Research on Multi-condition Bearing Fault Diagnosis Based on Improved Semi-supervised Deep Belief Network

With the development of industrial automation, the processes in the production of rotating systems are becoming more and more sophisticated and complex. When multiple processes are executed, the internal bearing speed and load are constantly changing. For this type of multi-working conditions, the original fault diagnosis method trained in a single working condition is no longer applicable. Therefore, a bearing fault diagnosis method for bearings under multiple operating conditions is proposed. This method is based on a semi-supervised deep belief network (Semi-supervised deep belief network, SSDBN). Using a small amount of labelled data to conduct the fault classification and judgment, and it can improve the accuracy of fault diagnosis under varying working conditions. First, set the source and target domains to the same load, different speed and damage size data set, and process the source and target domain signals by wavelet packet decomposition. Then the maximum mean discrepancy algorithm ((Maximum mean discrepancy, MMD) is used as the evaluation indicator of the difference between the distribution of data features in the source and target domains, and the smaller distribution difference is selected. Feature sample data, make the model easier to compare features, improve the classification accuracy of the target domain data. Use the improved semi-supervised deep belief network to train less labelled data and a large number of unlabelled data, improve the classification accuracy of the data to be tested. Take the bearing data set of Western Reserve University as an example to verify the diagnostic accuracy of the model under the same load at different speeds and damage size conditions, as well as the diagnostic accuracy of the model under different load, speed and damage size conditions. The results show that the method can improve the accuracy of the fault diagnosis of the bearing under the constraints of multiple operating conditions, reduce the false alarm rate of the fault diagnosis, and reduce the gradient disappearance during the training model, and increase the probability of successful failure classification. In addition, the proposed method does not need to be considered the sensitivity of various features does not need to rely on expert knowledge and experience. It has strong universality and compatibility, which is helpful to find and replace damaged bearings in time and ensure the safe and reliable operation of machinery and equipment. © 2021 Journal of Mechanical Engineering.