Prediction of fatigue crack propagation driven by data and model collaboration with probability updates

Abstract

During the service life of engineering structures, fatigue crack propagation is inherently influenced by a multitude of uncertainties, encompassing material inhomogeneities, load fluctuations, and environmental corrosion, making accurate prediction a pivotal challenge in safeguarding structural safety and reliability. This paper proposes a data and model collaborative driving method based on probability updates to achieve high-precision prediction and uncertainty quantification of fatigue crack propagation. Construct a framework by integrating fracture mechanism models, finite element simulations, Bayesian theory, and data processing techniques. Select effective parameter groups based on experimental data confidence intervals, combine maximum likelihood estimation and Akaike information criteria to determine the optimal distribution characteristics of parameters, and use a kriging model to efficiently fit complex nonlinear relationships. Leveraging probability update mechanisms and Markov chain Monte Carlo sampling techniques, the dynamic evolution of parameter distributions from prior to posterior is achieved, thereby enabling quantitative characterization of system uncertainties. Experimental verification shows that the model prediction error is less than 4%, and reliability analysis reveals the decreasing trend of reliability with the number of cycles. The results provide theoretical methods for fatigue life assessment and reliability optimization of engineering structures, and have important guiding significance for structural design and maintenance strategy formulation.