Research On Sensor Fault Diagnosis Method Of Control System Based On Data Analysis

The safety and reliability of the control system are the guarantee of the normal operation of the production process.Among them,the sensor fault is one of the main factors that cause the abnormal system.Therefore,how to through the data analysis method to mining information system of historical data or real time data,and does not need to invest in other equipment,it can be used to monitor the state of the control system,to prevent the occurrence of sensor faults,has a very important significance for the normal operation of the control system.First,the feature information of the fault data is extracted.Considering the noise in the data sample,the wavelet packet is used to denoise.Because of the high data dimension and the coupling between variables,fault diagnosis can be made even if the data is used directly.In order to reduce the data dimension and remove redundant information,we use the mean value kernel principal component analysis to extract the feature data of the control system's fault.Secondly,the fault detection is carried out.In this paper,the method of class mean kernel principal component analysis is used to detect the fault,and the two statistics T~2 and SPE are calculated to determine whether the system fails.Due to the different kernel function will affect the classification results of the data,with blindness and the selection of kernel parameters,so this paper selects a composite kernel polynomial kernel and RBF kernel function is composed of Gauss,the particle swarm algorithm of three parameters on the composite kernel function is optimized,and then use the class mean kernel principal component analysis algorithm fault detection.The fault data of the multivariable control system is taken as an example,and the superiority of the method is proved by the comparison with the principal component analysis and the kernel principal component analysis algorithm.At last,the type of fault is identified.This paper uses the method of BP neural network to identify the fault type,using part of fault feature extraction after the training set into the BP neural network,to construct a fault model,and then the rest of the sample set for fault identification to send the trained network identification,calculate the failure rate,and compared with the BP neural network diagnosis based on the characteristics of the original fault,verify the validity of the method.