Analysis of The Effect of Alloying Elements On The Martensite-Start Temperature of
Analysis of The Effect of Alloying Elements On The Martensite-Start Temperature of
Analysis of The Effect of Alloying Elements On The Martensite-Start Temperature of
Department of Physical Metallurgy, Centro Nacional de Investigaciones Metalrgicas (CENIM) Consejo Superior de Investigaciones Cientficas (CSIC), Avda. Gregorio del Amo, 8. 28040 Madrid, Spain. www.cenim.csic.es.
Corresponding author. Tel.: 00-34-915538900; fax: 00-34-915347425 E-mail addresses: [email protected] (C.Capdevila), [email protected] (F.G.
Abstract
Making the transformation from austenite to martensite difficult is called stabilisation of austenite, a phenomenon that occurs in many cases. The straightforward method to analyse the influence of a specific factor on the stabilisation of austenite is through its influence on the martensite start (Ms) temperature. This work outlines the use of an artificial neural network to model the Ms temperature of engineering steels from their chemical composition and austenite grain size. The results are focussed on analysing the role in the stabilisation of austenite of alloying elements in steels including less common elements such as V and Nb, as well as the austenite grain size. Moreover, a physical interpretation of the results is presented.
1. Introduction
Microalloyed steels in the as-forged condition are commonly used in the automotive industry. For heavy-duty applications such as that for diesel engine crankshafts, a surface induction hardening heat treatment is carried out in critical regions of the components to enhance their service performance through adding strength and fatigue resistance. The goal of the induction hardening heat treatment is to form a fully martensitic structure in the outer surface of the component to locally increase the hardness and tensile strength. Thus, factors affecting martensitic transformation are of vital importance in the design of industrial processes of these engineering steels. Making the transformation from austenite to martensite difficult is called stabilisation of austenite, which is due to a change in chemical composition, heat treatment or plastic deformation.
1
therefore, the influence of the chemical composition on the martensite start (Ms) temperature has been extensively reported in the literature for low alloy steels and several empirical equations have been proposed. 2-7 However, these empirical equations are not sufficiently general and are known to provide inaccurate answers for microalloyed steels, or steels whose compositional range are out of bounds from those used to formulate the equations. On the other hand, Olson and Cohen 8 developed a model for heterogeneous martensitic nucleation that obviates the need for pre-existing embryos with martensitic structure, but requires a suitable nucleating defect in austenite. The initial defect might be a group of dislocations in an austenite-austenite interface quenching form austenitisation temperature.
10 9
lattice imperfections may also act as nucleation sites and contribute to make the
austenite phase unstable. On the contrary, they can also contribute to the stabilisation of the austenite phase by hindering the growth of the transformation product.
10
Which of
these various contributions predominates depends on the chemical composition and the nature of the imperfections. It is followed that to find out how the austenite grain size and alloying elements, including those used as microalloying elements (V and Nb), can affect the Ms temperature is an important issue to be investigated. The aim of this work is to develop an artificial neural network model to predict the Ms temperature of steels and interpret the influence of the chemical composition and the austenite grain size. Neural networks are useful whenever the intricacy of the problem is overwhelming from a fundamental perspective and where simplification is unacceptable. They represent a powerful method of non-linear regression modelling.
The definition of the Ms temperature in any model ideally requires a complete description of the chemical composition and the austenite grain size. A search of the literature
9-15
composition, prior austenite grain size (PAGS), and Ms values were reported. Table l shows the list of 15 input variables used for the Ms temperature analysis. It was possible to find 270 cases where all of these variables were reported with the exception of nitrogen concentrations. It would have been unreasonable to set this latter concentration to zero when its value is not reported. Steels inevitably contain this impurity element in practice. For cases where the N value was missing its concentration was set to the mean value calculated for the 320 cases of the database. For the other
elements such as Mn, Ni, etc, the concentrations can truly be set to zero when they are not reported. This is because they would not be deliberate added or would have concentrations close to the limits of chemical analysis with the techniques generally used 16.
The aim is to be able to estimate the Ms temperature as a function of the variables listed in Table 1. The analysis was carried out using variables normalised between +0.5 and 0.5; this normalisation is not necessary for the analysis, but allows a convenient comparison of the influence of individual input variables on an output. The normalisation procedure is expressed quantitatively as
xN = x x min 0 .5 x max x min
(1)
where xN is the normalised value of x which has maximum and minimum values given by xmax and xmin respectively. The network consisted of 15 input nodes (Table 1), a number of hidden nodes, and an output node representing the Ms temperature (Fig. 1). The network was trained using a randomly chosen of 170 examples from a total of 320 available; the remaining 150 examples were used as new experiments to test the trained network. Linear functions of the inputs xj are operated by a hyperbolic tangent transfer function
( hi = tanh wij1) x j + i(1) j
(4)
so that each input contributes to every hidden unit. The bias is designated i and is analogous to the constant that appears in linear regression. The strength of the transfer
function is in each case determined by the weight wij. The transfer to the output y is linear y = wi( 2 ) hi + ( 2)
i
(3)
This specification of the network structure, together with the set of weights, is a complete description of the formula relating the inputs to the output. The weights were determined by training the network and the details are described by MacKay
17-18
. The
training involves a minimisation of the regularised sum of squared errors. The term v used below was the framework estimation of the noise level of the data. The complexity of the model was controlled by the number of hidden units (Fig. 2). Figure 2 shows that the inferred noise level decreases monotonically as the number of hidden units increases. However, the complexity of the model also increases with the number of hidden units. A high degree of complexity may not be justified, and in an extreme case, the model may in a meaningless way attempt to fit the noise in the experimental data. MacKay 19-20 has made a detailed study of this problem and defined a quantity (the evidence) which comments on the probability of a model. In circumstances where two models give similar results for the known data, the more probable model would be predicted to be that which is simpler; this simple model would have a higher value of evidence. The evidence framework was used to control v. The number of hidden units was set by examining performance on test data. A combination of Bayesian and pragmatic statistical techniques were therefore used to control the complexity of the model. 19-21 Figure 3 shows that a large number of hidden units did not give significantly lower values of v; indeed, the test set error has a minimum at three hidden units. Therefore, three hidden units were found to give a reasonable level of complexity to represent the variations of Ms temperature as a function of the input variables of Table 1. The levels
of agreement for the training and test data are shown in Fig. 4(a) and 4(b); good predictions occur in both instances.
The main advantage of the neural network model as compared with other empirical models is the ability of analysing separately the influence on the stabilisation of austenite of each one of the alloying elements. In this sense, the role of microalloying elements such as V and Nb on Ms temperature has been analysed in this section. Figure 5 shows the influence of V and Nb on Ms temperature for three different grades of carbon and considering a constant PAGS of 20 m. It is clear from this figure that the higher microalloying content, the higher Ms temperature. This effect is more pronounced as carbon concentration increases. It is possible to get a physical understanding of these results. According to their chemical properties, V and Nb can be classified as very strong carbide formers. This behaviour may be attributed to the influence of alloying elements on the activity of carbon in the solid solution. Keeping this in mind, we can expect that interactions between carbon and carbide former elements tend to weaken the role of carbon rising thus Ms. Likewise, these interactions are stronger as carbon content increases. This is consistent with the following thermodynamic considerations. In the thermodynamic approach presented by Bhadeshia
22-23
, martensitic transformation is
said to be triggered when the chemical driving force (G) achieves a critical value at
the Ms temperature ( GC ' ). Following Olson and Cohen,
24
proposed that the total energy describing the heterogeneous semicoherent nucleation process of martensite is given by the sum of a defect dislocation energy and a fault energy. At sufficient driving force (obtained by cooling or an applied stress), the fault energy becomes negative and gives rise to a barrierless condition where a nucleus can grow spontaneously, at a rate controlled by the interfacial mobility. The critical condition for semicoherent nucleation is then given by a balance between the negative fault energy and the interfacial frictional work. Therefore, Gosh and Olson proposed
that GC ' is the addition of two terms. The former includes the fault energy, and the
latter is the interfacial frictional work between the austenite matrix and martensite nucleus which is composition dependent. The critical value in J mol-1 of the driving force needed to trigger martensitic transformation is:
0 0 0.5 0 0.5 0. G C ' = 1010 + 4009c C.5 + 1879c Si.5 + 1980c Mn + 172c C.5 + 1418c Mo + 1868c Cr5 0 0. 0 0.5 0 0. + 1618cV .5 + 752c Cu5 + 714cW.5 + 1653c Nb + 3097c N.5 352c Co5
(7)
where c0.5 are the square root of the alloying elements concentration in mole fraction. The coefficients were obtained by Ghosh and Olson by establishing the c0.5 dependence and fitting over a wide range of compositions: the maximum concentrations were approximately 2 wt.-% for carbon and nitrogen, 0.9 wt.-% vanadium and about 2-28 wt.-% for all the other alloying elements. 25 The thermodynamic calculations involved here have been performed using the commercial software package, MTDATA.
26
27
was used to
express the Gibbs free energies of ferrite and austenite phases. The first sublattice is occupied by substitutional atoms and the second is occupied by interstitial atoms and vacancies. The Gibbs free energies of austenite, G, and ferrite of the same composition,
G, were calculated separately by allowing only one phase to exist in the system. Then,
the molar Gibbs free energy differences, G = G G, at different temperatures were obtained. The Gibbs free energies of both phases include unitary terms of free energies, mixing entropies, excess free energies describing the deviation from the regular solution model, and magnetic contributions. However, to calculate G also requires an estimation of the Zener ordering energy,
28
can in some circumstances order on one of available sublattices of octahedral interstitial sites, thereby changing the symmetry of the lattice from bcc to bct. The ordering temperature, Tc, is a function of the carbon concentration
29
. If the Ms temperature
exceeds Tc then the martensite is bcc, but when it is below Tc martensite is bct. The ordering energy is a complicated function of temperature and carbon concentration, and was calculated as in Ref. 29. The required free energy is then given by G=
G + GZener.
Figures 6 and 7 show the evolution of G for different grades of V and Nb for carbon concentration of carbon C=0.1 wt.-% and C=0.8 wt.-%. It is suggested from these figures that the influence of the microalloying elements on Ms temperature is small for the lower value of carbon concentration (C=0.1 wt.-%) meanwhile the higher carbon content (C=0.8 wt.-%), the more influence on rising Ms temperature.
Figure 8 shows the influence of PAGS on Ms temperature in a Fe-C steel. It is clear from this figure that an increase on Ms is achieved if the PAGS increases. Likewise, Fig. 8 shows that the lower carbon content, the higher increase in Ms temperature.
As it was mentioned in the previous section, the temperature of initiation of martensitic transformation, and its progress are controlled by the chemical and non-chemical free energies of the system. The chemical free energy difference (G) is the driving force of the transformation and is converted to non-chemical free energy. The latter partly goes into the energy of lattice imperfections inevitable upon transformation. 10 In carbon steels, the morphology of martensite changes with the carbon content. The martensite consists of bundles of laths (lath-shaped martensite) with a high density of dislocations inside each lath in low carbon steels and, as carbon content increases, it changes to lenticular (lens-shaped martensite) with a midrib and a high density of dislocations as well as internal twins. 30 Several studies
31-34
grain size in bursting-type Fe-Ni-C alloys. These authors concluded that the Ms temperature in these alloys is influenced by the PAGS because of the interference with the autocatalytic nature of the burst-type martensitic transformation. However, they describe the martensite morphology as lath-shaped martensite, and lath martensite transformation is often associated with grain boundaries.
32
A nucleation argument,
therefore, would suggest higher Ms temperature (i.e. easier nucleation) as grain size is decreased since grain boundary area increases. In the present analysis, the opposite effect is observed. On the other hand, Brofman and Ansell 33 proposed that the Hall-Petch strengthening of the austenite explains the depression of Ms temperature as PAGS decreases. Hirth 35 has reviewed various theories of grain size strengthening in metals and reported some experimental evidence for the expression
1 D
(8)
where is the dislocation density and D is the austenite grain diameter. If this argument is valid, Hall-Petch strengthening results from increasing dislocation density with decreasing grain size, thus directly strengthening the matrix, i.e. the austenite. Therefore, a reduction of the grain size will increase the resistance of the austenite to plastic deformation locally as well as macroscopically. This increased austenite resistance will directly impede the martensite transformation by increasing the nonchemical free energy opposing the transformation.
36
Ms temperature as PAGS decreases. However fails to explain why Ms temperature increases faster as PAGS increases in low carbon (C=0.1 wt.-%) than in a high carbon (C=0.8 wt.-%) steels. An alternative explanation is outlined considering that higher austenitisation temperatures are required to achieve the same PAGS in low carbon steels than in high carbon steels. Therefore, a more likely cause of rising the Ms temperature as PAGS increases is the reduction of the energy needed for the complementary shear during transformation, which originates in the elimination of lattice imperfections due to higher austenitisation temperature. 37 Likewise, the nucleation of martensite may be boosted by an increase of frozen-in vacancies into the austenite grain due to higher quenching temperatures in low carbon steels. The increase in vacancies makes the austenite phase less stable by increasing the nucleation sites 9,37. Therefore, assuming the austenitisation temperature and the quenching temperature identical, Ms temperature should increase as austenitisation temperature increases since a higher quenching temperature produces more frozen-in vacancies and hence more nucleation sites.
4.3 Validation of the model Figure 9 shows a comparison between the neural network model predictions and the experimentally measured Ms temperatures carried out in five very different alloys whose actual compositions are listed in Table 2. S1 and S2 are low carbon HSLA steels microalloyed with Nb, S3 and S4 are medium carbon forging steel with and without Nb as microalloying element, respectively, and S5 is a commercial martensitic stainless steel. All of these are used for commercial purposes, and therefore, their Ms temperature is a critical parameter whose determination is important in the processing route of the steel. Hence, its accurate determination is very interesting from an industrial point of view. It could be concluded from the figure that the neural network model presents an excellent accuracy on Ms temperature prediction.
5. Conclusions
A neural network method based on a Bayesian framework has been used to rationalise an enormous quantity of published experimental data on Ms temperature of steels. It is now possible, therefore, to estimate the role of elements such as V and Nb whose use as microalloying elements has recently increased due to the good combination of mechanical properties that microalloyed steels present. The formulated neural network model has been also applied towards the understanding of the role of the austenite grain size on stabilising the austenite or, by contrast, on triggering the martensite transformation. The increase of Ms temperature as PAGS increases seems to reveal that the reduction of lattice imperfections and the increase in frozen-in vacancies due to higher austenitisation temperatures increases martensite nucleation sites and then enhance the transformation. It has been demonstrated that the
model gives a good estimation of the experimental Ms temperatures obtained in different kinds of commercial steels.
6. Acknowledgements
The authors acknowledge financial support from Comisin Interministerial de Ciencia y Tecnologa (PETRI 1995-0436-OP). F.G. Caballero would like to thank the Consejera de Educacin, D.G. de Investigacin de la Comunidad Autnoma de Madrid (CAM) for the financial support in the form of a Postdoctoral Research Grant. C. Capdevila would like to express his gratitude to the Consejo Superior de Investigaciones Cientifcas for financial support as a Post-Doctoral contract (I3P PC-2001-1). The authors gratefully acknowledge to Prof. H.K.D.H. Bhadeshia F.R.S of the University of Cambridge for his fruitful discussions.
7. References
1. J.W. Christian: Theory of phase transformations in metals and alloys, 803; 1965, Oxford, Pergamon press. 2. L.A. Capella: Metal. Prog., 1944, 46, 108-118. 3. J. Wang, P.J. van der Wolk and S. Van der Zwaag: Mater. Trans. JIM, 2000, 41, 761-768. 4. R.A. Grange and HM. Stewart: Trans. AIME, 1946, 167, 467-494. 5. P. Payson and H. Savage: Trans ASM, 1944, 33, 261-280. 6. C.Y. Kung and J.J. Rayment: Hardenability Concepts with Applications to Steels, 229; 1978, Warrendale, PA, TMS-AIME. 7. K.W. Andrews: JISI, 1965, 203, 721-727. 8. G.B. Olson and M. Cohen: Metall. Trans. A, 1976, 7A, 1897-1923. 9. G. Ghosh and G.B. Olson: Acta Metall. Mater., 1994, 42, 3361-3370. 10. Z. Nishiyama: Martensitic Transformation, (ed.: M. E. Fine et al.), 13; 1978, New York, Academic Press. 11. M. Economopoulos, N. Lambert, and L. Habraken: Diagrames de transformation des aciers fabriques dans le Benelux, 80; 1967, Bruxelles, Centre National de Reserches Metakllurgiques. 12. M. Atkins: Atlas of continuous cooling transformation diagrams for engineering steels, 17; 1985, Sheffield, British Steels Corporation. 13. J. Wang, P. Van der Wolk and S. Van der Zwaag: ISIJ Int., 1999, 39, 1038-1046. 14. F.G. Caballero, H.K.D.H. Bhadeshia, K.J.A. Mawella, D.G. Jones and P. Brown: Mater. Sci. Technol., 2001, 17, 517-522.
15. F.G. Caballero, H.K.D.H. Bhadeshia, K.J.A. Mawella, D.G. Jones and P. Brown: Mater. Sci. Technol., 2001, 17, 512-516. 16. T. Cool, H.K.D.H. Bhadeshia and D.J.C. MacKay: Mat. Sci. Engineer., 1997, A233, 186-200. 17. D.J.C. MaKay: Neural Comput., 1992, 4, 698-705. 18. D.J.C. MaKay: Darwin college J., 1993, 81-93. 19. D.J.C. MaKay: Neural Comput., 1992, 4, 415-422. 20. D.J.C. MaKay: Neural Comput., 1992, 4, 448-460. 21. H.K.D.H. Bhadeshia: ISIJ Int., 1999, 39, 965-979. 22. H.K.D.H. Bhadeshia: Met. Sci., 1981, 15, 175-177. 23. H.K.D.H. Bhadeshia: Met. Sci., 1981, 15, 178-180. 24. G.B. Olson and M. Cohen: Metall. Trans. A, 1976, 7A, 1905-1914. 25. G.B.Olson and M. Cohen: Metall. Trans., 7A, 1976, 1915-1923. 26. Metallurgical and Thermochemical Databank, 1996, Teddington, Middlessex, National Physical Laboratory. 27. M. Hillert and L.I. Staffansson: Acta Chem. Scand., 1970, 24, 3618-3625. 28. C. Zener: Trans. AIME, 1946, 167, 513-550. 29. J.C. Fisher: Trans. AIME, 1949, 185, 688-690. 30. Z. Nishiyama: Martensitic Transformation, (ed.: M. E. Fine et al.), 28; 1978, New York, Academic Press. 31. M. Unemoto and W.S. Owen: Metall. Trans., 1975, 5, 2041-2053. 32. O.A. Ankara, A.S. Sastri, and D.R.F. West: J. Iron and Steel Institute, 1966, May, 509-511. 33. P.J. Brofman and G.S. Ansell: Metall. Trans. A, 1983, 14A, 1929-1931. 34. T. Maki, S. Shimooka, and I. Tamura: Metall. Trans., 1971, 2, 2944-2955.
35. J.P. Hirth: Metall. Trans., 1972, 3, 3047-3065. 36. E.M. Breinan and G.S. Ansell: Metall. Trans., 1970, 1, 1513-1520. 37. Z. Nishiyama, Martensitic Transformation, (ed.: M. E. Fine et al.), 307; 1978, New York, Academic Press.
Table 1. Variables that influence Ms temperature. SD is standard deviation Range Average SD C, wt.-% Mn, wt.-% Si, wt.-% Cr, wt.-% Ni, wt.-% Mo, wt.-% V, wt.-% Co, wt.-% Al, wt.-% W, wt.-% Cu, wt.-% Nb, wt.-% B, wt.-% N, wt.-% PAGS, m 0 1.62 0 1.98 0 - 3.400 0 - 17.98 0 - 27.20 0 - 5.10 0 - 4.55 0 - 12.27 0 - 1.10 0 - 12.99 0 - 0.28 0 - 0.2 0 - 0.001 0.0001-0.0140 5 - 339 0.3774 0.8652 0.2328 1.4255 1.9705 0.1850 0.0758 0.0820 0.0187 0.1269 0.0388 0.0004 0.0000 0.0015 55.1731 0.2098 0.4882 0.3836 3.2575 5.3646 0.4611 0.3579 0.8145 0.1183 1.0665 0.0778 0.0048 0.0001 0.0039 37.3855
Table 2. Chemical compositions and PAGS of the steel tested. %C S1 S2 S3 S4 S5 0,07 0,20 0,31 0,60 0,37 % Mn 1,50 1,10 1,22 0,49 0,34 % Si 0,37 0,34 0,25 0,35 0,94 % Cr 0,039 0,011 0,138 14,52 4,80 % Ni 0,49 0,02 0,09 0,17 0 % Mo 0,021 0,008 0,03 0,57 1,34 %V 0,004 0,009 0,004 0,18 1,19 % Nb 0,027 0,003 0 0 0 PAGS, m 9 10 21 50 30
Figure 1.- Neural Network model used in this study Figure 2. Variation of V as a function of hidden units. Figure 3. Variation of test error as a number of hidden units. Figure 4. Comparison between predicted and experimental Ms values: (a) training data and (b) test data. Figure 5. Effect of (a) V and (b) Nb on Ms temperature. PAGS=20 m.
Figure 6. Effect of V on G and GC ' for (a) C=0.1 wt.-% and (b) C=0.8wt.-%. Horizontal lines represent GC ' . Figure 7. Effect of Nb on G and GC ' for (a) C=0.1 wt.-% and (b) C=0.8wt.-%. Horizontal lines represent GC ' .
Figure 8. Effect of PAGS on Ms temperature for a Fe-C steel. Figure 9. Comparisson between experimental and calculated Ms temperature for the steels listed in Table 2.
1 0.8
Test Error
Measured Normalised Ms
(a) 0.5 Predicted Normalised Ms . 0.25 0 -0.25 -0.5 -0.5 -0.25 0 0.25 0.5 Measured Normalised Ms (b) Figure 4. Comparison between predicted and experimental Ms values: (a) training data and (b) test data.
900
700
Ms, K
500
C=0.1 w t.-% C=0.4 w t.-% C=0.8 w t.-%
300 0 1 (a)
900
V, wt.-%
700
Ms, K
500
C=0.1 w t.-% C=0.4 w t.-% C=0.8 w t.-%
300 0
0.05
0.1
0.15
0.2
Nb, wt.-%
(b) Figure 5. Effect of (a) V and (b) Nb on Ms temperature. PAGS=20 m.
C=0.1 wt.-%
-1 '
G , J mol
-1000
-2000 V=4 wt.-% -3000 273 373 473 573 V=0 wt.-% 673 773
Temperature, K
(a)
0
C=0.8 wt.-%
G , J mol
-1000
'
-1
-2000 V=4 wt.-% V=0 wt.-% -3000 273 373 473 573 673
Temperature, K
(b)
Figure 6. Effect of V on G and GC ' for (a) C=0.1 wt.-% and (b) C=0.8wt.-%.
C=0.1 wt.-%
-1 '
G , J mol
-1000
-2000 Nb=0.08 wt.-% -3000 273 373 473 Nb=0 wt.-% 573 673 773
Temperature, K (a)
0
C=0.8 wt.-%
-1 '
G , J mol
-1000
-2000 Nb=0.08 wt.-% -3000 273 373 473 Nb=0 wt.-% 573 673
Temperature, K
(b)
Figure 7. Effect of Nb on G and GC ' for (a) C=0.1 wt.-% and (b) C=0.8wt.-%.
850 800 750 700 650 600 550 500 450 0 100 200 300
C=0.8 wt.-% C=0.1 wt.-%
MS, K
PAGS, m
750
Predicted Ms, K
650
550
Measured Ms, K
Figure 9. Comparisson between experimental and calculated Ms temperature for the steels listed in Table 2.