2026/07/07 by Aditya Konale, Vikas Srivastava
Computer Science · Engineering · #Contact Mechanics and Variational Inequalities #Adhesion, Friction, and Surface Interactions #3D Shape Modeling and Analysis
paper · doi:10.1115/1.4072262
Abstract Elastomers are highly deformable, important engineering materials with applications ranging from sealing to wearable technology, where they undergo large deformation contact with rigid surfaces. The finite element method (FEM) is the modeling standard for these interactions. There is significant interest in developing and applying new neural network-based methods for solving mechanics problems. The application of physics-informed neural networks (PINNs) to contact is limited to small deformations. There is only one PINN developed for large deformation contact modeling. However, this energy-based PINN uses the direct formulation, a relatively complex surface contact potential, and requires numerical integration for the energy evaluation. The mixed formulation and automatic differentiation (AD) remain underutilized, which can enable efficient and practical PINNs for large deformation contact problems. We present a mixed formulation PINN with the gap function approach for modeling large deformation frictionless contact between elastomers and rigid bodies. In this work, several methods leveraging AD to evaluate deformed tangential vectors are proposed and assessed, which is nontrivial for large deformation contacts. We show the method and its successful applicability to various contact surface configurations, including flat-flat, convex-flat, flat-concave, and concave-convex, by comparing them with baseline solutions from FEM. The F (Deformation gradient) tangent approach was found to be the most robust performer for complex surface contacts. The proposed modeling framework and neural network architecture should motivate PINNs to model large deformation contact in a variety of materials. The codes are available at https://github.com/SrivastavaResearchLab/ContactPINN2026.