Automating plastic bearing fault diagnosis using continuous wavelet transform and neural networks

Abstract

Faulty plastic bearing is an initial alarm for bearing failure that can cause massive losses in the production line. The losses include restarting of production, producing of defective products, and even human casualty. This paper therefore aims to propose an automated plastic bearing fault detection and classification system. The system begins by transforming the bearing vibration signals into coefficients with continuous wavelet transform. The coefficients are then filtered by coefficients reduction and smoothing thereafter. Then, the filtered coefficients are classified by two ANN classifiers i.e. feed forward backpropagation (FFB) and recurrent neural network (RNN). The performance of both classifiers are finally measured and compared. The best overall performance is 90% detection rate by FFB. This system prevents bearing failure by giving an early alarm for faults detection and making the corrective action easier. Also, the single stage data processing and single signal type increase the data processing efficiency.

Keywords

• bearing fault diagnosis
• continuous wavelet transform
• exponential moving average
• neural network
• principle component analysis