A Fully Connected Neural Network Enhanced Peridynamic Framework for Efficient Simulation of Hydraulic Fracturing in Fractured Rocks

ABSTRACT Hydraulic fracturing involves strong coupling among fluid flow, rock deformation and fracture evolution. Peridynamics can naturally represent crack initiation and propagation without explicit crack tracking, but repeated nonlocal seepage calculations make hydro‐mechanical simulations computationally demanding. This study develops a fully connected neural network‐enhanced peridynamic (FCNN–PD) framework in which a data‐driven surrogate predicts pore‐pressure fields and is coupled with the peridynamic mechanical and damage calculations. The network learns the relationship between nodal states, hydraulic loading, fracture geometry and pressure response. A case‐balanced sampling strategy is used, and several candidate input representations are compared to identify the information required for accurate prediction. The selected input representation includes coordinates, permeability, hydraulic conditions, fracture descriptors and spatial‐distance features. Independent tests yield a mean normalised root‐mean‐square error of 0.804% and a mean coefficient of determination of 0.99247, demonstrating high fidelity to the peridynamic reference solution. Sensitivity and generalisation analyses indicate stable performance within the investigated parameter range. Comparisons involving matrix flow, fracture flow, crack deformation and hydraulic fracturing further show that the coupled framework reproduces pressure, displacement and fracture responses with acceptable accuracy, while the predicted critical pressure differs from the experimental value by 8.27%. The proposed approach reduces the computational burden of peridynamic hydro‐mechanical analysis while preserving its capability to model fracture evolution, providing an efficient route for simulating fractured porous media. Further work should address complex fracture networks, broader loading histories and explicit conservation constraints.