Quantitative analysis of mechanical properties associated with aging treatment and microstructure in Mg-Al-Zn alloys through machine learning

doi:10.1016/j.jmst.2021.07.045

Abstract

Abstract:

The present study proposes a methodology for predicting the mechanical properties of AZ61 and AZ91 alloys associated with microstructure, texture and aging parameters and estimating predictor importance. For this, we investigate quantitative correlations between microstructure, texture and mechanical properties of aged AZ61 and AZ91 rods through machine learning. This regression analysis focuses on the precipitation behavior of Mg₁₇Al₁₂ as the main second phase of Mg-Al-Zn alloys with respect to aging conditions. To simplify data generation, only SEM images were used to quantify the features of discontinuous and continuous precipitates. To overcome the lack of data and make the most of the measured data, we devised a method to extend the existing dataset by a factor of 9 using the mean and standard deviation of the measured data. Artificial neural networks predicted tensile and compressive yield strengths and resultant yield asymmetry with a high accuracy of over 98% using 11 predictors for a total of 288 datasets. Decision tree learning quantitatively assessed the importance of predictors in determining the mechanical properties of aged AZ61 and AZ91 rods.

Key words: Magnesium alloy, Aging treatment, Microstructure, Mechanical properties, Machine learning

Joung Sik Suh, Byeong-Chan Suh, Sang Eun Lee, Jun Ho Bae, Byoung Gi Moon. Quantitative analysis of mechanical properties associated with aging treatment and microstructure in Mg-Al-Zn alloys through machine learning[J]. J. Mater. Sci. Technol., 2022, 107: 52-63.

Figures/Tables 16

Table 1 Ranges of input and output variables.

Variable		Minimum	Maximum
Input	Al (wt.%)	6.0	9.0
	T (° C)	25.0	250.0
	log₁₀(h) (h)	0.000	2.380
	DP-A (-)	0.003	0.996
	CP-A (-)	0.000	0.979
	CP-L (µm)	0.001	1.825
	CP-D (µm)	0.000	4.043
	GS (µm)	7.30	29.10
	KAM (-)	0.517	0.669
	I_max (m.r.d.)	3.42	6.20
	m_basal (-)	0.121	0.224
Output	TYS (MPa)	168.3	294.4
	CYS (MPa)	121.8	283.9
	YA (-)	0.711	0.996

Fig. 1. Proposal of data expansion approach: Avg means average and SD means standard deviation.

Fig. 2. Age-hardening curves of AZ61 and AZ91 rods.

Fig. 3. Development of DP and CP in extruded and aged AZ61 rods: (a) as-extruded, (b) measurement of DP in (a), (c) 150 °C for 2 h, (d) 200 °C for 24 h and (e) 250 °C for 240 h, red means DP and yellow means CP.

Fig. 4. Development of DP and CP in extruded and aged AZ91 rods: (a) as-extruded, (b) measurement of DP in (a), (c) 200 °C for 2 h, (d) 200 °C for 8 h, (e) 200 °C for 96 h, (f) 200 °C for 240 h (g) measurement of length of CP and (h) measurement of distance between CP, red means DP and yellow means CP.

Fig. 5. Microstructure and texture development of AZ61 rods: (a, e) IPF map, (b, f) KAM map, (c, g) (0001) pole figure, (d, h) Schmid factor for basal 〈a〉 slip, (a-d) for extruded and (e-h) for aged at 200 °C for 24 h.

Fig. 6. Microstructure and texture development of AZ91 rods: (a, e) IPF map, (b, f) KAM map, (c, g) (0001) pole figure, (d, h) Schmid factor for basal texture, (a-d) for extruded and (e-h) for aged at 200 °C for 24 h.

Fig. 7. Engineering stress-strain curves of aged AZ61 (a) and aged AZ91 (b) rods in uniaxial tension and compression along the ED.

Fig. 8. Prediction of mechanical properties of aged AZ61 and AZ91 rods: (a) TYS, (b) CYS, (c) YA.

Table 2 Comparison of prediction accuracy of training, validation and testing data.

	TYS			CYS			YA
	R²	RMSE	MAPE	R²	RMSE	MAPE	R²	RMSE	MAPE
Training	0.995	2.888	0.708	0.995	4.073	1.225	0.986	0.012	0.768
Validation	0.996	2.538	0.859	0.993	4.673	1.035	0.981	0.014	0.953
Testing	0.998	1.810	0.426	0.995	3.771	1.014	0.982	0.014	1.002

Fig. 9. Estimation of predictor importance for aged AZ61 and AZ91 rods: (a) TYS, (b) CYS, (c) YA.

Fig. 10. Estimation of predictor importance for aged AZ61 rods: (a) TYS, (b) CYS, (c) area fraction of precipitates and distribution of YS with different aging temperatures and times, AE means as-extruded state.

Fig. 11. Contour maps of mechanical properties of AZ61 rods aged with temperature and time: (a) TYS, (b) CYS and (c) YA.

Fig. 12. Estimation of predictor importance for aged AZ91 rods: (a) TYS, (b) CYS, (c) area fraction of precipitates and distribution of YS with different aging temperatures and times, AE means as-extruded state.

Fig. 13. Contour maps of mechanical properties of AZ91 rods regarding area fraction and length of CP: (a) TYS, (b) CYS, (c) SEM image at 150 °C for 24 h, (d) SEM image of CP at 250 °C for 240 h, (e) magnification of CP in (c), (f) magnification of CP in (d), yellow and red mean CP and DP, respectively.

Fig. 14. Contour maps of mechanical properties of AZ91 rods aged with temperature and time: (a) TYS, (b) CYS and (c) YA.

References 30

[1]	J. Wu, L. Jin, J. Dong, F. Wang, S. Dong, J. Mater. Sci. Technol., 42 (2020), pp. 175-189. DOI URL
[2]	J. Zhang, S. Liu, R. Wu, L. Hou, M. Zhang, J. Magnes. Alloys, 6 (2018), pp. 277-291.
[3]	F. Pan, M. Yang, X. Chen, Mater. Sci. Technol., 32 (2016), pp. 1211-1221.
[4]	C.S. Roberts, Magnesium and Its Alloys, John Wiley & Sons, New York (1960).
[5]	E.A. Ball, P.B. Prangnell, Scr. Metall. Mater., 31 (1994), pp. 111-116. DOI URL
[6]	W.F. Hosford, The Mechanics of Crystals and Textured Polycrystals, Oxford University Press, Oxford(1993).
[7]	J. Jain, W.J. Poole, C.W. Sinclair, M.A. Gharghouri, Scr. Mater., 62 (2010), p. 310.
[8]	J.F. Nie, Metall. Mater. Trans. A, 43 (2012), pp. 3891-3939.
[9]	C.R. Brooks, Heat Treatment, American Society for Metals (1984), Metals Parks.
[10]	D.B. Williams, E.P. Butler, Int. Mater. Rev., 26 (1981), pp. 153-183. DOI URL
[11]	D. Duly, M.C. Cheynet, Y. Berchet, Acta Metall. Mater., 42 (1994), pp. 3843-3854. DOI URL
[12]	S. Celotto, T.J. Bastow, Acta Mater., 49 (2001), pp. 1211-1221.
[13]	K.N. Braszczyńska-Malik, J. Alloys Compd., 477 (2009), pp. 870-876. DOI URL
[14]	S. Yu, Y. Gao, C. Liu, X. Han, J. Alloys Compd., 477 (2015), pp. 431-436.
[15]	S.H. Kim, J.U. Lee, Y.J. Kim, J.H. Bae, B.S. You, S.H. Park, J. Mater. Sci. Technol., 34 (2018), pp. 265-276. DOI URL
[16]	J.U. Lee, S.H. Kim, Y.J. Kim, S.H. Park, Mater. Sci. Eng. A, 714 (2018), pp. 49-58. DOI URL
[17]	S.Y. Lee, S. Byeon, H.S. Kim, H. Jin, S. Lee, Mater. Des., 197 (2021), pp. 1-16.
[18]	Z. Tong, L. Wang, G. Zhu, X. Zeng, Metall. Mater. Trans. A, 50 (2019), pp. 5543-5560. DOI URL
[19]	R. Willumeit, F. Feyerabend, N. Huber, Acta Biomater., 9 (2013), pp. 8722-8729. DOI PMID
[20]	S. Bingöl, W. Misiolek, J. Mech. Eng., 61 (2015), pp. 610-617.
[21]	S.A. Sani, G.R. Ebrahimi, H. Vafaeenezhad, A.R. Kiani-Rashid, J. Magnes. Alloys, 6 (2018), pp. 134-144. DOI URL
[22]	D.J.C. MacKay, Neural Comput., 4 (1992), pp. 415-447. DOI URL
[23]	F.D. Foresee, T.H. Martin, Gauss-Newton approximation to Bayesian learning, Proceedings of the International Joint Conference on Neural Networks, IEEE (1997).
[24]	A.D. Aczel, Complete Business Statistics, McGraw-Hill Higher Education, Irwin (1989).
[25]	C.D. Lewis, Industrial and Business Forecasting Methods, Butterworths, London(1982).
[26]	L. Breiman, J.H. Friedman, R.A. Olshen, C.J. Stone, Classification and Regression Trees, Chapman & Hall, Boca Raton, FL(1984).
[27]	L. Rokach, O. Maimon, IEEE Trans. Syst. Man Cybern. Part C, 35 (2005), pp. 476-487. DOI URL
[28]	M.G. Jiang, H. Yan, R.S. Chen, Mater. Sci. Eng. A, 621 (2015), pp. 204-211. DOI URL
[29]	C.C. Zhang, H.Y. Wang, M. Zha, C. Wang, J.H. Li, Z.Z. Yang, Q.C. Jiang, Materials, 11 (2018), p. 895.(Basel). DOI URL
[30]	J.F. Nie, Scr. Mater., 48 (2003), pp. 1009-1015. DOI URL