MoSA Framework Bridges Real-to-Sim Gap in Continuum Dynamics

The newly introduced MoSA (Motion-constrained Stress Adaptation) framework tackles the gap between real-world and simulated learning of continuum dynamics through visual cues. This technique focuses on addressing residual anisotropy and heterogeneity found in actual objects following the calibration of a nearly isotropic backbone. By employing an isotropic model as a foundational physics principle, MoSA learns residual stress operators to account for deviations, thereby enhancing accuracy while retaining physical priors. This method seeks to address the shortcomings of conventional calibration methods that rely on assumptions of homogeneity and isotropy, as well as the inefficiencies and overfitting issues associated with black-box neural networks. The research can be accessed on arXiv under ID 2605.22597.