Extraction of the plastic properties of metallic materials from scratch tests using deep learning

Powered by machine learning and computer technology, neural networks have opened new paths for solving engineering problems. In this paper, the plastic parameters, i.e., the yield stress and strain hardening index, of metallic materials are extracted from scratch tests using deep learning methods. Using a dataset generated by finite element simulations, three network models, i.e., the classical multi-output multi-layer perceptron (MLP), a single-target approach (ST-MLP) and the parameter sharing-based deep network (DMTR), are adopted to determine the relationship between scratch responses and plastic parameters. According to the test dataset re-sults, the DMTR performs better than the MLP and ST-MLP. The trained DMTR is verified by comparing the plastic parameters of 18CrNiMo7-6 alloy steel, 304 stainless steel, and brass obtained from scratch tests with those under tension. This work is expected to provide an alternative method for determining the plastic pa-rameters of metallic materials.

Automated summary

This paper utilizes neural networks and deep learning methods to extract plastic parameters of metallic materials from scratch tests, establishing the relationship between scratch responses and plastic parameters using various network models, with the DMTR model outperforming the other two.