, 0.06 Cu, 3.12 Ni, 12.5 Cr, 0.02 Mo,
0.009 Nb, 0.03 V, 0 W, 0.04 Co.
1000
C
950
o
Predicted Ac1 /
900
850
800
750
700
0 10 20 30 40
o -1
Heating rate / C s
Figure 8.12: Predicted and experimental Ac1 temperatures as a function of heating rate for steel
T91.
Material C Si Mn Cr Mo Ni N Nb V
T91 0.105 0.43 0.37 8.20 0.97 0.13 0.051 0.075 0.2
Table 8.2: Chemical composition of steel T91. Compositions in wt % [94].
8.4.c Application to supermartensitic steels

A range of supermartensitic stainless steels has been chosen to make prediction
over their Ac1 temperatures. The selection of alloys (table 8.3) is ranked as increas-
ing (Cr+Mo)/Ni ratio to separate high (small ratio) and low alloy steels. Steels A,
B and C that have been studied in the previous chapters are included in the list:
(Cr+Mo)/Ni ratio of 2.27, 7.04 and 4.01 respectively.
Figure 8.13 (a) shows the results of all predicted Ac1 temperatures for a heating
rate of 1 ◦ C s−1 . As expected steels possessing a (Cr+Mo)/Ni ratio between 2
and 3 (high nickel alloys) had lower Ac1 temperatures compared with leaner alloys.
According to these predictions, tempering at the common temperature of 650 ◦ C
would lead to partial reaustenitisation in high alloy steels but it would not be so
at a heating rate of 10 ◦ C s−1 (figure 8.13 (b)). Faster heating rates allow therefore
the use of higher tempering temperatures and as a consequence of atoms higher
mobility, shorter tempering time. This illustrates the importance of taking heating
rate into account when tempering those steels.
8.4.d Effect of the microstructure and discussion

In practice, the HAZ regions that need tempering are those that have been
completely reaustenitised during welding. However, the additional experimental
Ac1 temperatures have been obtained on steels in their as-received conditions. It
is therefore interesting to estimate the effect of the initial microstructure on Ac1 in
supermartensitic stainless steels.
To investigate the effect of the microstructure on Ac1 , a dilatometric test using

two cycles has been performed. The retained austenite-free as-received microstruc-
Cr+M o
Ni
C Si Mn Cu Ni Cr Mo Nb V Co Ti N
2.27 0.01 0.26 0.46 0.03 6.46 12.2 2.48 0.01 0.03 0.03 0.09 0.007
2.30 0.009 0.20 0.43 0.03 6.40 12.2 2.51 0.009 0.03 0.03 0.12 0.005
2.47 0.012 0.18 0.27 0.03 5.90 12.4 2.19 0.009 0.03 0.01 0.024 0.012
2.70 0.017 0.20 1.77 0.51 4.87 11.7 1.39 0.005 0.02 0.0 0.007 0.011
2.71 0.013 0.12 0.54 0.04 5.10 11.8 2.03 0.009 0.04 0.06 0.004 0.009
2.80 0.013 0.16 0.65 0.03 5.10 13.5 0.78 0.009 0.04 0.03 0.088 0.006
2.93 0.021 0.17 1.79 0.47 4.40 11.5 1.39 0.009 0.02 0.03 0.008 0.012
3.12 0.02 0.30 0.46 1.24 3.98 11.3 1.12 0.04 0.06 0.04 0.02 0.008
3.52 0.02 0.22 0.39 0.01 3.91 12.8 0.97 0.009 0.02 0.03 0.009 0.051
4.01 0.01 0.19 0.24 0.06 3.12 12.5 0.02 0.009 0.03 0.04 0.01 0.012
4.23 0.011 0.19 0.30 0.08 2.91 12.3 0.02 0.009 0.04 0.02 0.013 0.009
4.34 0.011 0.10 1.77 0.48 2.45 10.6 0.06 0.006 0.02 0.0 0.006 0.006
4.67 0.011 0.11 1.83 0.50 2.26 10.5 0.06 0.009 0.02 0.01 0.004 0.006
7.04 0.01 0.18 1.14 0.49 1.55 10.9 ≤ 0.01 0.009 0.05 0.04 0.01 0.006
Table 8.3: Chemical composition of steels used to make predictions ranked as increasing
(Cr+Mo)/Ni ratio. All compositions in wt% [30, 102, 8].
ture of steel B was heated up to 1000 ◦ C using a heating rate of 1 ◦ C s−1 and was
held at temperature for 10 minutes before cooling (cycle 1). A second cycle on the
same specimen allowed the measurement of the Ac1 temperature of fresh martensite
(cycle 2).
As figure 8.14 (a) shows there is no significant differences between the two cycles
i.e. Ac1 is about 730 ◦ C. This result indicates that Ac1 is not greatly affected by
the level of tempering of the microstructure when retained austenite is absent. The
model prediction for steel B at a heating rate of 1 ◦ C s−1 is in good agreement with
the experimental values i.e. 751 ± 24 ◦ C.
C
C
o
o
900 900
Predicted Ac1 at 10oC s-1 /

Predicted Ac1 at 1oC s-1 /
a) b)
850 850
800 800
750 750
700 700
650 650
600 600
550 550
500 500
2 3 4 5 6 7 8 2 3 4 5 6 7 8
(Cr+Mo)/Ni (Cr+Mo)/Ni
Figure 8.13: Steels predicted Ac1 temperatures versus (Cr+Mo)/Ni ratio for heating rates of
1 ◦ C s−1 (a), and 10 ◦ C s−1 (b).
On the other hand, steel A contained 13 % retained austenite in its as-received

microstructure (chapter 4). To determine the extent of the effect of austenite pres-
ence on Ac1 another dilatometric experiment similar to the one reported before has
been conducted.
Figure 8.14 (b) shows an increase of 125 ◦ C in the Ac1 temperature between the
two cycles with 610 ◦ C and 735 ◦ C, respectively. This difference originates from the
fact that when present, austenite precludes the need for nucleation during heating
and thus significantly lowers the temperature of the onset of the transformation.
The model prediction for the same heating rate is 624 ± 23 ◦ C. The latter result
clearly shows that the neural network model underestimates the critical temperature
of the quenched microstructure of high alloy supermartensitic stainless steels. The
presence of retained austenite plays therefore an important role and represents a
source of error in the model.
Nevertheless, the model can be used as a guide to estimate the post-weld heat-
treatment temperature in lean and medium alloyed supermartensitic welds. It should
also be remembered that in the case of multipass welds the situation is further
complicated by reheating effects and those have been neglected in this work.
8.5 Summary 167
1.4 1.4
1.2 1.2
1 First cycle 1
Second cycle
0.8 0.8
dl/lo / %
dl/lo / %
First cycle
0.6 0.6
Second cycle
0.4 0.4
0.2 0.2
0 0
300 400 500 600 700 800 900 300 400 500 600 700 800 900 1000
o o
Temperature / C Temperature / C
Figure 8.14: Dilation curves of the two cycles performed on steels B (a) and A (b). Heating rate
of 1 ◦ C s−1 and maximum temperature of 1000 ◦ C.
8.5 Summary
A new neural network model for predicting Ac1 as a function of chemistry and
heating rate has been developed focussing on martensitic stainless steels. Although
few metallurgical trends were identified, the model represents a significant improve-
ment compared with linear regression equations. The model can be used to estimate
the PWHT temperature of lean and medium supermartensitic stainless steels and
is available freely on the world wide web at http://www.msm.cam.ac.uk/map/steel
/programs/ac1new.html.
Future analysis should consider the level of retained austenite in the starting
microstructure as an input parameter.
Although not explored in this work, the new model can also be employed to
make predictions on low alloy steels since most of the data in the original database
have been obtained on those steels [94].
Chapter 9
Summary and future work
The work presented in this thesis was concerned with the evolution of microstructure
in the high temperature HAZ of three recently designed supermartensitic stainless
steels, with particular attention to austenite and the often neglected δ-ferrite phase
which is potentially detrimental to the properties.
Dilatometry was employed to measure the phase transformation temperatures of

the different alloys and thus helped identify the different regions of the heat-affected
zone. In each steel, the temperature ranges of δ-ferrite formation were short and
thus explained the small size of the dual-phase HAZ in the experimental welds. On
the other hand, MT-DATA successfully predicted the steels to have the capacity to
become fully austenitic and fully ferritic before reaching the melting temperature.
A comparison between the actual and equilibrium predicted temperatures revealed
a systematic underestimation of the Ae4 temperatures. Finally, a comparison of the
predicted temperature ranges of δ-ferrite formation and growth obtained using both
equilibrium and paraequilibrium conditions revealed contradicting trends.
Detailed examination of a series of autogenous bead-on-plates revealed the pres-

ence of δ-ferrite films in dual-phase and coarse-grained HAZ regions and showed the
greater tendency of the most corrosion resistant alloy to retain δ-ferrite. The effect
of cooling rates on δ-ferrite volume fraction was studied on furnace heat-treated
coupons. This revealed that while fast cooling rates favour δ-ferrite retention in
partially ferritised specimens, the opposite applies to specimens cooled from tem-
169
peratures above Ac5 . On the other hand, it was possible to correlate the δ-ferrite
volume fraction in dual-phase HAZ with the employed heat input. The volume per-
centage of δ-ferrite in dual-phase HAZ decreased with heat input and never exceeded
15 %. Furthermore, observations of actual multipass welds showed a non-uniform
distribution of δ-ferrite between the root and the cap of the HAZ. In fact reheating
effects contributed to δ-ferrite dissolution at the root of the HAZ.
The methods available to estimate the tendency of the steel to retain δ-ferrite
have been compared. The Balmforth diagram was found suitable to make compar-
isons between different supermartensitic alloys. The expressions of the Creq and Nieq
indicated that the balance between molybdenum and nickel plays a major role on
δ-ferrite retention. In fact, predictions made using a published neural network model
indicate that it is difficult to balance the ferrite stabilising effect of molybdenum by
only adding the minimum amount of nickel required to obtain a fully austenitic steel
at high temperature. Finally, based on the predicted phase transformation temper-
atures and Rosenthal’s equations, an attempt has been made to predict the extent
of the dual-phase HAZ. Unfortunately, due to the lack of accurate thermodynamic
data the agreement with experiments was poor. Improvement of the thermodynamic
databases is therefore required.
Heat-treated specimens allowed the effects of δ-ferrite on Charpy impact tough-

ness to be investigated in the alloy most prone to ferrite retention. The results
showed that even relatively large amount of δ-ferrite in dual-phase HAZ have little
effect on the ductile to brittle transition temperature. Testing of Gleeble simulated
HAZ confirmed the weak influence of δ-ferrite on the behaviour of the whole HAZ.
In fact, the impact toughness of the coarse-grained HAZ was found to be a more
serious issue because of the large grain size that can be observed in this area. How-
ever, the problem might be more acute for lean supermartensitic stainless steels
since MT-DATA predicted those alloys to possess larger coarse-grained HAZ.
Finally, a neural network model has been proposed to predict the Ac1 tempera-
ture of steels and in particular martensitic stainless steels as a function of chemical
composition and the employed heating rate. Predictions made with this model have
170
been shown to be able to reproduce correctly known metallurgical trends and to be

superior to conventional extrapolation methods. Experimentally, the Ac1 tempera-
tures were seen to be significantly affected by the presence of retained austenite in
the tested microstructure. This represents a potential source of error in the neural
network model and consequently future analysis should take the retained austenite
volume fraction into account in the input variables. Nevertheless, the neural network
model can be used as a guide to estimate the post-weld heat-treatment temperature
of single-pass welds in martensitic stainless steels and also carbon steels.
The effect of the presence of HAZ δ-ferrite on corrosion resistance was not inves-
tigated in this work. However, work is being performed on the topic at TWI. So far
it has not been clearly demonstrated that δ-ferrite presence is responsible for crack
initiations.
APPENDIX ONE
MAP STEEL AC1TEMP
This appendix presents the model described in chapter 8 and associated documenta-
tion following the MAP format, http://www.msm.cam.ac.uk/map/mapmain.html.
1 Provenance of Source Code

Dominique Carrouge
Phase Transformations and Complex Properties Group,
Department of Materials Science and Metallurgy,
University of Cambridge,
Cambridge, CB2 3QZ U.K.
The neural network program was produced by:

David MacKay,
Cavendish Laboratory,
University of Cambridge,
Madingley Road,
Cambridge, CB3 0HE, U.K.
Added to MAP: September 2002.
2 Purpose
A program for the prediction of the Ac1 temperatures of steels, including martensitic
stainless steels, as a function of chemical composition and heating rate.
3 Specification
Language: C
Product Form: Source Code (Unix version only)
Operating System: Tested on Solaris, Linux and PC.
Appendix One 172
4 Description
MAP STEEL AC1 TEMP contains the programs which enable the user to estimate
the Ac1 temperature of steels, including martensitic stainless steels, as a function of
chemical composition and heating rate. It makes use of a neural network program
called generate44, which was developed by David MacKay and is part of the bigback5
program. A total of eighty three experimental data relevant to 9-13 wt% Cr steels
were added to the original database from Gavard et al. work [1] and the updated
database “database ac1 martensitic.txt” is provided. Following a Gaussian process
analysis by Bailer-Jones et al. [2], the elements considered to affect the Ac1 temper-
ature in this neural network model are : C, Si, Mn, Cu, Ni, Cr, Mo, Nb, V, W and
Co. The model presented here combines the output of two different models which
had five hidden units and different initial weights. The programs automatically cal-
culate the result of each model and average those results to produce a prediction of
the Ac1 temperature and an error estimate.
A Linux/Unix and a PC version of the program are provided. To make predictions

with the PC version alter the test.dat file and double click on model.exe. The results
are in the file Model RESULT.dat. In the Linux/Unix version, the source code
is accompanied by a program to install the program, which should run on most
versions of UNIX. Once uncompressed, and once the ‘install’ program runs, the
directory contains :
MINMAX
A text file containing the minimum and maximum limits of each input and output
variable. This file is used to normalise and unnormalise the input and output data.
test.dat
The input file that needs to be edited to make predictions. The variables should be
entered in line, separated by a single space and in the following order : C, Si, Mn,
Cu, Ni, Cr, Mo, Nb, V, W, Co and Heating rate.
Appendix One 173
model.gen
This is a UNIX shell file containing the commands required to run the model. It
can be executed by typing ‘sh model.gen’ at the command prompt. This shell file
normalises the input data, executes the neural network program, unnormalises the
results and combine them to produce the final committee result.
.normalise
Hidden executable file, to normalise the input data.
.generate44
Hidden executable file, for the neural network program. It reads the normalised
input file and also uses a configuration file spec.t generated by .generate spec and
the weight files, located in the subdirectory c.
.generate spec
Hidden executable file, generates the configuration file read by .generate44.
.gencom
Hidden executable file, combines the output of the different models in a committee
result.
.treatout
Unnormalise the results.
SUBDIRECTORY c
w*f
The weight files of the different models.
*.lu
Files containing information for calculating the size of the error bars for the different
models.
Appendix One 174
c*
Files containing information about the perceived significance value [3] for each
model.
R*
Files containing information about the noise, test error and log predictive error [3]
for each model.
SUBDIRECTORY d
outran.x
A normalised output file which was created during the building of the model. It is
accessed by .generate44.
SUBDIRECTORY outprdt
out1, out2, etc.

The normalised output files for each model.
com.dat
The normalised output file containing the committee results. It is generated by
.gencom.
5 References
1. L. Gavard, H. K. D. H. Bhadeshia, D. J. C. MacKay and S. Suzuki, Bayesian
Neural Network Model for Austenite Formation in Steels, Materials Science and
Technology 12:453-463, 1996.
2. C. A. L. Bailer-Jones, H. K. D. H. Bhadeshia and D. J. C. MacKay, Gaussian

Process Modelling of Austenite Formation in Steel, Materials Science and Technology
15:287-294, 1999.
Appendix One 175
3. D. J. C. MacKay, Probable Networks and Plausible Predictions - A Review

of Practical Bayesian Methods For Supervised Neural Networks, 1995, Web site :
http://wol.ra.phy.cam.ac.uk/mackay/BayesNets.html.
6 Input parameters
As mentioned before, the input variables are C, Si, Mn, Cu, Ni, Cr, Mo, Nb, V,
W, Co and Heating rate. The maximum and minimum values for each variable are
given in the file MINMAX.
7 Output parameters
The output is written in the file result in the Linux/Unix version and in
Model RESULT.dat for the PC version. This file contains the predicted Ac1 temper-
atures (in ◦ C), the error bar value, the predicted value - the error and the predicted
value + the error.
Program data
See sample file test.dat
Program results
See sample file result.txt
Keywords
Neural networks, Ac1 temperature, heating rate, martensitic stainless steels, carbon
steels.
Bibliography
[1] A. G. Haynes. Some Factors Governing the Metallurgy and Weldability of

13 % Cr and Newer Cr-Ni Martensitic Stainless Steels. In Supermartensitic
Stainless Steels ’99’, pages 25–32, Belgium, 1999.
[2] P. E. Biagiotti and J. S. Reichman. Justifying the Use of 13Cr Steels For Cor-
rosive CO2 Operations. In CORROSION 1995, paper 81, NACE International,
1995.
[3] J. Charles. How to Specify CRA’s Cost Effectively. In Stainless Steel World
2002, pages 13–27, Houston USA, 2002.
[4] D. William and Jr. Callister. Materials Science and Engineering an Introduc-
tion. John Wiley and Sons, Inc, New York, USA, 1999.
[5] L. M. Smith and M. Celant. Martensitic Stainless Steel Flowlines, Do They

Pay. In Supermartensitic Stainless Steels ’99’, pages 66–73, Belgium, 1999.
[6] J. R. Davis. Selection of Wrought Martensitic Stainless Steels. ASM, Metals

Handbook, 6:432–442, 1971.
[7] M. Ueda, H. Amaya, K. Kondo, K. Ogawa, and T. Mori. Corrosion Resistance

of Weldable Super 13Cr Stainless Steel in H2 S Containing CO2 Environments.
In CORROSION 1996, paper 58, NACE International, 1996.
[8] E. Deleu, A. Dhooge, and Dufrane J. J. Weldability and Hot Deformability of

Different Supermartensitic Stainless Steel Grades by Weld Simulation testing.
In Supermartensitic Stainless Steels ’99’, pages 232–240, Belgium, 1999.
Bibliography 177
[9] C. Dillon. Compositional Effects in Corrosion of Stainless Steels in Chemical

Environments. Corrosion, 16:433t–440t, 1960.
[10] E. Folkhard. Welding Metallurgy of Stainless Steels. Springer-Verlag, Wien,

New York, 1988.
[11] R. Castro and J. J. De cadeney. Welding Metallurgy of Stainless and Heat-

Resisting Steels. Cambridge University Press, Cambridge, 1974.
[12] K. Bungardt, E. Kunze, and E. Horn. Untersuchungenüber den Aufbau des

Systems Eisen-Chrom-Kohlenstoff. Arch. eisenhüttenwes, 29:193–203, 1958.
[13] M. Kimura, Y. Miyata, Y. Yasuyoshi, T. Takaaki, and N. Yoshifumi. Corrosion

Resistance of High Strength Modified 13Cr Steel. In CORROSION 1997, paper
22, NACE International, 1997.
[14] K. kondo, M. Ueda, K. Ogawa, H. Amaya, H. Hirata, and H. Takabe. Alloy

Design of Super 13 Cr Martensitic Stainless Steel (Development of Super 13
Cr Martensitic Stainless Steel for Line Pipe-1). In Supermartensitic Stainless
Steels ’99’, pages 11–18, Belgium, 1999.
[15] H. Asahi, T. Muraki, H. Inoue, and H. Tamehiro. High Chromium Martensitic

Stainless Linepipes. Pipeline Technology ASME 1996, V:223–230, 1996.
[16] H. Asahi, T. Hara, A. Kawakami, and A. Takahashi. Development of Sour

Resistant Modified 13Cr OCTG. In CORROSION 1995, paper 79, NACE
International, 1995.
[17] T. G. Gooch. Welding Martensitic Stainless Steels. Welding Institute Research

Bulletin, 18:343–349, 1977.
[18] Sumitomo Metal Industries Ltd. European Patent Specification : Martensitic

Stainless Steels for Use in Oil Wells. Bulletin 1997/30, 1997. Patent number
EP 0 565 117 B1.
[19] Y. Miyata, M. Kimura, T. Koseki, T. Toyooka, and F. Murase. Martensitic

Stainless Steel Seamless Linepipe With Superior Weldability and CO2 Corro-
sion resistance. In CORROSION 1997, paper 19, NACE International, 1997.
Bibliography 178
[20] A. Miyasaka and H. Ogawa. Influence of Metallurgical Factors on Corrosion

Behaviors of Modified 13 %Cr Martensitic Stainless Steels. In CORROSION
1990, paper 67, NACE International, 1990.
[21] J. Pleva and S. Nordin. Properties of Different MMA Welds on Modified Type
329 Ferritic-Austenitic Stainless Steel. In Duplex Stainless Steels ’82’, pages
603–629, St. Louis, 1982.
[22] Ø. Grong. Metallurgical Modelling of Welding. The Institute of Materials,

London, 1994.
[23] Standard MR0175. Sulphide Stress Cracking Resistant Metallic Materials for
Oilfield Equipment. Nace International, 2001.
[24] P. P. Koistinen and R. E. Marburger. A General Equation Prescribing The Ex-

tent of The Austenite-Martensite Transformation in Pure Iron-Carbon Alloys
and Plain Carbon Steels. Acta Metallurgica, 7:59–60, 1959.
[25] Honeycombe, R. W. K. and Bhadeshia, H. K. D. H. Steels Microstructures

and Properties. Edward Arnold, London, 1995.
[26] T. Maki, K. Tsuzaki, and I. Tamura. The Morphology of Microstructure

Composed of Lath Martensites in Steels. ISIJ, 20:207–214, 1980.
[27] K. Takano, M. Sakakibara, T. Matsui, and S. Takaki. Effects of Alloying El-

ements and Microstructure on Corrosion Resistance and Hardness of Marten-
sitic Stainless Steels. Tetsu to Hagane, 86:123–130, 2000.
[28] J. J. Dufrane. Metallurgical Basis for the Development of Weldable Marten-

sitic Stainless Steels. In Supermartensitic Stainless Steels ’99’, pages 19–24,
Belgium, 1999.
[29] J. A. Ribenolt and W. J. Harris. Effect of Alloying Elements on Notch

Toughness of Pearlitic Steels. Transactions of the american society for metals,
43:1175–1201, 1951.
[30] P. Woollin. The Weldability of Improved Martensitic Stainless Steels. Tech-

nical Report 5673/9A/98, Welding Institute Internal Report, 1998.
Bibliography 179
[31] J. K. Solberg, E. V. Ladanova, G. Rorvik, and P. E. Kvaale. Post Weld Heat

Treatment Response of Coarse Grained Heat Affected Zone in a Supermarten-
sitic Stainless Steel. In Supermartensitic Stainless Steels ’99’, pages 56–63,
Belgium, 1999.
[32] K. Nose and H. Asahi. Effect of Microstructure on Corrosion Resistance of

a Martensitic Stainless Linepipe. In CORROSION 2000, paper 145, NACE
International, 2000.
[33] M. Kimura, Y. Miyata, and T Toyooka. Effect of Retained austenite on Cor-

rosion Performance for Modified 13% Cr Steel Pipe. In CORROSION 2000,
paper 137, NACE International, 2000.
[34] C. Leymonie, M. C. Ottmann, R. Risacher, and G. Thauvin. Contribution a

l’etude des Transformations Structurales des Aciers a 13%Cr - 4%Ni. Revue
de Metallurgie, 12:815–826, 1979.
[35] L. Dong-Seok, L. Yong-Deuk, J. Joong-Hwan, and C. Chong-Sool. Amount

of Retained Austenite at Room Temperature After Reverse Transformation
of Martensite to Austenite in an Fe-13%Cr-7%Ni-3%Si Martensitic Stainless
Steel. Scripta materialia, 45:767–772, 2001.
[36] I. Varga, E. Kuzmann, and A. Vertes. Kinetics of α-γ Phase Transition of Fe-
12Cr-4Ni Alloy Aged Between 500-650 ◦ C. Hyperfine Interactions, 112:169–
173, 1998.
[37] I. Varga, E. Kuzmann, M. Kaldor, and A. Vertes. Mössbauer Study of the

Kientic Behavior of Anomalous Austenite Formation in Fe-12Cr-4Ni Steel. J.
Radioanal. Nucl. Chem., 165:115–126, 1992.
[38] T. G. Gooch, P. Woollin, and A. G. Haynes. Welding Metallurgy of Low

Carbon 13 %Cr Martensitic Steels. In Supermartensitic Stainless Steels ’99’,
pages 188–195, Belgium, 1999.
[39] A. Kulmburg, F. Korntheuer, O. Gründler, and K. Hutterer. Umwand-

lungsund Ausscheidungsverhalten von Weichmartensitischen und Hochfesten
Korrosionsbeständigen Stählen. Berg. u. hüttenm. Mh., 126:104–108, 1981.
Bibliography 180
[40] C. Leymonie, M. C. Lecocq, and M. C. Ottmann. Relations entre la Structure

et les Proprietes Mecaniques de Deux Aciers a 16%Cr - 4%Ni. Traitement
thermiques, 165:57–62, 1982.
[41] H. J Niederau. Entwicklungsstand Nichtrostender Weichmartensitischer

Chrom-Nickel-Stähle unter besonderer Berücksichtigung des Stahles X 5
CRNi13 4. Stahl u. Eisen, 98:385–392, 1978.
[42] P. D. Bilmes, M. Solari, and C. L. Llorente. Characteristics and Effects of

Austenite Resulting From Tempering of 13Cr-Ni-Mo Martensitic Steel Weld
Metals. Materials Characterisation, 46:285–296, 2001.
[43] M. C. Balmforth and J. C. Lippold. A Preliminary Ferritic-Martensitic Stain-

less Steels Constitution Diagram. Welding Journal, 77:1s–7s, 1998.
[44] J. C. M. Farrar and A. W. Marshall. Super Martensitic Stainless Steels -

Overview and Weldability. IIW-IX-H-423-98, pages 1–16, 1998.
[45] S. Hashizume. A new 15 %Cr Martensitic Stainless Steel Developed for OCTG.
In CORROSION 1991, paper 28, NACE International, 1991.
[46] Y. Ishizawa, T. Shimada, and M. Tanimura. Effect of Microstructure on the

Sulphide-Stress Corrosion Cracking Resistance of AISI 410 and 420 Steels. In
NACE International, paper 124, Houston, Texas, 1982.
[47] T. Hara and H. Asahi. Effect of δ-Ferrite on Sulphide Stress Cracking in a

Low Carbon 13 mass % Chromium Steel. ISIJ International, 40:1134–1141,
2000.
[48] TWI web site. www.twi.co.uk.
[49] K. Easterling. Introduction to the Physical Metallurgy of Welding. Butterworth

& Co Ltd, 1983.
[50] D. P. Smith, D. J. Carney, L. W. Eastwood, and C. E. Sims. Gases in Metals.

American Society for Metals, Cleveland, Ohio USA, 1953.
Bibliography 181
[51] A. W. Marshall and J. C. M. Farrar. Welding of Ferritic and Martensitic

11-14% Cr Steels . Welding in the world, 45:32–55, 2001.
[52] J. F. Lancaster. Metallurgy of Welding, fourth edition. Allen and Unwin Ltd,
40 Museum street, London WC1A 1LU, UK, 1987.
[53] J. Enerhaug, Ø. Grong, and U. M Steinsmo. Factors Affecting Initiation of

Pitting Corrosion in Super Martensitic Stainless Steel Weldments. Science
and Technology of welding and Joining, 6:330–338, 2001.
[54] L. Coudreuse, M. Verneau, and J. J. Dufrane. Sulphide Stress Cracking Re-

sistance of Weldable Super-Martensitic Stainless Steels. In Supermartensitic
[55] G. Rørvik, P. E. Kvaale, and O. M. Akselsen. Sources and Levels of Hydrogen

in TIG Welding of 13% Cr Martensitic Stainless steels. In Supermartensitic
[56] P. Woollin and D. Carrouge. Heat Affected Zone Microstructures in Super-

martensitic Stainless Steels. In Supermartensitic Stainless Steels 2002, Bel-
gium, 2002. Paper 027.
[57] T. G. Gooch and B. J. Ginn. Heat Affected Zone Toughness of MMA Welded
12 %Cr Martensitic-ferritic Steel. Technical Report 373, Welding Institute
Members Report, 1988.
[58] C. Grobler. Weldability Studies on 12 % and 14 % Chromium Steels. PhD

thesis, University of Pretoria, 1987.
[59] J. J. J. Zaayman and G. T. Van Rooyen. The Toughness of the Heat Affected
Zone of Welds in 11.5 per cent Chromium Steels. In Proceedings of the 1st
International Chromium Steels and Alloys Congress, Volume 2, SAIMM, pages
137–142, Cape town, Johannesburg, 1992.
[60] L. Karlsson, S. Rigdal, A. Dhooge, E. Deleu, M. Goldschmitz, and J. Van den

Broek. Degassing Response of Supermartensitic Weld Metals. Technical Re-
port IIW-IX-H-514-2001, International Institute of Welding, 2001.
Bibliography 182
[61] N. Petch. Fracture. John Wiley & son, New York, USA, 1959.
[62] J. F. Knott. Fundamentals of Fracture Mechanics. Butterworths, London,

UK, 1988.
[63] R. Fenn and M.F. Jordan. Microstructure of Weld Heat-Affected Zone of

12Cr-Mo Steel. Metals Technology, 9:327–337, 1982.
[64] H. Hoffmeister and R. Mundt. Untersuchungen zum Einfluß des Kohlenstoffs

und des Stickstoffs auf die δ-γ Umwandlung ferritisch-austenitischer Chrom-
Nickel- Stäle. Arch. eisenhüttenwes, 52:159–164, 1981.
[65] J. Singh, G. R. Purdy, and G. C. Weatherly. Microstructural and Microchem-

ical Aspect of The Solid State Decomposition of Delta-ferrite in Austenitic
Stainless Steels. Metallurgical Transactions, 16A:1363–1369, 1985.
[66] H. Hemmer. A Process Model for Electron Beam Welding of Duplex stainless
steels. PhD thesis, Norges teknisk-naturvitenskapelige universitet, 1999.
[67] W. T. Delong. Ferrite in Austenitic Stainless Steels Weld Metal. Welding

Journal, 53:273s–286s, 1974.
[68] C. G. Long and W. T. Delong. The Ferrite Content of Austenitic Stainless

Steel Weld Metal. Welding Journal, 52:281s–297s, 1973.
[69] D. J. Kotecki. Welding Parameter Effects on Open-Arc Stainless Steel Weld

Metal Ferrite. Welding Journal, 57:109s–117s, 1978.
[70] D. J. Kotecki. Extension of the WRC Ferrite Number System. Welding Jour-
nal, 61:352s–361s, 1982.
[71] R. H. Kaltenhauser. Improving the engineering properties of ferritic stainless

steels. Metals Engineering Quaterly, 11:41–47, 1971.
[72] J. Hewitt. High-Chromium Controlled-Hardenability Steels. In Proceedings

of the 1st International Chromium Steels and Alloys Congress, Cape Town.
Volume 2. Johannesburg, SAIMM, pages 71–88, 1992.
Bibliography 183
[73] T. G. Gooch. Corrosion Behaviour of Welded Stainless Steels. Welding Jour-

nal, 75:135s–154s, 1996.
[74] A. L. Schaeffler. Constitution diagram for stainless steel weld metal. Metal
Progress, 56:680–680B, 1949.
[75] M. C. Balmforth and J. C. Lippold. A New Ferritic-Martensitic Stainless

Steels Constitution Diagram. Welding Journal, 79:339s–345s, 2000.
[76] T. A. Siewert, C. N. McCowan, and D. L. Olson. Ferrite Number Prediction to

100 FN in Stainless Steels Weld Metal. Welding Journal, 67:289s–298s, 1988.
[77] D. J. Kotecki and T. A. Siewert. WRC-1992 Constitution Diagram for Stain-

less Steels Weld Metals : a Modification of the WRC-1988 Diagram. Welding
Journal, 71:171s–178s, 1992.
[78] J. M. Vitek, Y. S Iskander, and E. M. Oblow. Improved Ferrite Number

Prediction in Stainless Steel Arc Welds Using Artificial Neural Networks -
Part 1 : Neural Network Development. Welding Journal, 79:33s–40s, 2000.
[79] J. M. Vitek, Y. S. Iskander, and E. M. Oblow. Improved Ferrite Number

Prediction in Stainless Steel Arc Welds Using Artificial Neural Networks -
Part 2 : Neural Network Results. Welding Journal, 79:41s–50s, 2000.
[80] M. Vasudevan, M. Murugananth, and A. K. Bhaduri. Application of Bayesian

Neural Network For Modelling and Prediction of Ferrite Number in Austenitic
Stainless Steel Welds. Mathematical Modelling of Weld Phenomena, 6:1079–
1099, 2002.
[81] P. Woollin. Metallurgical Examination of Martensitic Stainless Steel Weld-

ments. Technical Report 620886/1/95, Welding Institute Internal Report,
1995.
[82] E. Beraha and B. Shpigler. Color Metallography. American Society for Metals,
1977.
[83] B. D. Cullity. Elements of X-ray Diffraction. Addison-Wesley, 1978.

Bibliography 184
[84] BS EN 10045-1. Charpy Impact Test on Metallic Materials. Part 1: Test

Method. British Standard, 1990.
[85] R.D. Campbell and D.W. Walsh. Weldability Testing. ASM, Metals Handbook,
6:603–613, 1971.
[86] G. L. Leone and H. W. Kerr. The Ferrite to Austenite Transformation in

Stainless Steels. Weld. J. res. suppl., 61:13s–22s, 1982.
[87] J. C. Lippold and W. F. Savage. Solidification of Austenitic Stainless Steels

Weldments Part I- a Proposed Mechanism. Weld. J. res. suppl., 58:362s–374s,
1979.
[88] G. Blanc and R. Tricot. Solidification, Segregation et Homogeneisation des

Aciers Inoxydables Austenitique Contenant de la Ferrite Delta. Mem. Sci.
Rev. Met., 58:735–753, 1979.
[89] NKK Weldable 13 %Cr Seamless Line Pipe Material. Technical report, NKK
Japan, 1999.
[90] R. D. Knutsen and R. Hutchings. Occurence of Non-metallic Inclusions in

3CR12 Steel and Their Effect on Impact Toughness . Materials Science and
Technology, 4:127–135, 1988.
[91] M. M. McDonald and Ludwigson D. C. Fractographic Examination in the

Scanning Electon Microscope as a Tool in Evaluating Through-Thickness Ten-
sion Test Results. Journal of Testing and Evaluation, 11:165–173, 1983.
[92] D. P. Fairchild, D. G. Howden, and W. A. T. Clark. The Mechanism of

Brittle Fracture in a Microalloyed Steel : Part I. Inclusion-Induced Cleavage.
Metallurgical and Materials Transactions A, 31A:641–652, 2000.
[93] D. P. Fairchild, D. G. Howden, and W. A. T. Clark. The Mechanism of Brittle

Fracture in a Microalloyed Steel : Part II. Mechanistic Modelling. Metallurgical
and Materials Transactions A, 31A:653–667, 2000.
Bibliography 185
[94] L. Gavard, H. K. D. H. Bhadeshia, D. J. C. MacKay, and S. Suzuki. Bayesian

Neural Network Model for Austenite Formation in Steels. Materials Science
and Technology, 12:453–463, 1996.
[95] D. J. C. MacKay. Probable Networks and Plausible Predictions - A

Review of Practical Bayesian Methods For Supervised Neural Networks.
http://wol.ra.phy.cam.ac.uk/mackay/BayesNets.html, 1995.
[96] H. K. D. H. Bhadeshia. Neural Network in Materials Science. ISIJ Interna-

tional, 39:966–979, 1999.
[97] C. A. L. Bailer-Jones, H. K. D. H. Bhadeshia, and D. J. C. MacKay. Gaussian

Process Modelling of Austenite Formation in Steel. Materials Science and
Technology, 15:287–294, 1999.
[98] C. Downs and H. K. D. H. Bhadeshia. Unpublished work. 2001.
[99] A. Ball, Y. Chauhan, and G. B. Schaffer. Microstructure Phase Equilibria and

Transformations in Corrosion Resistant Dual Phase Steel Designated 3CR12.
Materials Science and Technology, 3:189–196, 1987.
[100] A. Alamo, J. C. Brachet, A. Castaing, F. Lepointtevin, and F. Barcelo. Phys-

ical Metallurgy of FeCrWTaV Low Activation Martensitic Steels. Effect of
Chemical Composition. Journal of Nuclear Metallurgy, 258-263:1228–1235,
1998.
[101] D. J. C. MacKay. Bayesian Non-Linear Modelling Neural Networks.

http://wol.ra.phy.cam.ac.uk/mackay/BayesNets.html, 1996.
[102] P. Woollin. Optimising The Properties of Weldments in Weldable Martensitic

and Supermartensitic Stainless Steel. Technical Report 12097/5/01, Welding
Institute Internal Report, 2001.

Complete Final

Uploaded by

Copyright:

Available Formats

Complete Final

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Complete Final

Uploaded by

Copyright:

Available Formats

Preface

This dissertation is submitted for the Degree of Doctor of Philosophy in Natural

Part of this work has been presented in the following publications :

D. Carrouge, H. K. D. H. Bhadeshia and P. Woollin, Microstructural change

P. Woollin and D. Carrouge, Heat-aﬀected zone microstructures in supermarten-

1 Introduction to supermartensitic stainless steels 1

2 Welding and heat-aﬀected zone microstructures and properties 24

4 Characterisation and predictions 54

5 Microstructure in high temperature heat-aﬀected zone 80

6 Predicting δ-ferrite retention and dual-phase HAZ width 112

7 Eﬀect of HAZ δ-ferrite on Charpy impact toughness 127

8 Neural network model for estimating the Ac1 temperature 147

9 Summary and future work 168

APPENDIX ONE 171

1.1.a Materials selection

1. Adequate corrosion resistance to a range of species found in water that may

3. Good low temperature impact toughness in case of emergency closure of the

4. The capacity to be easily and reliably joined by welding (good weldability)

A comparison of the costs of typical materials used in the manufacture of pipelines

Amongst the diﬀerent families of stainless steels, ferritic/austenitic (duplex)

Carbon steel 0.26 0 0 0 1000

Compared with their ancestors, supermartensitic stainless steels have enhanced

403 0.15 max 1.00 0.5 11.5 - 13.0 0 0

1.1.b Constituents of supermartensitic stainless steels

Solid mark : SSC

Corrosion rate / mm year −1

Alloy Y.S. (MPa) T.S. (MPa)

11Cr 2.5Ni 598 753

22Cr 5Ni 3Mo 480 680

• When the chromium content is below approximately 12 wt% it is possible to

• A chromium content greater than approximately 14 wt% gives a completely

microstructure consisting partly of δ-ferrite which remains unchanged following

Figure 1.3: Schematic view of the body cen-

iii Carbon and nitrogen

v Phase balance in 0.01 C wt% Fe-Cr-Ni-Mo system

0 0.2 0.4 0.6 0.8 1

MS (◦ C) = 540 − 497C − 6.3Mn − 36.3Ni − 10.8Cr − 46.6Mo wt% (1.1)

Austenite stabilising elements

• Manganese, is added in concentrations up to 2 wt % to substitute some of

• Copper, is believed to improve corrosion resistance in CO2 or CO2 /H2 S en-

Ferrite stabilising elements

• Silicon, is a strong ferrite former, essential as a deoxidiser during reﬁning of

• Titanium. By forming very stable carbo-nitrides, Ti(C,N), titanium has an

1.2 Types of corrosion

1.2.a Uniform attack

P RE = Cr + 3.3Mo + 16N wt% (1.2)

Element Composition range (wt %)

1.2.c Stress-corrosion cracking

1.2.d Hydrogen embrittlement

similar to stress-corrosion in that a normally ductile metal experiences brittle frac-

1.2.e Sulﬁde stress cracking

Figure 1.9: Mechanism of hydrogen absorption in sulﬁde stress-corrosion cracking [22].

The presence of sulfur from H2 S accelerates embrittlement by retarding the for-

1.3 Phases present and mechanical properties

Martensitic transformation in supermartensitic steels is athermal, meaning that

Vm = exp(−0.011 × (Ms − Tq )) (1.3)

where Vm is the fraction of martensite and Tq is the quenching temperature below

The displacement associated with martensite formation is an invariant-plane

Figure 1.10: Schematic illustration of the microstructural fea-

i Martensite hardness and toughness