Elastic Modulus Prediction from Indentation Using Machine Learning: Considering Tip Geometric Imperfection

Abstract

Instrumented indentation technique provides a simple and quick means to investigate mechanical properties such as hardness and elastic modulus near the material surface. However, accurately predicting plastic pileup/sink-in during indentation remains a hurdle in calibrating real contact depth, affecting precise material property evaluation, especially in metallic materials. This study utilizes machine learning on extensive finite element analysis (FEA) data to exclusively predict elastic modulus from indentation curves. Leveraging comprehensive FEA data from sharp and spherical indentations across diverse material properties, our neural network-based models showcase impressive accuracy, achieving approximately 0.65 and 1.72% Mean Absolute Percentage Error for spherical and sharp indentations, respectively. Furthermore, we address the impact of indenter geometry imperfections on prediction accuracy. Through data normalization and subsequent transfer learning, we effectively minimize the MAPE deviation in predicted elastic modulus between results obtained from perfect and imperfect indenters.

Graphical Abstract