Data-driven inverse identification of mixed-mode fracture for clad interfaces

Accurate characterization of interfacial fracture is essential for assessing the reliability of corrosion-resistant alloy (CRA) clad sheets, yet direct identification of mixed-mode cohesive zone model (CZM) parameters is challenging due to the coupled effects of plasticity and interface failure. In this study, a finite element method–neural network (FEM–NN) inverse identification framework was developed to determine mixed-mode CZM parameters for three industrial clad systems: 316L/A516, 304L/A516, and 825/X60. Plasticity models for the cladding and interface-adjacent base steel were calibrated from tensile tests, and finite element analyses were conducted for T-peel and lap shear tests to generate training datasets with varying CZM parameters. A neural network combining gated recurrent unit (GRU) encoders, physically interpretable curve features, and a multilayer perceptron (MLP) regressor was trained to predict mixed-mode CZM parameters from paired force–displacement curves. The identified CZM parameters accurately reproduced the experimental T-peel and lap shear responses of all three clad systems. Subsequently, a non-conventional circular-lap debonding (CLD) test was newly designed as an independent mixed-mode fracture test for clad sheets. The FE analysis based on the optimized CZM parameters successfully predicted the CLD responses, validating the interface fracture properties under mixed-mode loading. Additional analyses on local normal–shear force histories and CZM parameter sensitivity revealed that the CLD test generates a spatially varying mixed-mode fracture state. The proposed AI-assisted inverse identification framework provides an efficient and physically interpretable approach for mixed-mode interfacial fracture characterization of industrial clad sheets.