Proper Orthogonal Decomposition-Deep-Learning-based Prediction of Plastic Properties Using Indentation Surface Displacement

The purpose of this study is to predict the plastic properties of material using surface displacement caused by indentation. Depending on the material properties, the surface displacement exhibits different aspects. The surface displacement can be measured using a digital image correlation method, which is a noncontact measurement method. The surface displacement database for each plastic property necessary for property prediction was constructed through numerical simulation, with plastic properties represented by yield strength and the strain hardening exponent. The database was compressed through proper orthogonal decomposition (POD) for deep learning data. POD is a preprocessing technique that compresses data according to the greatest influence. The trained deep learning model predicts plastic properties by indentation surface displacement. By using this technique, the yield strength was predicted within an average error rate of 2.7%, and the strain hardening exponent was predicted within an average error rate of 5.7%.