A belief function theory based approach to combining different representation of uncertainty in prognostics

Abstract

In this work, we consider two prognostic approaches for the prediction of the remaining useful life (RUL) of degrading equipment. The first approach is based on Gaussian Process Regression (GPR) and provides the probability distribution of the equipment RUL; the second approach adopts a Similarity-Based Regression (SBR) method for the RUL prediction and belief function theory for modeling the uncertainty on the prediction. The performance of the two approaches is comparable and we propose a method for combining their outcomes in an ensemble. The least commitment principle is adopted to transform the RUL probability density function supplied by the GPR method into a belief density function. Then, the Dempster's rule is used to aggregate the belief assignments provided by the GPR and the SBR approaches. The ensemble method is applied to the problem of predicting the RUL of filters used to clean the sea water entering the condenser of the boiling water reactor (BWR) in a Swedish nuclear power plant. The results by the ensemble method are shown to be more satisfactory than that provided by the individual GPR and SBR approaches from the point of view of the representation of the uncertainty in the RUL prediction.

Remaining useful life estimation in heterogeneous fleets working under variable operating conditions

“…This can be very useful, since it allows identifying the adjuvant 5 operating condition that may enhance the degradation process towards failure and, thus, scheduling proactively the proper maintenance interventions [43] 2) quantifying the uncertainty affecting the RUL predictions of the equipment, due to the variable operating conditions experienced by the fleet equipment, whose information is used for the HDTFSSMM parameters identification. This uncertainty assessment, which describes the expected mismatch between the real and predicted equipment failure times, can be used by the maintenance decision maker to plan maintenance interventions with the required confidence [43];…”

“…In fact, the large width of the prediction interval is due to the large variability of the fleet transitions and of the temperature conditions experienced by the capacitors during the early stages of their lives. With respect to the capability of the fuzzy similarity-based method to estimate a confidence interval, it is worth mentioning that in [43] the considered fuzzy similaritybased approach has been combined with a Belief Function Theory (also called Dempster-Shafer or evidence 19 theory [80]) for providing RUL predictions with the associated uncertainty. Other similarity-based approaches for prognostics have been developed and their capabilities of quantifying the uncertainty affecting the RUL predictions have been demonstrated [30].…”

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

“…In industries such as nuclear, oil and gas, chemical and transportation, unforeseen equipment failures are extremely costly in terms of repair costs, lost revenues, environmental hazards and human fatalities [1]. To anticipate failures and mitigate their consequences, predictive maintenance approaches are being developed, based on the assessment of the actual equipment degradation condition and on the prediction of its evolution for setting the optimal time for maintenance [1]- [4].…”

“…To anticipate failures and mitigate their consequences, predictive maintenance approaches are being developed, based on the assessment of the actual equipment degradation condition and on the prediction of its evolution for setting the optimal time for maintenance [1]- [4].…”

Prediction of industrial equipment Remaining Useful Life by fuzzy similarity and belief function theory

“…The performance of the proposed method is verified with respect to three performance indicators (i.e., Mean Square Error ( ) for estimating the accuracy of the predictions, Coverage ( ) for the reliability of the prediction intervals and Mean Amplitude ( ) for their precision (Baraldi et al, 2015). For comparison, the Kernel Density Estimation (KDE) (Botev et al, 2010) and the Mean-Variance Estimation (MVE) (Nix & Weigend, 1994) methods which have already been successfully applied for estimating predictions uncertainty in different prognostic applications on industrial components such as turbofan engines (Wang, 2010) and turbine blades (Baraldi et al, 2012a), are applied to the same case studies and their results are compared to those obtained by the proposed method.…”