Research On Intelligent Rotating Machinery Fault Diagnosis Method Of Rolling Bearing With Multi-failure Modes

The bearings,which are key components in machinery,may occur structural failures affected by alternating loads,high temperature and high pressure in the long-term operation,which will lead to equipment damage and safety accidents if not found in time.Therefore,the research of bearing fault diagnosis is of great significance and engineering application value.The current fault diagnosis methods mainly based on fault mechanism,signal processing and data-driven.Among them,the data-driven fault diagnosis method,which has the advantages of not relying on accurate physical models and priori diagnosis knowledge,is the most promising method at present.However,it relies on the balanced dataset to obtain ideal diagnosis result,but the actual data is collected in the normal operation,and there are many normal samples but few fault samples,so the positive and negative samples are seriously imbalanced.Therefore,this paper proposes an intelligent fault diagnosis method based on novel feature extraction method and deep learning model under data imbalance condition,to explore the correlation relationship between signals and faults and enable intelligent diagnosis of single and compound faults.The main research contents include:(1)Research on bearing fault mechanism and fault signal characterization method.To study the occurrence mechanism of bearing faults,and evaluate the anti-interference ability,the difficulty of signal acquisition and the volume of fault information,then screen out vibration signal to carry out the subsequent diagnosis research.To study the feature patterns of different faults in the time domain,frequency domain and time-frequency domain based on vibration signals,and to identify the relationship between faults and signals.Through reading other references,we build our own Shandong University bearing fault diagnosis platform,and collect data to conduct diagnosis experiments,so as to realize an independent and complete closed-loop research from data acquisition to diagnosis methods.(2)Research on the intelligent diagnosis method of single fault under data imbalance condition.To address the problem that small volume samples under data imbalance can hardly provide sufficient fault features,the bearing single fault intelligent diagnosis method based on Mel Frequency Cepstral Coefficents and optimized convolutional neural network is proposed.Using Mel Frequency Cepstral Coefficents to mine more obvious and differentiated features and building a convolutional neural network optimized by channel attention and submodular normalization,reducing the fault feature shifts caused by imbalanced data,mining the features which are helpful for fault diagnosis,improving the accuracy and generalization ability of the model.Diagnosis experiments are conducted on the CWRU dataset and the Shandong University single fault dataset to verify that the proposed method can effectively improve the performance of single fault diagnosis under data imbalance conditions.(3)Research on the intelligent compound fault diagnosis method under data imbalance.To address the problem that different components in compound faults are mutual concealing and coupling which makes compound fault hard to identify,an intelligent diagnosis method based on Frequency-domain Gramian Angular Field and improved convolutional neural network is proposed.Using the Frequency-domain Gramian Angular Field to extract more obvious and differentiated features,and building a convolutional neural network optimized by channel attention and instance normalization,fully preserving the fault style of feature map,deeply mining and enhancing compound fault features,and weakening the negative impact of data imbalance.Diagnosis experiments are conducted on simulated compound fault dataset and Shandong University compound fault datasets,and it is verified that the proposed method can effectively improve the compound fault diagnosis effect under data imbalance.(4)Validation of single and compound fault diagnosis of rolling bearings under working condition changing.The effectiveness of the fault diagnosis method is also affected by the working condition changing,so the extended experiments are conducted to verify the generalization ability of the proposed diagnosis method.Based on the CWRU and self-built compound fault dataset,the diagnosis experiments under different speed variation conditions are conducted to verify the diagnosis effect of the proposed single fault and the compound fault diagnosis methods,and the diagnosis results are analyzed and discussed.This paper uses Mel Frequency Cepstral Coefficents and Frequency-domain Gramian Angular Field to achieve differentiated feature extraction for single and compound faults of bearings,weaking the influence of data imbalance and enhancing the fault information before input to model.The diagnosis model based on advanced normalization and attention feature enhancement network is designed to alleviate the interference of data imbalance and working condition changing,which helps to deeply mine fault features and improve the accuracy and generalization ability of the model.The experiments based on the CWRU dataset,Shandong University dataset and simulated vibration signal dataset have fully validated the fault diagnosis effect of the proposed method under the data imbalance conditions.