Uncertainty-Quantified Damage Identification for High-Rate Dynamic Systems

Abstract

This paper proposes an uncertainty-quantified damage identification framework that models a high-rate dynamic system using a Gaussian process regression embedded within a nonlinear autoregressive (GP-NARX) model. The output of the GP-NARX model consists of a set of normal distributions at prediction time points with first-order statistics, facilitating the development of a dynamic threshold for outlier detection. A novel weight-updating approach is introduced which weighs the outliers with real-time frequency changes to evaluate and quantify the probability of multiple damage features as time evolves. The framework is further studied using experimental measurements from shock-loaded electronic packages. The results show that the proposed approach can effectively classify different structure characteristics, including damage-sensitive features.