Determination of Material Parameters from Single Spherical Indentation Data Using Artificial Neural Networks

This study presents a novel artificial neural network approach for determining material parameters from single spherical indentation test data, addressing the fundamental non-uniqueness problem inherent in indentation-based material property identification. A multi-layer perceptron neural network architecture was developed to directly process force-displacement curves and residual imprint profiles. Three neural network models utilizing different input configurations, force-displacement curves only (FD), residual imprint profiles only (RI), and both data types combined (FD+RI), were systematically compared to predict six material parameters: elastic modulus, yield strength, tensile strength, and three Voce equation parameters (σy0,Q,β). The results demonstrate that residual imprint data alone proved sufficient for achieving high prediction accuracy across most material parameters, while force-displacement curves alone exhibited significant limitations in enabling unique material property determination. Although substantial prediction errors in flow stress occurred for austenitic stainless steels and nickel alloys due to their extremely small saturation rate parameters, applying β correction using predicted material parameters dramatically improved both R2 values and normalized integral absolute errors. This methodology successfully resolves the non-uniqueness challenge in indentation-based material characterization and establishes that unique determination of material parameters is achievable using only residual imprint profiles from single spherical indentation tests.