Machine Learning for Metal Processing

We propose applications based on the use of a deep neural network as fast surrogate models for local stress calculations in inhomogeneous non-linear materials. We show that deep neural networks can predict the local stresses with &lt;4% mean absolute error for the case of heterogeneous elastic media and a mechanical contrast of up to factor of 1.5 among neighboring material domains, while performing 2 orders of times faster than spectral solvers or finite element solvers. Deep neural network models prove suited for reproducing the stress distribution in geometries different from those used for training. In the case of elasto-plastic materials with up to 4 times mechanical contrast in yield stress among adjacent regions, the trained model simulates the micromechanics with an error &lt;7% in one single forward evaluation of the network, without any iteration. The results reveal an efficient approach to solve non-linear mechanical problems, with an acceleration up to a factor of nearly four orders of magnitude for elastic-plastic materials compared to conventional solvers (1). We also show similar applications of the case of heat treatment of metals.<br/><br/>(1) Mianroodi, J.R., H. Siboni, N. & Raabe, D. Teaching solid mechanics to artificial intelligence—a fast solver for heterogeneous materials. npj Comput Mater 7, 99 (2021).