ANN strategies for the stress–strain analysis of metallic materials: Modeling, database, supervised learning, validation and performance analysis

Abstract

Artificial neural networks (ANN) are developed and employed to characterize a wide range of metallic materials. Focus is given to the evaluation of stress–strain behavior via sphere-to-flat indentation. Each ANN is trained using a supervised machine learning procedure comprised of two steps: (i) generation of a training dataset via calibrated finite element model, and (ii) validation using experimental data retrieved from indentations tests. The developed frameworks aim to establish a fast and low-cost tool for the assessment of loading conditions in industrial applications. The best proposed solution is able to predict stress–strain behavior with a quasi-instantaneous response and errors of less than 3%. Moreover, outputs are attained with minute costs (processing and memory bandwidth) when compared to finite element simulations.