Clustering Model Based On Deep Learning And Its Application In Fault Diagnosis Of Rotating Machinery

In recent years,Deep learning is developing very rapidly and applied to various fields,the fault diagnosis of mechanical equipment is gradually becoming intelligent and informationized from the original signal processing and manual judgment.This paper improved the network commonly used in deep learning: Stacked Denoising Autoencoder(SDAE),Deep Belief Network(DBN),Convolutional Neural Networks(CNN)and combined with the clustering algorithm which is applied to the fault diagnosis of rotating machinery.1)The structure and training method of SDAE network are researched theoretically,and the SDAE network is improved by Isometric Mapping(ISOMAP)method.A deep learning network based on SDAE is proposed: ISMSDAE network.Combining the ISMSDAE network with the Density-Based Spatial Clustering of Application with Noise(DBSCAN),the ISMSDAE-DBSCAN clustering model is proposed and applied to the diagnosis fault of rotating machinery,and obtained a good diagnostic effect.For the ISMSDAE-DBSCAN clustering model,the K value cannot be set,the fuzzy C-Means(FCM)algorithm is introduced into the ISMSDAE network.The ISMSDAE-FCM clustering model is proposed and applied to the rotor fault diagnosis,compared with the ISMSDAE-DBSCAN clustering model,the ISMSDAEFCM clustering model has better clustering effect.2)Introduced the Reconstruction Independt Component Analysis(RICA)method into the DBN network and proposed the RIDBN network.combined the RIDBN network with the FCM algorithm to propose the RIDBN-FCM clustering model which is applied to gear fault diagnosis.The clustering effect of RIDBN-FCM clustering model on gear faults under different K values and different ambiguities m are researched.3)The structure and training method of CNN network are researched.The tDistributed Stochastic Neighbor Embedding(t-SNE)method is introduced into the CNN network.A new network based on CNN is proposed: TSCNN network.Combining the TSCNN network with the K-Means Clustering algorithm,the TSCNN-KMeans clustering model is proposed and applied to the bearing fault diagnosis.The effect of TSCNN-KMeans clustering model on bearing fault diagnosis under different K values is researched.The time series input and time-frequency feature input of bearing signals are compared in the TSCNN-KMeans clustering model.The experimental results showed that the time-frequency features had better recognition in the deep learning method and have have practical reference significance for the fault diagnosis of rotating machinery.