Research On Compressed Sensing And Fusion Method For Remote Bearing Fault Diagnosis

For a long time,with the rapid development of Internet technology and sensor technology,the data generated in the industrial production process have increased explosively.Although more data bring new diagnostic ideas for fault diagnosis,they also bring some new problems.Effective and real-time fault diagnosis and monitoring methods for intelligent rotating machinery have always been research hotspots.The bearing,as one of the most important parts of contemporary machinery and equipment,plays a vital role in the smooth and healthy operation of machinery and equipment.Accurate and real-time classification and early warning of faulty bearings are important in terms of economy and safety.In the scenario of using the cloud for data processing,t he mechanical equipment and data acquisition sensors are located at the far end of the cloud.When the data are transmitted from the collection end to the cloud,the limited network bandwidth and the transmission rate delay will greatly affect the real-time diagnosis influences.In response to this problem,this thesis proposes a compressed sensing method and a multi-sensor information fusion method to achieve an intelligent fault diagnosis method that requires low bandwidth and meets the requirements of re al-time diagnosis and diagnosis accuracy.The main research work is described as follows.This research first verifies the sparseness of the vibration signal to prove that the bearing signal meets the prerequisites for using compressed sensing.Then,based on the traditional tracking algorithm,a perceptual signal reconstruction algorithm is proposed,which is faster in use,searches superior gray wolf pack intelligent optimization algorithm in sub-space selected by track algorithm to select the optimizatio n results under different strategies,expands the dictionary atom set constructed by the matching algorithm,and improves the success rate of reconstruction through a larger atom search range,and reduces reconstruction error.Subsequently,the framework o f fault diagnosis on the reconstructed data is carried out on the diagnosis side.The framework includes data preprocessing and fault classification.In the data preprocessing stage,a feature ext raction method PLC is proposed.Feature extraction method is used to reduce data redundancy and speed up model recognition.In order to make full use of the advantages of two traditional feature extraction methods with different dimensionality reduction strategies,features suitable for diagnosis model input are extracted,and canonical correlation analysis is performed on the features extracted by the two methods.The two sets of new features with the greatest correlation are calculated as the input of the subsequent diagnosis model.Finally,in order to make full use of the multi-sensor data to improve the accuracy of the diagnostic model,a multi-sensor information fusion method based on the Gaussian kernel function is proposed.It is inspired by the kernel function in the SVM.The covariance matrix is used to perform multi-dimensional Gaussian distribution fusion for each sample point of the multi-channel signals.According to the fluctuation of the mean value of each sample,the fused one-dimensional fusion signal is obtained.The signal contains more obvious fault information to train the model and improve the diagnosis accuracy.Through the research content and simulation verification,the proposed compressed signal reconstruction method and multi-sensor data fusion technology can effectively reduce the impact of the diagnosis process on the condition of satisfying the diagnosis accuracy in the working condition of the remote bearing fault diagnosis.Therefore,the fault diagnosis method combining CS and multi-sensor information fusion has great research value and potential in the scenario of remote fault diagnosis.