Research On Equipment Condition Monitoring And Nonparametric Bayesian Diagnostic Method For Variable Condition

With the significant increasing in data type,data volume and complexity of equipment status monitoring data,when fault diagnosis is performed,modeling the fault data with finite parameters or using a single model will become unreliable.A novel fault diagnosis mechanism is needed to make the model no longer limited to training samples,but to adaptively change and adjust the existing model as the training sample changes.In this paper,the fault diagnosis method based on nonparametric Bayesian is proposed,and the main contents and conclusions are as follows:(1)Feature extraction method of vibration signal based on homogeneous multimodality.The experiment of crank pin wear fault of reciprocating compressor is carried out.By analyzing the experimental data of different fault degree,it is verified that the diagnosis effect of the proposed method is superior to the single mode including the time domain,the frequency domain and the time frequency domain.(2)Fault Feature dimension reduction based on GDBM.In order to solve the problem that the high dimensional data set contains too much redundant information and increase the data processing time,the feature dimension reduction model based on GDBM is constructed by studying the relationship between feature transformation and dimensionality reduction and the essence of GDBM learning process.The experimental results have shown that the original feature space is reduced based on GDBM and the diagnosis effect is improved obviously.(3)Fault diagnosis method of mechanical equipment based on Dirichlet process mixture model(DPMM).Aiming at the problem that the number of clusters in the traditional clustering model needs to be determined artificially,a fault diagnosis method based on DPMM is proposed.By constructing a set of sub-data sets with different classes as the training set and all classes data sets as the testing set,it has been verified that the DPMM model could adaptively determines the number of clusters with the change of data.In particular,when a category in the test set is not found in the training set,DPMM can dig it in the form of a new class.(4)Research on diagnosis method of equipment variable condition based on Hierarchical Nonparametric Bayesian(HNB).Considering the few fault samples and the complex and changeable operation condition of mechanical equipment,the HNB model which can realize the knowledge transfer is put forward.Through two experiments,it is verified that the fault recognition effect of HNB is better than that of traditional SVM,BP neural network and KNN under a small number of fault samples.