A switching ensemble approach for remaining useful life estimation of electrolytic capacitors

Abstract

We consider the problem of predicting an equipment Remaining Useful Life (RUL), based on data from a fleet of similar components working under different operating conditions. Two data-driven prognostics models are considered, an Homogeneous Discrete-Time Finite-State Semi-Markov Model (HDTFSSMM) and a Fuzzy Similarity-Based (FSB) model. The predictions from the two models are ensembled with a switching approach that selects the most accurate model at each stage of the equipment life to guarantee the best performance throughout the degradation progression. A real case study is worked out, regarding a fleet of aluminum electrolytic capacitors used in electric vehicles powertrains. In this specific case, the HDTFSSMM is selected at the early stages of the equipment degradation and the FSB model at the last stages of life. Results are compared with those achieved by each individual model, showing that the proposed ensemble approach is able to provide more accurate RUL predictions throughout the entire equipment life.

A locally adaptive ensemble approach for data-driven prognostics of heterogeneous fleets

“…The adaptive switching ensemble model 45 (sketched in Figure 8) entails first an offline selection of the optimal switching time t opt among all the possible switching times t sw = [ t sw min , t sw max ] , where t sw min . is the first measurement time and t sw max is the longest end of life that minimizes the accuracy index (AI) over the P valid validation trajectories, that is, the relative error of the RUL prediction.…”

“…The alternative approach is an adaptive switching ensemble approach for data-driven prognostics that selects the HDTFSSMM at early stages of life and the FSB model at the last stages of life. 34,45…”

“…The approach is structured in two phases [45]: an offline selection of the optimal switching time before which the HDTFSSMM is used for providing the RUL estimates and after which the FSB is used (the interested reader may refer to Appendix A for further details on the procedure) and an online phase that relies on to switch between the HDTFSSMM and the FSB for predicting the RUL of the = 25 capacitors.…”

“…For further comparison, an alternative ensemble approach is applied to the case study and its results are compared to those obtained by the individual models and the proposed ensemble approach. The alternative approach is an adaptive switching ensemble approach for data-driven prognostics that selects the HDTFSSMM at early stages of life and the FSB model at the last stages of life [34], [45].…”

“…The adaptive switching ensemble model [45] (sketched in Figure 8 Then, an online usage for predicting the RUL of pieces of equipment. In other words, the optimal switching time represents the time up to which HDTFSSMM is used for providing the RUL estimates at the early stage of the equipment life and beyond which FSB is used when the equipment approaches the end-of-life.…”