AI extracts interpretable constitutive laws directly from solid-mechanics data

Researchers at the Eastern Institute of Technology in Ningbo have created a innovative approach to uncovering fundamental physical laws governing material behavior. This breakthrough graph-based method efficiently analyzes experimental data to uncover concise, accurate constitutive equations without relying on preconceived mathematical structures.

By representing equations as directed graphs, the system can simultaneously optimize both the structure of the equations and the specific material parameters. Experiments with diverse materials like alloy steels, lithium metal and filled rubbers demonstrated that the GraphED framework outperformed traditional empirical models. It successfully discovered explicit equations governing strain-rate dependence and strain-hardening behaviors for steels, while providing superior predictions compared to the widely used Johnson–Cook model.

For lithium metal, the method generated concise plastic-flow constitutive equations with better agreement to experimental data than existing approaches. The framework was also applied to filled rubbers, yielding compact hyperelastic equations that maintained accuracy across varying compositions and temperatures. Dongxiao Zhang, a leading researcher on the project, emphasized that this new approach provides a powerful tool for deriving rigorous mathematical descriptions in cases where conventional models fall short.

The method's ability to uncover interpretable physical laws directly from data has broad potential applications beyond computational mechanics, potentially transforming various fields that rely on understanding complex material behaviors from experimental observations.