Probabilistic Fatigue Life Prediction for Welded Joints with Multiple Cracks Based on an Equivalent Initial Flaw Size Distribution

Abstract

This study introduces a methodology that integrates an improved K-T method and a dynamic Bayesian network (DBN) to calculate the EIFSD. The prior EIFSD, derived from the improved K-T method, enables to calculate the EIFS for components with arbitrary crack shapes and provides initial value and initial probability distribution for the state variable. Using the particle filter algorithm in a DBN, deviations from the initial state are corrected through diagnostic and prognostic adjustments, and the prior probability density function is continuously updated with experimental data to reduce prediction errors. This approach is independent of the load history, applicable to components with complex-shaped cracks, and addressing the limitations of backward methods. The generalized EIFSD model was developed to enhance accuracy and reduce stress dependence. The proposed EIFSD methodology was combined with fatigue crack growth analysis of welded joints with multiple cracks, an EIFSD-based probabilistic fatigue prediction method is proposed to investigate the fatigue life distribution of welded joints. The predicted lifetimes are validated against the experimental data for Q420qFNH weathering steel–welded joints, demonstrating the high accuracy of the EIFSD-based prediction method. Results indicate that, under constant stress ranges, the fatigue life of Q420qFNH welded joints follows a log-normal distribution.

Keywords: equivalent initial flaw size distribution, Fatigue life prediction, dynamic Bayesian network, reliability, multiple cracks propagation