Bayesian-Informed Fatigue Life Prediction for Shallow Shell Structures

Abstract This study introduces a Bayesian-informed framework for fatigue life prediction in shallow shell structures. The methodology focuses on inferring the Equivalent Initial Flaw Size Distribution (EIFSD), a critical parameter for structural durability. Bayesian inference, combined with a Co-Kriging surrogate model, enables statistically robust predictions while accounting for uncertainties in material properties, geometry, and loading. The Dual Boundary Element Method (DBEM) is employed for crack propagation due to its efficiency and re-meshing-free modelling. To improve inference efficiency, an iterative parameter space narrowing strategy is proposed. Instead of exhaustively sampling the entire space, the method begins with coarse discretisation to locate high-probability EIFSD regions, then refines them adaptively. A numerical example involving a fuselage window under cabin pressure demonstrates the method. Surrogate models trained on DBEM-generated data significantly reduce computational cost. The proposed strategy achieves high-precision inference, with only 0.059% error in the inferred mean and 5.2% in standard deviation, while reducing CPU time by 52% compared to dense sampling.