An approach for predicting fatigue crack growth in polymers based on high-fidelity multidimensional feature mapping and physical constraints

Reliable residual life assessment of safety–critical components relies on accurate prediction of fatigue crack growth (FCG), which directly controls damage tolerance and failure risk. While machine learning offers a powerful data-driven approach for FCG prediction, its effectiveness is often limited by data scarcity, physical model limitations, and the black-box nature of neural networks. To address these challenges, this study proposes a failure-mechanism-related FCG prediction strategy driven by multi-dimensional fusion, integrating in-situ experiments, multi-scale characterization, high-fidelity simulation, data augmentation, and physics-informed constraints. The results show that the plastic zone at the crack tip expands significantly with increasing stress ratio and crack length, and the features of multi-level steps and curled ligaments on the microscopic fracture surface confirm interlaminar separation and intense viscoplastic dissipation mechanisms. Given these failure characteristics, a high-fidelity finite element framework was constructed by coupling an anisotropic visco-elasto-plastic constitutive model with a cyclic cohesive zone model. After validating the model’s effectiveness at two scales, microscopic local strain fields and macroscopic FCG rates, a seven-dimensional microscopic feature space containing local physical information was established, achieving a dimensionality leap and data enhancement. Subsequently, the FCG mechanics equation was explicitly embedded into the neural network loss function as a physical constraint to guide the model optimization direction. The resulting hybrid architecture integrates data, simulation, and physics, achieving an order–of–magnitude reduction in computational cost while maintaining high predictive fidelity. This work establishes a mechanistically grounded framework for efficient and reliable FCG prediction.