هانيه شيرين

Fatigue-induced crack propagation represents a primary threat to the structural integrity an‎d operational longevity of bridges subjected to cyclic loading. While the Paris Law remains a cornerstone of classical fracture mechanics for predicting fatigue crack growth rates, its integration into large-scale Finite Element (FE) simulations is often hindered by the prohibitive computational expense of iterative parameter extraction. Moreover, the predictive performance of data-driven surrogates is heavily contingent upon the quality an‎d representativeness of the underlying datasets. The core contribution of this study is the development of a robust data-driven framework designed to generate high-fidelity datasets from numerical simulations, subsequently replacing the conventional Paris Law with a machine learning-based surrogate model. Accordingly, a representative interior steel bridge girder was developed in Abaqus under dynamic fatigue loads, with crack geometries of varying initial lengths automated via Python scripting. To enhance data efficiency, clustering techniques were employed to filter redundant information an‎d emphasize critical crack growth regimes, resulting in an optimized dataset for training. A Multi-Layer Perceptron (MLP) neural network was then implemented to predict crack growth rates as a computationally efficient proxy for the Paris Law. Performance metrics on unseen datasets indicate that the proposed methodology significantly enhances computational throughput an‎d reduces reliance on complex fracture mechanics derivations without compromising predictive precision.

خستگي , مكانيك شكست , يادگيري ماشين , قانون پاريس , پل

Fatigue , Fracture Mechanics , Machine Learning , Bridge , Paris Law

Hanieh Shirin

Vahid Broujerdian