Novel deep learning approach for practical applications of indentation

Abstract

The instrumented indentation technique has been investigated to efficiently evaluate the mechanical responses of materials with few limitations on the shape and size of the specimen. There have been attempts to discover a direct correlation between the stress-strain curve and the indenting load-displacement curve by introducing the concept of representative strain and stress. However, it is still difficult to find relible parameters and to distinguish similar load-displacement curves that correspond to different stress-strain curves with a limited number of experimental datasets. The present study introduces a finite element method (FEM)-based simulation that can output various load-displacement datasets corresponding to intrinsic properties of materials, including strain rate; these datasets are validated using experimental indentation results for diverse metallic materials at different indenting speeds (0.6, 0.9, 1.2 mm/min). In addition, an autoencoder (AE)-shaped artificial neural network (ANN) model is designed to efficiently characterize those datasets. Then, the indenting load-displacement datasets are extracted into effective physically meaningful datasets by introducing a data post-processing procedure. The proposed indentation FEM-AE-shaped ANN model demonstrates that a long-range true stress-strain curve can be attained even from a noisy experimental load-displacement dataset.

Keywords

• Spherical indentation
• Finite element method
• Deep learning
• Mechanical property
• Strain rate sensitivity