Machine learning-aided cohesive zone modeling of fatigue delamination

The objective of this paper is to develop a machine learning (ML)-aided cohesive zone model (CZM) for fatigue delamination. A viscodamage model is implemented in a string-based cohesive zone modeling framework to capture load- and path-dependent, pure and mixed mode fatigue delamination. An associated implicit integration scheme is developed and implemented in Abaqus/Standard to generate necessary training data. Subsequently, a series of ML models are trained as surrogates for computationally expensive finite element simulations required during model calibration, and these surrogates are implemented in the Dakota toolkit for parameterized and automated calibration. The present framework is validated by calibrating the present CZM via double cantilever beam and end-notched flexure tests on unidirectional E-glass fiber/E722 composite beams. The ML-aided calibration method is found to significantly enhance calibration efficiency while maintaining accuracy. The present CZM’s capability in handling mixed-mode delamination is demonstrated through simulating a series of mixed-mode bending tests. The present CZM can be implemented in a multiscale modeling framework for fatigue life prediction of composites.