Fault detection of taper roller bearings using tunable Q-factor wavelet transform and fault classification using long–short-term memory network

Abstract

Abstract Taper roller bearing is a widely used moving component in heavy industrial machinery. Hence, early detection and repair of even minor faults in taper roller bearing is a fault diagnosis and prognosis strategy followed by modern industries. Although many methods for this exist today, the penetration of artificial intelligence and big data analysis into modern industries opens up the possibility of developing better fault diagnosis methods. Such a fault diagnosis and fault classification strategy is going to be discussed in this article. For that, a Tunable Q-factor Wavelet Transform (TQWT) is employed for signal processing, and a Long–Short-Term Memory (LSTM) network is employed for fault classification in this work. It is clear from the experimental findings that the TQWT and LSTM combination can very efficiently and reliably diagnose the faults present in the bearings, and it can classify the types of faults with one hundred percent accuracy. Also, the superiority of the method proposed in this article is confirmed by the fact that it is able to produce better results when compared with the other four combinations of Variational Mode Decomposition (VMD) and Convolutional Neural Network (CNN).