Topological Optimization of Artificial Neural Networks to Estimate Mechanical Properties in Metal Forming Using Machine Learning

The ability of a metal to be subjected forming processes depends mainly on its plastic behavior and, thus, the mechanical properties belonging this region stress–strain curve. Forming techniques are among most widespread metalworking procedures in manufacturing, and aluminum alloys great interest fields as diverse aerospace sector or food industry. A precise characterization is crucial estimate capability equipment, but also for robust numerical modeling processes. Characterizing material very relevant task which large amounts resources invested, paper studies how optimize multilayer neural network able make, through machine learning, accurate predictions about wrought alloys. This study focuses determination ultimate tensile strength, closely related strain hardening material; more precisely, methodology developed that, by randomly partitioning input dataset, performs training prediction cycles that allow estimating average performance each fully-connected topology. In way, trends found networks, it established networks with at least 150 perceptrons their hidden layers, predictive error stabilizes below 4%. Beyond point, no really significant improvements found, although there an increase computational requirements.