Machine learning-based predictions of fatigue life and fatigue limit for steels

doi:10.1016/j.jmst.2021.02.021

Abstract

Abstract:

To predict the fatigue life for oblique hyperbola- and bilinear-mode S-N curves of metallic materials with various strengths, a machine-learning approach for direct analysis was employed. Additionally, to determine the fatigue limit of the utilized materials (AISI 316, AISI 4140 and CA6NM series) with different S-N curve modes using finite-fatigue life data, a Bayesian optimization-based inverse analysis was performed. The results indicated that predictions of the fatigue life for the utilized datasets via the random forest (RF) algorithm for AISI 4140 and CA6NM, and artificial neural network (ANN) for AISI 316, distribute within 2 factor error lines for most data. In the Bayesian optimization-based inverse analysis, the specific explanatory variables corresponding to the optimized maximum fatigue life were treated as the fatigue limits. The predicted fatigue limits either approximated to or slightly underestimated the experimental results, except for several cases with large errors. Using the inverse analysis to predict the fatigue limit for both S-N curve modes is applicable for current employed data-set. However, the explored maximum fatigue lives via BO corresponding to the predicted fatigue limit were underestimated for AISI 4140 and CA6NM, and was overestimated for AISI 316 because of effect of shape of S-N curves. By combining the ANN or RF direct and BO inverse algorithms, whole S-N curves (including the fatigue limit) were evaluated for the S-N curve shapes of the oblique hyperbola and bilinear modes.

Key words: Machine learning, Fatigue life prediction, Inverse analysis, Steels

Lei He, ZhiLei Wang, Hiroyuki Akebono, Atsushi Sugeta. Machine learning-based predictions of fatigue life and fatigue limit for steels[J]. J. Mater. Sci. Technol., 2021, 90: 9-19.

Figures/Tables 14

Table 1 Mechanical properties of the utilized materials.

	Tensile stress σ_TS (MPa)	Yield stress σ_YS (MPa)	Hardness Hv	Number of data
AISI 316	592	264	176	8	Present work
	570	263	162	6	Previous work [16]
	587.5	259.1	203.5	10	Literature [21]
2RM2	968	928	327	8	Previous work [18]
CA6NM	830*	600*	280*	10*	Previous work [19]
	830	600	280	11	Previous work [17]
	918	575	277	53	Literature [22]
CA15	630*	420*	186.7*	7*	Previous work [20]
AISI 4140	950	800	314	13	Present work
	1019	976	348	12	Literature [23]
	1030	949	337	10
	1062	975	339	12
	1027	942	336	11

Table 2a Chemical composition of AISI 316 (mass%).

	C	Si	Mn	P	S	Ni	Cr	Mo	Co	Cu	N	Fe
AISI 316	0.04	0.25	1.32	0.037	0.03	10.04	16.91	2.02	0.30	0	0	69.053	Present work
	0.04	0.30	1.25	0.038	0.025	10.13	16.01	2.00	0	0	0	70.207	[16]
	0.009	0.39	1.75	0.029	0.002	10.20	16.31	2.07	0.16	0.23	0.11	68.740	[21]

Table 2b Chemical composition of CA6NM series (mass%).

	C	Si	Mn	P	S	Ni	Cr	Mo	N	Al	Fe
2RM2	0.023	0.40	0.56	0.006	0.003	5.14	12.63	0.31	0	0	80.928	[18]
CA6NM	0.049	0.50	0.83	0.04	0.004	3.62	12.82	0	0.027	0.01	82.100	[17,19]
CA6NM	0.06	1.00	1.00	0.04	0.03	4.00	12.75	0.52	0	0	80.600	[22]
CA15	0.11	0.40	0.67	0.021	0.009	0.08	12.68	0	0.017	0.004	86.009	[20]

Table 2c Chemical composition of AISI 4140 (mass%).

	C	Si	Mn	P	S	Cu	Ni	Cr	Mo	Fe
AISI 4140	0.42	0.22	0.80	0.008	0.006	0.01	0.02	1.06	0.16	97.296	Present work
AISI 4140	Max:0.43 Min:0.38	Max: 0.35 Min: 0.15	Max:0.85 Min:0.60	0.03	0.03	0.30	0.25	Max: 1.2 Min: 0.9	Max: 0.30 Min: 0.15	--	[23] *

Fig. 1. S-N curve of AISI 316 and AISI 4140.

Fig. 2. Variables selected results via Lasso function: (a) AISI 316 (b) AISI 4140 (c) CA6NM series.

Fig. 3. Machine learning results of AISI 316 with dataset ratio: (a) ANN (train/test) 9:1 (b) SVR 9:1(c) RF 9:1 (d) ANN 8:2 (e) SVR 8:2 (f) RF 8:2 (g) ANN 7:3 (h) SVR 7:3 (i) RF 7:3. Shaded area: dataset without variables selected.

Fig. 4. Machine learning results of AISI 4140 with dataset ratio: (a) ANN (train/test) 9:1 (b) SVR 9:1(c) RF 9:1 (d) ANN 8:2 (e) SVR 8:2 (f) RF 8:2 (g) ANN 7:3 (h) SVR 7:3 (i) RF 7:3. Shaded area: dataset without variables selected.

Fig. 5. Machine learning results of CA6NM series with dataset ratio: (a) ANN (train/test) 9:1 (b) SVR 9:1(c) RF 9:1 (d) ANN 8:2 (e) SVR 8:2 (f) RF 8:2 (g) ANN 7:3 (h) SVR 7:3 (i) RF 7:3. Shaded area: dataset without variables selected.

Fig. 6. Accuracy of constructed model by various algorithms: (a) AISI 316-ANN (b) AISI 4140-RF (c) CA6NM series-RF.

Fig. 7. Comparison of the fatigue life between the experimental and prediction results:AISI 316; (b) AISI 4140; (c) CA6NM series.

Fig. 8. Inverse analysis based on BO: (a) AISI 316, and (b) AISI 4140, (c) CA6NM series.

Fig. 9. Comparison of the fatigue limit between the experimental and prediction results.

Fig. 10. S-N curves of experimental and prediction results with fatigue limits. (a) AISI 316; (b) AISI 4140; (c) CA6NM [17].

References 39

[1]	S.S. Tschegg, Int. J. Fatigue 60 (2014) 2-17. DOI URL
[2]	Z.Y. Huang, D. Wagner, C. Bathias, J.L. Chaboche, Int. J. Fatigue 33 (2011) 115-121. DOI URL
[3]	K. Shibanuma, K. Ueda, H. Ito, Y. Nemoto, M. Kinefuchi, K. Suzuki, M. Enoki, Mater. Des. 139 (2018) 269-282. DOI URL
[4]	NIMS Fatigue Data Sheet 95, National Institute of Materials Science, Japan, 2004.
[5]	K. Worden, G. Manson, Philos. Trans. Math. Phys. Eng. Sci. 365 (2007) 515-537.
[6]	A. Shoeb, J. Guttag, Application of Machine Learning to Epileptic Seizure Detection, in: Proceedings of the 27th International Conference on Machine Learning, Haifa, Israel, 2010.
[7]	H. Bhaskar, D.C. Hoyle, S. Singh, Comput. Biol. Med. 36 (2006) 1104-1125. PMID
[8]	P.J. García Nieto, E. García-Gonzalo, J.C. Álvarez Antón, V.M.González Suárezb R. Mayo Bayón, F. Mateos Martín, J. Comput. Appl. Math. 330 (2018) 877-895. DOI URL
[9]	P. Aminian, H. Niroomand, A.H. Gandomi, A.H. Alavi, M.A. Esmaeili, Neural Comput. Appl. 23 (2013) 119-131. DOI URL
[10]	X.Y. Li, C.C. Roth, D. Mohr, Int. J. Plast. 118 (2019) 320-344. DOI URL
[11]	A. Agrawal, P.D. Deshpande, A. Cecen, G.P. Basavarsu, A.N. Choudhary, S.R. Kalidindi, Integr. Mater. Manuf. Innov. 3 (2014) 90-108. DOI URL
[12]	H.Y. Wan, G.F. Chen, C.P. Li, X.B. Qi, G.P. Zhang, J. Mater. Sci. Technol. 35 (2019) 1137-1146. DOI
[13]	A. Agrawal, A. Choudhary, APL Mater. 4 (2016) 1-10, 053208.
[14]	A. Rovinelli, M.D. Sangid, H. Proudhon, W. Ludwig, Npj Comput. Mater. 35 (2018) 1-10.
[15]	M. Zhang, C.-N. Sun, X. Zhang, P.C. Goh, J. Wei, D. Hardacre, H. Li, Int. J. Fatigue 128 (2019), 105194. DOI URL
[16]	L. He, H. Akebono, M. Kato, A. Sugeta, Int. J. Fatigue 101 (2017) 18-26. DOI URL
[17]	M. Natsume, Y. Hayashi, H. Akebono, M. Kato, A. Sugeta, Procedia Eng. 2 (2009) 1273-1281. DOI URL
[18]	L. He, H. Akebono, A. Sugeta, Y. Hayashi, Fatigue Fract. Eng. Mater. Struct. 43 (2020) 1419-1432. DOI URL
[19]	J. Arakawa, M. Kakuta, Y. Hayashi, R. Tanegashima, H. Akebono, M. Kato, A. Sugeta, Surf. Eng. 30 (2014) 662-669. DOI URL
[20]	H. Akebono, S. Matsuzaki, M. Soﬁan, Y. Hayashi, M. Kato, A. Sugeta, Japan Soc. Mech. Eng. 75-759 (2009) 1585-1590, in Japanese.
[21]	J.P. Strizak, H. Tian, P.K. Liaw, L.K. Mansur, J. Nucl. Mater. 343 (2005) 134-144. DOI URL
[22]	L.B. Winck, J.L.A. Ferreira, J.A. Araujo, M.D. Manfrinato, C.R.M. da Silva, Surf. Coat. Technol. 232 (2013) 844-850. DOI URL
[23]	K. Tanaka, S. Nishijima, S. Matsuoka, T. Abe, F. Kouzu, Fatigue Fract. Eng. Mater. Struct. 4 (1981) 97-108. DOI URL
[24]	P.P. Milella, Springer press, 2013, pp. 150-155.
[25]	R.J. Schalkoff, New York, 1997.
[26]	M.A. Hearst, S.T. Dumais, E. Osuna, J. Platt, B. Scholkopf, IEEE Intell. Syst. Appl. 13 (1998) 18-28.
[27]	A. Liaw, W. Matthew, R News 2 (2002) 18-22.
[28]	M. Pelikan, D.E. Goldberg, E. Cant ´u-Paz, in: W. Banzhaf, J.M. Daida, A.E. Eiben, M. H. Garzon, V. Honavar (Eds.), Proc. of the 1st Annual Conf. on Genetic and Evolutionary Computation, Vol. 1, Morgan Kaufmann Publishers, Orlando, FL, 1999, p. 525.
[29]	Z.L. Wang, T. Ogawa, Y. Adachi, Adv. Theory Simul. 2 (2019), 1900110. DOI URL
[30]	Z.L. Wang, T. Ogawa, Y. Adachi, Adv. Theory Simul. 3 (2020), 1900177. DOI URL
[31]	L. He, H. Akebono, A. Sugeta, Int. J. Fatigue 116 (2018) 388-395. DOI URL
[32]	T. Sakai, A. Sugeta, JSMS, 2007.
[33]	M. Rahaman, W.Z. Mu, J. Odqvist, P. Hedstrom, Metall. Mater. Trans. A 50 A (2019) 2081-2091. DOI URL
[34]	R. Tibshirani, J. R. Stat. Soc. Ser. B 58 (1996) 267-288.
[35]	H. Akaike, Automat. Control 19 (1974) 716-723.
[36]	H.S. Bhat, N. Kumar, On the Derivation of the Bayesian information criterion, in: School of Natural Sciences, University of California, 2010.
[37]	M.M. Jin, P.H. Cao, M.P. Short, J. Nucl. Mater. 523 (2019) 189-197. DOI URL
[38]	Z.L. Wang, Y. Adachi, Mater. Sci. Eng. A 744 (2019) 661-670. DOI URL
[39]	Y. Hozo, H. Akebono, M. Kato, A. Sugeta, Japan Soc. Mech. Eng. (2012), 12-5 (M&M2012 conference) (in Japanese).