Fatigue life prediction of GFRP laminates using averaged Bayesian predictive distribution and Lamb wave velocity

Abstract

This paper presents a fatigue life prediction approach for GFRP laminates using laser ultrasonic technique for data acquisition and a hierarchical Bayesian model for damage characterization as well as life prediction. Fatigue damage is measured as the degradation of the Lamb wave velocity and described using a scalar damage variable. Due to the complexity of the damage mechanisms involved in the fatigue process and the difficulties in distinguishing different damage modes, i.e. fiber breakage, matrix cracks and delamination, from a single damage variable, multiple degradation models of various levels of complexity are proposed to form a meta-model. Bayesian inference is applied to obtain the distributions of model parameters to include uncertainty information. Then a stacking approach is applied to average the predictive distributions to obtain the optimal combination of each sub-model to generate a meta-prediction with the least diverge from the experimental observation. Experimental results of both traditional stiffness data and Lamb wave velocity data are used for validation. A credible interval (CI) based criterion is also proposed for fatigue life prediction, where consistent results are achieved with reasonable use of informative priors.