Research Of Principal Component Analysis And Its Application In Data Reconstruction Of Fault Sensor

With the increasing spring up of all kinds of intelligent device,the reliability and security of a system has become the key factor to guarantee economic benefit and social benefit. At the same time,various kinds of detecting struments and equipments go into more and more complicated. Sensor,as an important source of information for these devices and systems,plays an important role throughout the entire control operation and detecting system.In a highly sophisticated intelligent system, It is beyond the ability for manipulator to detect all kinds of faults in this system rapidly and timely so that unnecessary losses reduced and equipment attrition brought down.All this make automatic fault detection and diagnostic system emerge because of demand, opportunity etc.Fault diagnosis method,based on multivariate statistical process control,is an important branch of fault diagnosis field.For the past few years, various forms of principal component analysis get in-depth research and are widely applied in every fields of fault diagnosis and get good effects.In this dissertation,the fundamental theory of Principal Component Analysis(PCA) have been deeply analyzed. To eliminate the influence of dimension and whether selected principal components have namely representative,this thesis has presented Relative Principal Component Analysis(RPCA), and based on this, has proposed a data reconstruction method applied RPCA;Conventional PCA can not be used in the nonlinear systems,in order to overcome this deficiency,a nonlinear fault diagnosis method based on Kernel Principle Component Analysis (KPCA) has been introduced.But KPCA cannot be realized to identify the faulty variables,as it can only achieve failure testing. In this paper, KPCA is combined with a method of reconstruction for the raw data in original data space and a way based on it to identify the faulty variables.Specific issues that deserve more attention and deeper understanding have been presented.In the end,the proposed methods are verified by a simulation test for continuous annealing process of strip tension system.