An Integrated Data and Knowledge Model Addressing Aleatory and Epistemic Uncertainty for Oil Condition Monitoring

Abstract

Reliable operation of machinery is very desirable in engineering. To achieve this objective, the assessment of the lubrication oil state is necessary. However, due to the unpredictable variations, uncertainty detection and handling in the oil state has been a bottleneck in practice. A solution strategy is proposed in this paper that integrates information from the monitoring data and expert knowledge. On the other hand, since insufficient data and limited knowledge, two types of uncertainty are present, namely, aleatory and epistemic uncertainty. To handle these uncertainties, an integrated model with a three-layer structure is constructed that incorporates both expert knowledge and data. First, for the detection of stochastic data variation, the initial connection among the layers is assigned by membership probabilities as the characterization evidence. Second, the oil state that produces a unified output with various pieces of evidence is determined by evidential reasoning with knowledge-based rules. Third, to provide consistent monitoring adaptively, a knowledge-integrated neural network is established for determining the initial parameters from measurements. The effectiveness of the proposed model is demonstrated using both simulated and real-world data from industrial vehicles.