Physics-Informed Neural Networks (PINNs) offer a potent solution for inverse problems in mechanics, yet standard single-network formulations fundamentally fail to resolve the discontinuous fields inherent to fracture mechanics due to the breakdown of governing equations across geometric discontinuities. This study introduces a novel, meshless Multi-Region PINN framework that generalizes the solution of two-dimensional cracked elastic bodies. By decomposing the domain into continuous sub-regions stitched via strict traction and kinematic continuity constraints, the method restores the validity of the Navier–Cauchy equations within each partition. The framework is first validated against high-fidelity Finite Element Method (FEM) simulations, where it is shown to accurately capture high-gradient stress singularities using only sparse displacement data from the external boundaries. Quantitative comparisons demonstrate maximum relative errors below 1 percent for displacement fields and approximately 2 to 4 percent for strain and stress components. In this controlled environment, we further identify a systematic drift in the J-integral calculation inherent to soft-constraint formulations and propose a thermodynamic error-compensation strategy anchored to the far-field boundaries, yielding path-independent fracture energy estimates. To demonstrate practical utility, the method is subsequently applied to experimental Digital Image Correlation (DIC) data obtained from a Glass Fiber Reinforced Polymer (GFRP) Double Cantilever Beam (DCB) specimen subjected to Mode I loading. In this regime, the PINN acts as a rigorous physics-informed reconstructor, effectively recovering smooth, singular stress fields from noisy experimental data. To handle the inherent stochasticity of the experimental boundary data, the framework is adapted with a robust Huber loss formulation and a trainable masking mechanism that suppresses local outliers. The method’s reliability is confirmed by the close agreement between the PINN-calculated fracture energy (GI) and the analytical reference derived via Modified Beam Theory (MBT), establishing a robust methodology for hybrid fracture characterization that paves the way for the data-driven discovery of novel cohesive and bridging laws.