Machine learning based optimization method for vacuum carburizing process and its application

Abstract

This paper develops an optimized prediction method based on machine learning for optimal process parameters for vacuum carburizing. The critical point is data expansion through machine learning based on a few parameters and data, which leads to optimizing parameters for vacuum carburization in heat treatment. This method extends the data volume by constructing a neural network with data augmentation in the presence of small data samples. In this paper, the database of 213 data is expanded to a database of 2,116,800 data by optimizing the prediction. Finally, we found the optimal vacuum carburizing process parameters through the vast database. The relative error of the three targets is less than that of the target obtained by the simulation of the corresponding parameters. The relative error is less than 5.6%, 1%, and 0.02%, respectively. Compared to simulations and actual experiments, the optimized prediction method in this paper saves much computational time. It provides a large amount of referable process parameter data while ensuring a certain level of accuracy.

Keywords

Machine learning / heat treatment / neural networks / data augmentation / small sample

References

[1] George E. Totten. Quenching and distortion control. ASM Internaional; 1993.p.341.

[2] Canale LDC.Overview of distortion and residual stress due to quench processing. Part I: factors affecting quench distortion.IJMPT2005;24:4

[3] Zajusz M,Danielewski M.Modeling of vacuum pulse carburizing of steel.Surface and Coatings Technology2014;258:646-51

[4] Uchida F,Kamada K.Simulation of quenching crack generation in low alloy cast steel by using heat-treatment cae software.J Soc Mater Eng Resour Jpn2004;17:8-12

[5] Todo S,Imataka H. Development of application technology for vacuum carburizing. Shinnittetsu Sumikin giho;2016.p.14–19. Available from: https://www.nipponsteel.com/en/tech/report/nssmc/pdf/116-04.pdf [Last accessed on 20 Apr 2023]

[6] Inoue T,Okamura K.Metallo-thermo-mechanical simulation of carburized quenching process by several codes - a benchmark project -.KEM2007;340-341:1061-6

[7] Ju DY,Omori T. Residual stresses and distortion of a ring in quenching-tempering process based on metallo-thermo-mechanics. Proceeding of 2nd International Conference on Quenching and the Control of Distortion; 1996 Dec 194-257. Available from: https://www.osti.gov/biblio/484269 [Last accessed on 20 Apr 2023]

[8] Ju D.Simulation of thermo-mechanical behavior and interfacial stress of metal matrix composite under thermal shock process.Composite Structures2000;48:113-8

[9] Ju D,Zhang Y.Modeling and experimental verification of martensitic transformation plastic behavior in carbon steel for quenching process.Materials Science and Engineering: A2006;438-440:246-50

[10] Liu C,Liu Z.Ju D-Y. Bainitic transformation kinetics and stress assisted transformation.Materials Science and Technology2001;17:1229-37

[11] Wołowiec-korecka E.Modeling methods for gas quenching, low-pressure carburizing and low-pressure nitriding.Engineering Structures2018;177:489-505

[12] Lisjak D.Determination of steel carburizing parameters by using neural network.Materials and Manufacturing Processes2009;24:772-80

[13] Ju D. Simulation of carburizing of SCr420 by COSMAP and its verification. Quenching deformation simulation 60th Japan Heat Treatment Technology Association Lecture Summary; 2005.

[14] Ba JL,Hinton GE.Layer Normalization. 2016.p.arXiv:1607.06450v1.

[15] Xavier G,Yoshua B. Deep sparse rectifier neural networks. Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics; 2011 Feb PMLR 15:315-323. Available from: http://proceedings.mlr.press/v15/glorot11a [Last accessed on 20 Apr 2023]

[16] Vinod N. Rectified linear units improve restricted boltzmann machines. ICML’10: Proceedings of the 27th International Conference on International Conference on Machine Learning; 2010 Jun 807–814. Available from: https://dl.acm.org/doi/abs/10.5555/3104322.3104425 [Last accessed on 20 Apr 2023]

[17] Rumelhart DE,Williams RJ.Learning representations by back-propagating errors.Nature1986;323:533-6

[18] Kingma DP.Adam: a method for stochastic optimization. international conference on learning representations (ICLR); 2015 May13.