Deep Transfer Learning Based Bearing Fault Detection Method for Valve Cooling System

With the rapid development of industrialization, valve cooling system bearings have become vital in production. However, long-term operation and complex environment lead to faults that threaten the safety of the power system. Traditional fault detection methods are limited by a variety of factors, and their efficiency and accuracy need to be improved. Deep learning techniques provide a new way for valve cooling system bearing fault detection through cross-domain fusion, allowing efficient and accurate detection without complex modelling and expertise. However, existing deep learning methods have defects in network design and dataset, which affect the model performance. A valve cooling system bearing fault detection method is investigated by deep learning model. The proposed model is based on convolutional neural network (CNN), self-attention mechanism and migration learning, and combines Short-Time Fourier Transform (STFT) to extract the time-frequency map of motor vibration signals and improve the Resnet50 model. By using the bearing faults in the CWRU dataset, experiments show that the method achieves 100<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> accuracy, which significantly improves the fault detection efficiency and accuracy. The proposed method can provide a solution for the bearing fault detection for valve cooling system in high-voltage direct current (DC) power transmission.