Advanced dynamic Bayesian network-based real-time prediction of fatigue crack propagation under variable amplitude loading

This paper proposes a crack propagation prediction method under variable amplitude loading based on advanced dynamic Bayesian network (DBN) to address the problem that it is difficult to realize the accurate prediction of individual crack propagation prediction based on the traditional model fitted with historical data. An improved particle filtering algorithm based on MetropolisHastings (M-H) sampling is introduced to overcome the particle impoverishment problem of the traditional DBN algorithm. And the sensitivity analysis of the Wheeler model parameters is used to simplify the variable nodes of the DBN model from six to three, enhances computational efficiency while maintaining accuracy. The advanced DBN model is developed using ABAQUSFRANC3D coupled simulation results for accurate crack propagation prediction under variable amplitude loading. Fatigue crack propagation tests were conducted on three compact tension (CT) specimens subjected to distinct loading sequences. Experimental results demonstrate that the prediction error is 0.30 mm, with a relative prediction error of approximately 1.99 %, confirming the model's effectiveness. The advanced DBN method effectively mitigates uncertainty impacts, enabling robust and high-precision crack growth predictions under variable amplitude loading sequences.