Deep Learning Pattern Recognition And Its Application In Fault Diagnosis Of Rotating Machinery

The fault diagnosis of rotating machinery is a process of pattern recognition,and pattern recognition can be divided into clustering and classification.The commonly used classification algorithms are artificial neural network(Ann),support Vector machine(SVM),and so on.The commonly used clustering algorithms include K-means algorithm Density-Based Spatial Clustering of Applications with Noise-based DBSCAN algorithm and so on.However,these algorithms have some unsolved defects,such as the artificial neural network is prone to fall into the local optimal solution and K-means algorithm is very sensitive to noise and outliers,and so on.In recent years,due to the rapid development of the Internet and the advent of the big data era,the deep learning model has been paid more and more attention.Deep Belief Network DBN(Deep Belief Network DBN)can be used in feature extraction and pattern recognition by stacking restricted Boltzmann machines.This paper is supported by the National Natural Science Found ation of China(project No.: 51575168).The deep learning model and clustering algorithm are combined in the fault diagnosis of rotating machinery.The main research contents of this paper are as follows:1.Aiming at the deficiency of ASTFA(Adaptive and Sparsest Time-Frequency Analysis)method in dealing with the early crack signal of gear,the minimum entropy deconvolution Entropy deconvolution(MED)is used to reduce the noise of the gear early fault signal.A MED-ASTFA method is proposed,which combines the MED-ASTFA and the Average Sideband logarithmic ratio(ALR).The method is applied to the quantitative diagnosis of root crack at variable rotational speed.2.The existing typical deep learning models are analyzed,and the validity of deep belief network in pattern recognition is verified by simulation signals,and applied to the quantitative diagnosis of tooth root cracks.3.Combining the strong feature extraction ability of DBN and the advantage of clustering algorithm,combining DBN and K-means algorithm,a clustering model based on PCDBN(Principal Components Deep Belief Network PC DBN and K-means algorithm is proposed.A clustering model based on PCDBN and K-means is proposed and applied to the state recognition of rolling bearings.4.To solve the problem that both K-means algorithm and K-means++ algorithm need to set the number of K in advance,the iterative Self-organizing Data Analysis Techniques algorithm ISODATAA(iterative Self-Organizing data Analysis)algorithm is introduced to realize automatic data adjustment and clustering by changing the conditions of merging and splitting.In order to solve the problem that ISODATA algorithm is easy to fall into local optimal solution,an improvement is made.A clustering model based on PCDBN and IS ODATA is proposed.And applied to the status recognition of rolling bearings.5.In this paper,a flexible fuzzy partition method,fuzzy C-means fuzzy C-means-FCM(FCM)algorithm is studied.The combination of PCDBN method and FCM algorithm can achieve more accurate clustering recognition results by controlling the number of m and k.