Parameter Identification for Constitutive Models via Tree-Structured Parzen Estimator

Finite element method (FEM) is a widely adopted technique for simulating metal foaming processes. Constitutive models, which describe the flow behavior of materials, are critical for FEM accuracy. However, selecting appropriate parameters for these models is challenging, often requiring extensive trial and error. This work explores the use of the Tree-Structured Parzen Estimator (TPE) as an approach to automate the parameter identification process for constitutive models, particularly the Johnson-Cook and Zerilli-Armstrong models. By applying TPE to experimental data from two distinct materials, we demonstrate that TPE significantly improves sample efficiency compared to traditional genetic algorithm methods, reducing the computational requirement to less than 1% while achieving high accuracy of flow stress predictions. This work highlights the use of statistical methods to streamline the development of accurate constitutive models in FEM simulations.