Remaining Useful Life Estimation of Industrial Circuit Breakers by Data-Driven Prognostic Algorithms Based on Statistical Similarity and Copula Correlation

Michy Alice, Dejan Pejovski, Loredana Cristaldi

Abstract

Predicting the future behaviour of an item or a complex system based on its past history is the aim of data-driven algorithms. In our paper, we present two algorithms for predicting the Remaining Useful Life (RUL) of industrial circuit breakers (CB) which make use of on-site collected data related to CB’s health condition. In the first algorithm, a sub- fleet of CBs is identified by applying the two-sample Kolmogorov-Smirnov Test which relies on statistical similarity between the observations. Once chosen the sub-fleet, the algorithm attempts to exploit correlations between the variation of health condition and sampling time using copulas. The second algorithm models the correlation structure between the time at which a certain degradation level occurs and the item’s End of Life (EOL). Both algorithms are used to estimate the item’s Remaining Useful Life through the Monte Carlo method. The use of copulas attempts to exploit also the information on the correlation structure in the data in order to obtain a higher accuracy in the estimation.

Keywords

Remaining Useful Life, Copula correlation, Data-driven prognostics, Health condition variation

Download

IMEKO-TC10-2019-021.pdf

DOI

-

IMEKO TC

TC10 - Measurement for Diagnostics, Optimization and Control

Back to the proceedings