Abstract

Development of practical bearing prognostics plays a critical role to improve the reliability and safety of aero-engines. Bearings being suddenly failure without alert will be catastrophic. This paper proposes a robust method for bearing condition evaluation. It first acquires the time domain vibration signals of bearings and selects representative indices to construct the feature vector. These feature vectors coupled with their corresponding bearing condition stages will be taken as input and output of a RBF network. Once the RBF network is properly trained, it can be used for monitoring bearing condition. After that, new inputs and outputs of RBF network can constitute a case set, from which a lattice structure based extraction method is introduced for rule generation. The rule set can help us reveal the quantitative relationship between bearing vibration signals and bearing condition stages. Experimental results demonstrate that the RBF network model can effectively evaluate bearing condition with high generalization and adaptability. Furthermore, the rule extraciton method can accurately identify the degradation patterns of bearings.

Paper Details