Research On Fault Diagnosis And Anomaly Detection Of Mechanical Equipment Based On Siamese Network

As a widely used basic component,the working state of bearing is directly related to the stability of the whole equipment and plays an important role in mechanical equipment.Performance degradation assessment based on bearing fault diagnosis is a further development of fault diagnosis,which has certain theoretical significance and engineering application value.Thesis carried out research on fault diagnosis and anomaly detection of bearing based on siamese network.Siamese network was a model of meta-learning method,which can effectively solve the problem of small samples.Through siamese network and their variants,the characteristics of bearing vibration signals were extracted,and fault diagnosis model and anomaly detection model are built,which brought new solutions for fault diagnosis and anomaly detection of bearing.Specific research contents include:(1)Research on small sample fault diagnosis based on CNN-Bi GRU siamese network.Aiming at the problems that the fault sample was scarce and over-fitting in traditional deep neural network model in small samples and poor generalization performance,a fault diagnosis method based on CNN-Bi GRU siamese network was proposed.The siamese network was composed of convolution neural network and bidirectional gated recurrent unit that had the same structure and shared weights,the bearing sample pairs of the same category and different categories were constructed to input into the siamese network and the similarity was compared based on the L1 distance to achieve fault classification.A fault diagnosis experiment was carried out by using the measured bearing fault signal,and the results of compared with those of other deep neural network models.The experimental results show that the CNNBi GRU siamese network method still had superior diagnostic performance in the case of small samples.(2)Research on bearing fault diagnosis based siamese network structure.A framework for extending deep neural networks into siamese network structure was proposed to improve the fault diagnosis performance with small samples.Different from the traditional deep neural network,it was shown the siamese network adopts the method of sample pair training,so,the performance of bearing fault diagnosis was improved.The convolutional neural network and long short-term memory networks with different layers were respectively expanded into siamese network structure,carrying out a large number of fault diagnosis experiments on the public rolling bearing data set.The experimental results show that the accuracy of fault diagnosis results can be improved by expanding into siamese network structure,The accuracy of the Siamese CNN network was 2.44% higher than that of the corresponding CNN network,and the accuracy of the Siamese LSTM network was 9.27% higher than that of the corresponding LSTM network.(3)Research on anomaly detection based on DSC-Bi GRU siamese network.Aiming at the problem of bearing anomaly detection,an anomaly detection method based on the combination of depth wise separable convolution and bidirectional gated recurrent unit was proposed for the DSC-Bi GRU siamese network.The depth wise separable convolution was used to extract the time domain features of the vibration signal,and the bi-directional gated recurrent unit was used to extract the frequency domain features of the vibration signal,so that the bearing degradation features can be extracted more effectively and the degradation indicators can be constructed by fusing the time and frequency domain features.The anomaly detection experiments carried out by XJTU-SY full-life bearing data set.The survey results suggest that the constructed degradation index can intuitively reflect the bearing degradation trend,and was more sensitive to early abnormal phenomena,and had better generalization performance;the method of proposed anomaly detection can adaptively determine the threshold,with low false alarm rate and leakage rate,and can accurately detect anomalies;the designed model had lightweight characteristics,and the detection speed can meet the real-time requirements.