Data–driven Approach for Extracting Steady–state Data from Unsteady–state Flow Stress without Material Modeling

To accurately calculate the deformation behavior in forming simulations, it is essential to collect steady–state material properties and input them into the simulation software. For instance, in heated sheet metal forming processes, the temperature and strain rate change significantly. Hence, collecting data under isothermal and constant strain-rate conditions is crucial for representing such complex deformation behaviors. However, collecting steady–state data requires appropriate experimental apparatus or specimen geometry and precise control of the experimental environment. An alternative approach is the inverse analytical method, which identifies steady–state data by comparing forming simulation data with experimental measurements. However, this method requires material modeling that accurately represents the unsteady–state of a target. To overcome these challenges, we propose a simple method for directly extracting steady–state data by interpolating unsteady–state data using a machine learning method without material modeling. This paper describes a case study on the extraction of steady–state flow stress from high-temperature tensile experiments on a magnesium alloy sheet (AZ31) using Gaussian process regression. The results demonstrated that the flow stress extracted using the proposed method has predictive accuracy equivalent to that obtained through inverse analysis with a predefined material model that can express the dependency on the temperature and strain rate.