Remaining useful life prediction of lubrication oil by integrating multi-source knowledge and multi-indicator data

Abstract

Insufficient data restrains the accuracy of the remaining useful life (RUL) prediction for lubricating oil. For this problem, multi-indicator modeling can be an effective solution. However, the unknown coupling effects of the indicators still impose difficulties. To address this issue, a coupling model integrating both knowledge and data is proposed. Originally, multi-source knowledge is embedded to guide both the state characterization and the threshold setting for improving the accuracy of the multi-indicator oil RUL prediction. In particular, the multi-indicator evidences and the knowledge-based rules, are adopted to handle overlapping information and inconsistent decision-making by considering interactions between indicators. A three-layer hierarchical model integrating evidence and rules is firstly established to connect the indicator, attribute, and oil state. Furthermore, the rule-base containing multi-failure modes is constructed to extract the random thresholds from the training samples. Finally, an exponential Wiener stochastic process embedding the oil degradation mechanism is used to describe data fluctuation. After the training stage, the probability density function of RUL is estimated. Finally, the proposed method is validated with four sets of test data from a real-world hydraulic pump bench test.