Architected materials exhibit rich fracture behaviour, governed by microstructural interactions that are often difficult to capture. To overcome this challenge, we present a length-scale-informed, homogenised continuum framework in which nonlocal interactions are captured through a stochastic gradient estimator (SGE). The method substitutes explicit microstructural resolution with a radius of influence, which represents the communication distance between neighbours in a homogenised medium. This nonlocal continuum is embedded in a physics-informed neural network (PINN), which minimises a nonlocal energy functional involving displacement and phase field damage. PINN is adopted due to its computational efficiency compared to finite-element implementations of nonlocal models. Using two benchmark problems-a single-edge notched plate and a notched plate with an off-centre hole-we demonstrate that the present model recovers classic nonlocal fracture behaviours and reveals new geometric effects. In the notched plate, increasing the interaction radius produces the expected nonlocalisation of strain energy. In the plate with a hole, however, the same nonlocal mechanism leads to a geometry-induced shift in crack nucleation at the curved hole boundary, accompanied by a counterintuitive reversal of the expected stiffness reduction. This behaviour arises from curvature-assisted, nonlocal stress redistribution that cannot be produced by a local formulation. Dimensional benchmarking against existing damage models confirms the physical consistency and accuracy of the proposed approach. Overall, the study indicates that a homogenised nonlocal continuum, when coupled with an efficient solver, can capture both standard and nonintuitive fracture phenomena without explicit microstructural modelling.
扫码关注我们
求助内容:
应助结果提醒方式:
