Mixture-of-experts Surrogate Constitutive Models for Viscoplastic Creep Simulation of HT-9 Steel
Please login to view abstract download link
Mechanistic constitutive models that describe the viscoplastic deformation of polycrystals are often computationally expensive, as they involve coupled nonlinear equations that capture detailed microstructural processes. Such models are valuable for their high accuracy, but the associated high cost limits their use in optimization, calibration, and uncertainty quantification workflows that require a large number of model evaluations. To address this challenge, we develop a data-driven surrogate constitutive model for the viscoplastic response of HT-9 steel, a ferritic-martensitic alloy widely used in nuclear energy applications. The model adopts a mixture-of-experts framework that combines multiple local experts through a data-driven gating function. The surrogate is trained on data generated by the high-fidelity viscoplastic self-consistent (VPSC) model. The results show that the mixture-of-experts model achieves good predictive accuracy under creep loading, making it well suited for large-scale uncertainty quantification and design studies.
