Machine Learning Inversion of Layer-Wise Plasticity and Interfacial Cohesive Parameters in Multilayer Thin Films

This study proposes a fast material parameter evaluation method for multilayer thin-film structures based on machine learning technology to solve the problems of long time and low efficiency in the traditional material parameter inversion process. Nanoindentation experiments are first conducted to establish an experimental basis across film stacks. A two-dimensional elasto-plastic model of the indentation process is then built to generate a large set of load-depth curves, which serve as training data for a machine learning model. Trained on simulated curves and validated against measurements, the model enables fast inverse identification of layer-wise plastic parameters and interfacial cohesive properties. The experimental results show that the method has high accuracy and efficiency in the inversion of interlayer cohesion parameters, and the correlation coefficient R² is 0.99 or more. Compared with traditional methods, the pipeline supports batch analysis of multiple datasets and delivers parameter estimates within 1 h, substantially shortening turnaround time while improving result reliability. This method can not only effectively solve the challenges faced by traditional material evaluation, but also provide a new and effective tool for the performance evaluation and optimization design of multilayer thin-film materials. It has broad application prospects and potential value.