Physics-informed DeepONet predicts displacement fields from fracture geometry

The authors propose a physics-informed DeepONet framework designed to predict displacement fields in fractured elastic domains based on boundary conditions and fracture geometry. This approach utilizes a dedicated encoding strategy for the geometry and imposes traction-free conditions weakly through a localized penalty term, eliminating the need for finite-element-generated training data.

• The model serves as a fast surrogate for real-time structural health monitoring.
• It predicts displacement fields directly from boundary conditions and fracture geometry.
• Numerical examples demonstrate feasibility on a representative fracture geometry.

This work lays the groundwork for extending surrogate modeling to diverse fracture geometries, enabling physically consistent and rapid predictions without reliance on traditional simulation data.