Ensemble of model-based and data-driven prognostic approaches for reliability prediction

Accurate prediction of system reliability is of plenty of importance to engineering systems for accomplishing the designate function and system safety management. As the concerned system is getting complicated and more sufficient health monitoring measurement is available, the traditional reliability prediction schemes resorting to only one kind of prediction approaches, model-based or data-driven, begin to show their limitations. This paper proposes an ensemble prognostic method by combining traditional model-based with data-driven approaches. Initially, two member algorithms for state-of-health (SOH) prediction are weighted-sum by utilizing the historical prediction error. Then, to compare the performance of our ensemble prediction with the model-based, data-driven, kalman filter and hybrid approaches, experiments based on electrolytic capacitor capacitance degradation and lithium-ion battery capacity degradation are performed. Finally, by employing the available data as well as dynamically adjusting weights, smaller prediction error and standard deviation are derived. Based on the aforementioned analysis, the better prediction performance of the proposed method has been verified.