Convolutional Takagi-Sugeno-Kang-type Fuzzy Neural Network for Bearing Fault Diagnosis

Rotating machines are widely used in modern industry. In a mechanical system, rolling bearings are essential. Bearings must be able to operate in extreme environments, in which they are prone to various faults. To address the challenge related to accurately classify bearing fault types using vibration sensors, we propose a convolutional Takagi-Sugeno-Kang (TSK)-type fuzzy neural network classifier (CTFNNC) that comprises a convolutional layer and a TSK-type fuzzy neural network. In the CTFNNC, convolutional layers are used to extract the features of a vibration signal, and a TSK-type fuzzy neural network is used to classify bearing faults under various categories. In our experiment, the proposed CTFNNC was compared with other methods, such as a fuzzy neural network, an artificial neural network, and the LeNet-5 convolutional neural network. The experimental results indicate that the proposed CTFNNC has a bearing fault classification accuracy of 98.3% and requires half the number of parameters as LeNet-5.