基于表面应变数据与物理信息神经网络的PBX内部残余应力场重构OA

Accurate characterization of the internal residual stress field in polymer-bonded explosives(PBX)is crucial for evaluat-ing structural integrity and service safety.However,traditional inversion methods struggle to reconstruct this distribution efficient-ly and with high precision.A novel residual stress inversion method integrating distributed optical fiber sensing with physics-informed neural networks(PINNs)was proposed.TATB-based PBX cylindrical specimens measuring ϕ110×120 mm served as the research objects.Strain release data during the layer-removal test were acquired in real time using optical frequency-domain reflectometry(OFDR)technology.A PINNs inversion model was constructed using Gaussian temperature-field parameters as intermediate variables.Physical constraints,including static equilibrium equations and boundary conditions,were embedded into the loss function.This approach achieved an efficient mapping from limited surface strain data to the internal stress field.Results indicate that the model achieves a coefficient of determination of 0.954.The inverted results strictly satisfy fundamental solid mechanics constraints.Compared with traditional finite element optimization-based inversion methods,the PINNs model requires only about 5 min for a single inference after offline training.The reconstructed internal stress field exhibits a typical"tension inside,compression outside"distribution pattern.The maximum tensile stress reaches 1.03 MPa at the core,while the maximum compressive stress reaches-0.77 MPa at the middle of the lateral surface.This distribution pat-tern aligns perfectly with the physical mechanism of residual stress generation caused by non-uniform shrinkage during the cool-ing process of thermal pressing.