Calibration and validation of fatigue lifetime model in complex structures based on multi-level data

Abstract

A method of model calibration and validation is established for structures with complicated structural features in the presence of multi-level experiments and censored data. Fatigue lifetime models are established comprised of three levels, where Bayes network was introduced to connect model predictions and experimental data. To avoid iterative calculation, calibration models in component-like and component levels are reconstructed explicitly. Normal distributions are applied to approximate the observed censored data, and the scatter area metric is proposed for model validation. A simulation example and the reliability analysis of the bolt hole in the turbine disc are applied to verify the proposed method.