Rapid extraction of CZM parameters for Mode-Ⅱ delamination of plain-woven composites by invertible neural network

Accurately extracting cohesive zone model (CZM) parameters for Mode-II delamination is challenging due to the difficulty of observing crack propagation in end-notched flexure (ENF) tests. This study proposed a data-driven model to rapidly extract CZM parameters from experimental load–displacement curve based on invertible neural network (INN) framework, which supports bidirectional mapping, unifying forward parameter extraction and inverse structural response prediction within a single model. In the forward mode, CZM parameters are extracted directly from load-displacement curve; in the inverse mode, INN acts as a surrogate model for finite element (FE) simulations, enabling accurate prediction of structural response. Comparison with experimental and simulation results confirmed the high accuracy and robustness of the proposed data-driven model.