A Data-Driven Methodology for Obtaining the Stress–Strain Curves of Metallic Materials Using Discrete Indentation Tests

Determining the stress–strain curve and other plastic properties using instrumented indentation techniques has long been a topic of active study. The potential to use small, geometrically simple specimens and to characterize a component under service without the need to remove material for specimen preparation makes this methodology highly attractive to many industries. In this study, a data-driven approach that leverages machine learning and finite element analysis was used to construct a model called ‘Brilearn’ that predicts the stress–plastic strain curve of metallic materials. The framework consists of a novel model for predicting the hardening curve, the classical Tabor model for predicting the yield stress for materials with yield stress lower than 100 MPa, and an XGBoost model for predicting the yield stress for metals with yield stress higher than 100 MPa. The model was validated against experimental data on Al1100, Al6061-T6, Al7075-T6, and brass and copper alloys, features error predictions of 8.4 ± 8.5% for the yield stress and 3.2 ± 4% for a complete curve ranging from ε¯p=0 to ε¯p=0.15. The model is especially suited for the determination of the stress–plastic strain curves for components in service since only two simple indentation tests are required.