Research on Fault Diagnosis Method for Fan Rolling Bearing Based on Improved ResNet50

• Published in: 2024 2nd International Conference on Signal Processing and Intelligent Computing (SPIC) pp. 779 - 783
• Main Authors: Guan, Qixue, Zhang, Yongxin, Yang, Wei, Li, Xiyu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 20-09-2024
• Subjects: Accuracy; continuous wavelet transform; Continuous wavelet transforms; Convolution; Fault diagnosis; Neural networks; Residual Neural Network; Residual neural networks; Rolling bearings; squeeze-and-excitation networks; Vibrations; Wavelet domain; Wind turbines
• DOI: 10.1109/SPIC62469.2024.10691549

This paper presents a fault diagnosis technique for wind turbine rolling bearings, utilizing Continuous Wavelet Transform (CWT) and an Enhanced Deep Residual Network (ResNet). This method addresses the prevalent issues of low accuracy and slow speed in diagnosing faults in wind turbine rolling bearings. Initially, continuous wavelet transform is employed to generate wavelet images with time-frequency domain features, ensuring the preservation of temporal characteristics of one-dimensional vibration signals during convolutional processing. Subsequently, the Squeeze and Excitation Networks (SENet) structure, which is based on an attention mechanism, is incorporated into the residual neural network. This enhances the ResNet50 neural network, allowing it to train the wavelet image input model more effectively. Finally, the trained model is utilized for classifying bearing faults. The experimental results indicate that the proposed technique achieves superior accuracy and speed compared to conventional methods.