Online Robustness Degradation Analysis With Measurement Outlier

Abstract

Online degradation analysis requires an adaptive model parameter estimation. In addition, measurement errors and outliers are inevitable in real applications of degradation analysis. However, existing online models ignore the measurement error or assume the measurement error to be distributed as Gaussian for mathematical simplicity, which is vulnerable to measurement outliers. To deal with such problems, an online degradation analysis technique with robustness to measurement outliers is developed. More specifically, the underlying degradation is modeled with the Wiener process and the measurement error is modeled by constructing a modified Huber density to enhance the robustness against the outlier. For the adaptive estimation of model parameters, an online expectation-maximization (EM) algorithm is developed. Furthermore, procedures are provided for recursive degradation state identification by maximizing a posteriori based on the Laplace approximation. Numerical and two real case studies are carried out to validate the efficacy of the proposed model.