Machine learning-assisted process mapping and constitutive modeling of hot deformation in lightweight steel

Accurate modeling of hot flow behavior in advanced lightweight steels is critical for optimizing thermomechanical processes, yet remains challenging due to the complexity of deformation mechanisms across wide temperature and strain rate ranges. In this study, the hot deformation behavior of Fe-11Mn-6Al-0.1C steel was investigated using both a conventional Arrhenius-type constitutive model and a data-driven artificial neural network (ANN) approach. Phenomenological models such as the Arrhenius equation rely on temperature- and strain rate–sensitive parameters like activation energy and stress exponent, but their predictive accuracy declines significantly when deformation mechanisms vary, a common feature in lightweight steels. To overcome this limitation, a feed-forward ANN model with a single hidden layer of 30 neurons was trained on experimental hot compression data obtained in the temperature range of 100–1100 °C and strain rates of 0.001–0.1s −1 . The ANN model demonstrated superior predictive performance compared over the Arrhenius model, as validated by statistical metrics (R, RMSE, and AARE). Processing maps constructed via the Dynamic Materials Model (DMM) further revealed distinct instability domains and optimal hot-working conditions. This study highlights the advantages of machine learning-based constitutive modeling over traditional approaches, across wide thermomechanical domains particularly in materials exhibiting complex and temperature-sensitive deformation mechanisms. The results, for the first time in Fe–11Mn–6Al–0.1C lightweight steel, demonstrates that artificial neural networks can not only achieve superior predictive accuracy but also provide more reliable processing maps for guiding industrial hot working.