Effective data sampling strategies and boundary condition constraints of physics-informed neural networks for identifying material properties in solid mechanics

Abstract Material identification is critical for understanding the relationship between mechanical properties and associated functions. However, material a challenging task, especially when characteristic of highly nonlinear in nature, as common biological tissue. In this work, we identify unknown continuum solid mechanics via physics-informed neural networks (PINNs). To improve accuracy efficiency PINNs, develop efficient strategies to nonuniformly sample observational data. We also investigate different approaches enforce Dirichlet-type boundary conditions (BCs) soft or hard constraints. Finally, apply proposed methods diverse set time-dependent time-independent mechanic examples that span linear elastic hyperelastic space. The estimated parameters achieve relative errors less than 1%. As such, work relevant applications, including optimizing structural integrity developing novel materials.