About This Architecture

Physics-informed neural network (PINN) coupled with isogeometric analysis (IGA) for phase-field fracture simulation, integrating NURBS geometry, random Fourier features, and MLP backbone to predict displacement and crack evolution. The architecture normalizes control-point coordinates, projects them through learned random Fourier embeddings, and feeds enriched features into a fully-connected MLP that outputs trial fields for displacement and phase-field variables. IGA evaluation reconstructs fields at Gauss points, computes kinematic derivatives, and evaluates elastic and fracture energies with material gradation and irreversibility constraints. This hybrid approach combines the geometric precision of NURBS with neural network expressivity, enabling efficient surrogate modeling of complex fracture behavior in heterogeneous materials while respecting physical conservation laws. Fork this diagram on Diagrams.so to customize material parameters, optimizer sequences, or penalty term weights for your specific fracture mechanics application.

People also ask

How do you combine physics-informed neural networks with isogeometric analysis for fracture mechanics simulation?

This PINN-IGA architecture normalizes NURBS control-point coordinates, encodes them via random Fourier projections and MLP backbone, then reconstructs displacement and phase-field trial fields at Gauss points using IGA basis functions. Energy-consistent loss combining elastic, fracture, and penalty terms drives optimization through Adam→LBFGS→RPROP sequencing, with adaptive load-stepping ensuring