Research On Fault Detection And Diagnosis Based On Improved Principal Component Selection Method

With the improvement of the modernization level of the industrial system,the system in the production process have higher and higher requirements for the safety and reliability,so the fault detection and diagnosis technology has been developed rapidly.The purpose of fault detection and diagnosis aims to quickly detect the fault in the early stage of equipment fault occurrence and identify the fault type,to facilitate the follow-up processing decisions and avoid the deterioration of the fault causing unnecessary losses.The data-driven approach is a hot topic in fault detection and diagnosis at present.Its main task is to realize online monitoring of the operation process by analyzing and modeling the data in the industrial production process.Based on the traditional method,this thesis proposes an improved principal component selection method and applies it to fault detection and diagnosis.And the thesis mainly takes the following aspects:(1)In the principal component analysis algorithm,a new principal component selection method is proposed considering that the traditional cumulative percent variance selection does not take into account the important information contained in the small variance.ReliefF algorithm is used to select the principal component with significant contribution weight as the principal component,instead of just relying on the variance.(2)To improve the performance of traditional PCA in fault detection,a new decentralized weighted ReliefF-PCA(DWRPCA)method is adopted based on the improved principal component selection method and the importance of distinguishing different variables by the idea of decentralized modeling.(3)After fault detection,the ReliefF-PCA combined with continuous hidden Markov model(RPCHMM)is proposed:ReliefF-PCA is used to preprocess the data to reduce the size of the data,to improve the modeling and running speed of the model.To prove the validity and reliability of the fault detection method based on DWRPCA,this thesis used it in the simulation test of the TEP.Finally,the RPCHMM method is used to diagnose the detected fault,and the diagnosis results show the feasibility of the method.