Probabilistic fatigue life prediction for adhesively bonded joints via surrogate model

The paper is aimed at developing a probabilistic framework for fatigue life prediction in adhesively bonded joints by calibrating the predictive model, governing adhesive fatigue behavior, using the set of experimental data, and quantifying uncertainty in the model parameters. A cohesive zone model (CZM) is employed to simulate the fatigue damage growth (FDG) along the adhesive bondline and Bayesian inference is used for uncertainty quantification (UQ). The fatigue behavior predicted by FEA modeling for high cycle fatigue, in particular, is computationally intractable, not to mention the inclusion of UQ. To enhance the computational efficiency and yet retain accuracy, a rapid FDG simulator is developed for adhesively bonded joints, by replacing the computationally intensive strain field calculations with the artificial neural networks (ANNs) based surrogate model. The developed rapid FDG simulator is integrated with Bayesian inference and the integrated framework is verified by quantifying uncertainty in fatigue model parameters using the experimental fatigue life data of a single lap joint (SLJ) configuration under constant amplitude fatigue loading. The quantified parameter uncertainties are then used to predict the probabilistic fatigue life in the laminated doublers in bending joint configuration, fabricated using similar adhesive material as SLJ, and successfully comparing it with the experimental data.