A Novel Lightweight Customized Convolution Neural Network for Bearing Fault Classification

The rolling bearing is an essential component utilized in various types of rotating electrical machines. The implementation of defect detection techniques for bearings plays a crucial role in ensuring the secure operation these kinds of machinery. Research on bearing fault diagnostics using deep learning (DL) has been conducted over the past five years. Nevertheless, the utilization of Convolutional Neural Network (CNN) algorithms is complicated by their large memory requirements and computational expenses, hence constraining their use in low-cost microcontrollers. In this research, a novel approach for intelligent failure classification of rolling bearings using the Short Time Fourier Transform (STFT) and a Lightweight Customized Convolution Neural Network (LC-CNN) is proposed. The spectrogram images are converted from the vibration signals using STFT. Then, spectrogram images are given as a input to the LC-CNN for the bearing fault classification. The proposed model has a minimal memory requirement, making it suitable for implementation in cost-effective hardware devices such as microcontrollers. To validate the proposed model using Case Western Reserve University (CWRU) bearing dataset, resulting in an accuracy of 97.22%. Finally, the proposed model result is compared with GoogleNet.