A distribution-free stochastic physics-guided reliability analysis under polymorphic uncertainty

The paper proposes a stochastic model updating framework which possesses two key features: (1) the entropy-based Jensen–Shannon divergence as the distance function for the Approximate Bayesian Computation procedure, owing to its strengths in – (a) effectively capturing the discrepancy in both the relative mean and the variance information between two distributions, and (b) returning bounded finite values which avoids the issue of computational instability; and (2) the implementation of the Transitional Ensemble Markov Chain Monte Carlo to obtain posterior samples under affine-invariance. To the best of knowledge, the Jensen–Shannon divergence implementation for Approximate Bayesian Computation is under studied, providing an opportunity to study its robustness towards performing model calibration under varying data size. In addition, the challenge of performing a stochastic reliability analysis involving a system of coupled-equations is not widely investigated within the existing literature which further motivates such study. The proposed framework is validated through the 2008 SANDIA thermal problem involving a reactor slab material, whose thermal property reliability under a specific temperature condition is assessed. The robustness of the proposed framework is evaluated and verified against published results.