Hybrid system response model for condition monitoring of bearings under time-varying operating conditions

Condition monitoring (CM) plays a vital role in machine maintenance for ensuring the system's operating reliability and safety as fault detection and health degradation representation can be achieved through it. Nevertheless, Equipment such as wind turbines often operate under time-varying operating conditions (TVOCs), and traditional CM methods are challenged under these circumstances. This paper proposes a novel method for dealing with TVOCs in CM. The proposed method is based on a neural network and a state-space model(SSM), to build a hybrid system response model for describing the operating process of the equipment under TVOCs. Dual extended Kalman filtering is used to solve the dual parameters estimation problem. Finally, the estimated neural network parameters are used as the representation of the health state, and the health indicator (HI) is constructed for real-time monitoring through dimension reduction of the neural network parameters. Experiments on accelerated fatigue degradation of bearings validate the effectiveness and superiority of the proposed method, as an effective HI with TVOCs interference eliminated, compared with both the physical-based and data-driven methods.