Optimized SWPT and Decision Tree for Incipient Bearing Fault Diagnosis

As long as technological capabilities increase, the requirement of an integrated machinery health management (MHM) becomes an essential task in the operational life cycle of the industrial machines. The MHM can be investigated for anomaly detection, fault diagnosis and fault prediction, using up to-date data. The induction motor (IM) represents a critical part in the industrial processes. The overwhelming majority of IM faults are bearing faults. Therefore, the diagnosis of bearing faults has received high attentions, in order to minimize maintenance cost and unscheduled breakdowns. Since the bearing passes through different stages from healthy state to failure state, modern MHM should provide means of detecting the bearing fault, as well as identifying its severity. This paper uses a novel wavelet-based technique called Optimized Stationary Wavelet Packet Transform (Op-SWPT) in order to detect the IM bearing fault at an early stage and to provide identification of its severity. The usefulness of the proposed approach is tested on the well-known database of Case Western Reserve University (CWRU) for different levels of severity of the bearing fault. The results issued by the experiments verify the effectiveness of the proposed approach.