Uncertainty quantification in low cycle fatigue life model based on Bayesian theory

To quantify the uncertainties in the model for low cycle fatigue life prediction, the classic model calibration method is applied using Bayesian theory, and the error term was verified by the normality test. Posterior distribution of the model parameter samples is obtained by Markov Chain-Monte Carlo (MCMC) simulation. An application is presented where a 95% interval of fatigue life prediction well describes the dispersity in real tests with small data samples. Correlation analysis of the samples of parameters is conducted to establish the heteroscedastic regression model. Comparison of the two models shows that the heteroscedastic regression model is questionable in uncertainty quantification performance. Morris global sensitivity analysis method is applied to quantify the sensitivity of the parameters in Manson-Coffin model, indicating that the non-informative prior is reasonable if posterior distribution is sensitive to the prior.