Quantification of uncertainty information in remaining useful life estimation

Abstract

Accurate remaining useful life prediction is crucial for prognostics and health management of products. Model uncertainty is an important factor negatively affecting prediction performance. Existing methods fail to evaluate the predictive ability of a degradation model without the actual remaining useful life, and typically require adequate degradation data or prior information. They may be unable to ensure a reliable prediction performance in practical engineering scenarios with limited degradation data and mechanism knowledge. To address these issues, a model fusion based method is proposed using the uncertainty theory. Specifically, a comprehensive model performance evaluation method is proposed by simultaneously considering complexity, fitting ability, and predictive ability. The evolution process of the comprehensive model performance index is modelled by proposing a generalized arithmetic Liu process model that can flexibly depict characteristics of an uncertain process. A model fusion method is proposed by quantifying the similarity between a candidate model and the actual degradation process based on the predictive evolution analysis of the performance index. Then, a random initial solutions based algorithm is proposed to estimate remaining useful life from the fused model. Finally, three real cases and a numerical case are utilized to demonstrate the effectiveness and versatility of the proposed method by comparing it with existing methods.