Learning Johnson–Cook Parameters From Chip Formation and Cutting Forces Using a Multimodal Machine Learning Framework

Abstract The Johnson–Cook (J–C) plasticity and damage models are widely used for simulating machining processes under large strains, high strain rates, and high temperatures. However, determining the J–C parameters (A, B, n, C, and m) and associated damage parameters remains technically challenging since these parameters can hardly be individually measured. This article presents a multimodal machine learning method to learn the J–C parameters and a simplified damage parameter in response to chip-formation images and cutting force data from orthogonal cutting simulations. The proposed method employs convolutional neural networks to extract spatial features from chip images and utilizes a multilayer perceptron to learn from different data sources. Two numerical studies are conducted to test the method. Study 1 attempts to identify all five J–C parameters simultaneously. Although the non-uniqueness of the J–C model prevents a unique solution, the multimodal approach still provides the most reliable predictions. Study 2 determines the strain-rate hardening and thermal softening effects (C and m), as well as the fracture strain (ε¯Dpl) in the ductile damage criterion. With known A, B, and n from tensile testing, the results demonstrate that C, m, and ε¯Dpl can be uniquely determined. Overall, studies show that multimodal machine learning can be an effective method for inverse analysis in estimating the plasticity and damage model. Moreover, an experimental demonstration was conducted to estimate the parameters from the real-cutting images and force data. The practical aspects and limitations of the approach are also discussed.