Bearing Fault Diagnosis Based on Improved Residual Network

Abstract

In the wind power generation system, the bearing plays a very important role. Whether it can run stably directly determines the quality of the electricity produced and has a great influence on the efficiency of power generation. Due to the harsh working environment, the bearing has become one of the most vulnerable components in the entire wind turbine system. Therefore, bearings of wind turbines need to be maintained regularly. However, it needs to be shut down every time for maintenance, which will incur high maintenance cost. So, the fault diagnosis of the bearing is particularly important. A fault diagnosis method is proposed based on deep learning in this paper. This method is based on the residual module to construct a new ResNet model and embeds the attention mechanism in it to select information that is more critical to the current task goal from a lot of information. In addition, a long short-term memory is added to the network to extract the long-term dependence of the vibration signal and ensure that the information on the time series will not be lost as the training progresses. The experimental results show that the method proposed in this paper is very effective for the fault classification of fan bearings.