Physics-informed neural network–integrated point cloud decimation and 3D mesh modeling for thermo-mechanical simulation of rail fasteners in digital twin systems

High-density point cloud data of infrastructure components, such as railway fasteners, often contain excessive noise and redundant points, making them challenging to process for simulation and analysis. This study introduces an integrated framework that transforms raw point clouds into simulation-ready 3D meshes and couples them with PhysicsInformed Neural Networks (PINNs) for thermo-mechanical analysis, enabling applications within digital twin environments. The pipeline begins with a RANSAC-based iterative segmentation and outlier removal, achieving an 82.35% reduction in point count while preserving essential geometry. Poisson surface reconstruction and targeted post-processing then produce a high-fidelity, watertight mesh normalized for consistent simulation input. Leveraging this mesh, PINNs solve the steady-state heat equation to model thermal conduction and linear elasticity equations to estimate stress and displacement fields under defined Dirichlet boundary conditions. The resulting temperature and stress distributions are visualized directly on the mesh, providing interpretable, physics-consistent insights into fastener performance. By unifying geometric simplification with data-driven, physics-aware simulation, this approach supports accurate and computationally efficient digital twin development for railway asset monitoring and maintenance.