A Bayesian Surrogate Constitutive Model to Estimate Failure Probability of Rubber-Like Materials

Abstract

In this study, a stochastic constitutive modeling approach for elastomeric materials is developed to consider uncertainty in material behavior and its prediction. This effort leads to a demonstration of the deterministic approaches error compared to probabilistic approaches in order to calculate the probability of failure. First, the Bayesian linear regression model calibration approach is employed for the Carroll model representing a hyperelastic constitutive model. The developed model is calibrated based on the Maximum Likelihood Estimation (MLE) and Maximum a Priori (MAP) estimation. Next, a Gaussian process (GP) as a non-parametric approach is utilized to estimate the probabilistic behavior of elastomeric materials. In this approach, hyper-parameters of the radial basis kernel in GP are calculated using L-BFGS method. To demonstrate model calibration and uncertainty propagation, these approaches are conducted on two experimental data sets for silicon-based and polyurethane-based adhesives, with four samples from each material. These uncertainties stem from model, measurement, to name but a few. Finally, failure probability calculation analysis is conducted with First Order Reliability Method (FORM) analysis and Crude Monte Carlo (CMC) simulation for these data sets by creating a limit state function based on the stochastic constitutive model at failure stretch. Furthermore, sensitivity analysis is used to show the importance of each parameter in the probability of failure. Results show the performance of the proposed approach not only for uncertainty quantification and model calibration but also for failure probability calculation of hyperelastic materials.