Resolving physical complexities with machine intelligence

Understanding the relationships between material microstructures and their mechanical performance and using them to make predictions are pivotal topics in solid mechanics. From Galileo’s beam bending analysis, Cauchy’s stress definition to Arrhenius-based creep laws, theoretical and simulation frameworks find great success in addressing engineering problems. Yet, the spatiotemporal complexity challenges the conventional ‘observation-hypothesis-model’ approach for structural integrity in key industrial sectors such as aerospace, nuclear energy, and semiconductors. Recent progress and fusion of high-performance computing, high-throughput experiments, data science, and artificial intelligence provide a complementary solution to scientific discovery and engineering deployment on these issues. However, unlike their applications in vision and language domains, engineering science demands stronger data-model inference capabilities. High-quality, physically consistent databases and digital libraries are needed to enhance model performance, generalization, and interpretability. Concepts such as “physics transfer” and “reality reconstruction” offer guiding principles for modeling and predicting complex behaviors. With further support from cognitive science, intelligent agents and physical intelligence are increasingly capable of assisting, or even replacing, researchers in conducting exploration and reasoning in complex, dynamic scenarios. This paper reviews key insights of the complexities in solid mechanics and discusses active research areas through the lenses of learning theory and open science, with particular emphasis on multiscale mechanics and the long-term service behavior of materials and structures.