Can Neural Networks Simulate Metal Strength & Fatigue?

Understanding how metals perform under stress is key to developing new alloys for transportation, aerospace, aviation, and more. In a new publication, Exponent's Adnan Eghtesad and co-authors from academia explore how a new framework based on neural networks can predict the deformation behavior of metals and alloys. 

The authors use the neural network-based elasto-viscoplasticity (NN-EVP) constitutive modeling framework to predict the flow response in metals and alloys as a function of grain size. By integrating state-of-the-art deep learning tools in training the flow response of a metal/alloy, the trained model offers prediction and extrapolation of the elasto-viscoplastic deformation behavior to strains above the measured data. The modeling framework also offers the ability to predict the response of metals with microstructures containing different grain sizes than the existing experimental data. In this work, the team was able to provide insights into the generalization of material behavior under large plastic deformations induced by arbitrary loading states and microstructural grain size. 

From the publication: "The proposed NN-EVP model presented herein takes a further step in the prediction of flow responses in metals and improves the computational efficiency of structure-property-relationship simulations of metallic materials."