Research On Bearing Fault Diagnosis Based On Transfer Learning Of Deep Convolution Neural Network

At present,people’s requirements for the safety and reliability of mechanical equipment are increasing,and the health status of rolling bearings,as one of the most common and easily damaged components in rotating mechanical equipment,depends on the operation status of the whole rotating machinery.Most of the existing intelligent fault diagnosis methods require the same distribution of training data and test data,but it is difficult to collect a large number of effective label samples in actual production work,which leads to the performance of the theoretical methods being greatly affected in practical applications.This paper introduces migration learning theory,combined with deep convolutional neural network,to study the intelligent fault diagnosis method of rolling bearings across working conditions and equipment.Firstly,to address the problem of insufficient labeled samples and difficulties in training the recognition network,the paper introduces a model migration method and investigates a method based on model migration to achieve rolling bearing fault diagnosis for different working conditions of the same machinery and different mechanical devices.This method combines residual network and dense connection network for feature extraction,and uses a dataset containing a large amount of labeled data(called the source domain dataset)to train the diagnostic model,and then migrates the model to the target domain with insufficient labeled samples,and after fine-tuning,applies it to the fault diagnosis scenario in the target domain,so as to realize the fault diagnosis for different working conditions of the same machine and between The fault diagnosis between different working conditions of the same equipment and different mechanical equipment in the target domain is realized by the source domain diagnosis knowledge.Secondly,considering that there are few studies on faulty bearings in actual working scenarios and the utilization of existing datasets is low,the paper investigates a multi-source domain-assisted cross-machine rolling bearing fault diagnosis method.On the basis of feature migration,a multi-source domain migration network model is constructed to realize the diagnosis of bearing faults in actual working scenarios in the target domain by learning the diagnostic knowledge from multiple artificial source domain datasets.The method first preprocesses the vibration signal by means of time-frequency transformation,and then trains the classifier for the target domain using the source domain dataset through feature migration to finally achieve fault diagnosis of the target bearing.Several different datasets are used in the paper to verify the effectiveness of the method.The fault diagnosis method studied in this paper is validated on several publicly available datasets,and the influence of important parameters in the method on the diagnosis results is also investigated in the paper;in addition,the paper compares the studied method with commonly used fault diagnosis methods,and its comparison results show the effectiveness of the method.