Data-driven framework for extracting steady-state deformation resistance from transient experimental data without inverse finite element analysis

Accurate characterization of the steady-state deformation resistance of materials is a prerequisite for high-precision forming simulations. However, controlling the temperature and strain rate in high-temperature tensile experiments is difficult and can hinder data acquisition under ideal isothermal and constant strain-rate conditions. Although the inverse finite element method (iFEM) has been conventionally employed for this purpose, it suffers from limitations such as dependence on constitutive equations, non-uniqueness of solutions, and requirement for expert-level implementation. This study introduces a data-driven framework for extracting steady-state deformation resistance from transient experimental data without using iFEM. The proposed framework acquires transient experimental data, performs regression-based interpolation and optimal model selection, and extracts steady-state responses using regression techniques, which includes a pre-defined material model, artificial neural network, and Gaussian process regression. Two case studies with aluminum and magnesium alloys, each under distinct variable conditions, are conducted to assess validity and scalability. ANN and GPR enhance the prediction accuracy of interpolated values compared to that of the pre-defined material model. This finding validates the feasibility of accurately estimating steady-state deformation resistance through flexible model selection. The plausibility of the predicted responses is supported by visualizing the corresponding response surfaces. The proposed framework provides a flexible, scalable, and practical alternative to conventional iFEM-based approaches with promising future applicability to higher-dimensional scenarios. • Extracted steady-state data from transient experiments without inverse FEM • Selected framework for suitable regression models for data interpolation • Validated through case studies on aluminum and magnesium alloy experiments • Machine learning models demonstrated flexibility and scalability