Research On Fault Detection And Diagnosis Based On PCA And Bayesian Classifier

In the industrial production process,real-time and accurate monitoring of the entire process is one of the necessary conditions to ensure the safe operation of the industrial production process.The two core links of process monitoring are fault detection and fault diagnosis.It is time-consuming and labor-intensive to rely on humans to monitor the entire process.Therefore,some algorithms for fault monitoring and diagnosis need to be introduced instead of humans for monitoring.The traditional PCA fault detection method uses the regular statistics of each signal in the control process to divide the original data space into a main subspace and a residual subspace,and uses the T~2 statistics and Q statistics in their respective spaces to perform fault detection respectively.There are a large number of non-linear links in the industrial process.The kernel method is introduced into the PCA method.The original data is mapped to the high-dimensional space through non-linear mapping,and then PCA is processed in the high-dimensional space.In the traditional PCA-based fault detection method,when each dimension of a data sample obeys a similar probability distribution,the Q statistic can perform fault detection well in the residual space.However,when each dimension of the data obeys a different probability distribution,the detection effect of the Q statistic will be significantly reduced.In view of the disadvantages mentioned above,this paper divides the residual subspace after PCA decomposition into active residual subspace and stable residual subspace,and then proposes a weighted and combined index in the residual subspace.A variety of detection indicators are applied to the fault detection of typical TE control processes.Simulation results show that the comprehensive indicators proposed in this paper have higher accuracy in fault detection.After using various detection indicators to find a fault,the next step is to use the classifier to determine the type of fault.The Bayesian classifier is based on the knowledge of probability theory for classification.The model has a high degree of interpretation and classification accuracy,and is widely used in fault diagnosis.The fault diagnosis method based on Bayesian classifier is applied to the fault diagnosis of the superheated steam temperature control process in thermal power plants.PCA also has a function to reduce the dimensionality,calculation volume and correlation of the data;then using the training data set of each fault to train the naive Bayes classifier of continuous attributes,and use the kernel density estimation method to calculate the likelihood probability density;finally,input the test data set into the trained Bayesian classification It can determine the fault source of each fault data sample and evaluate its diagnostic accuracy.Simulation results show that the method has higher diagnostic accuracy.