Research On Fault Diagnosis Method Based On Independent Component Analysis

With the continuous development of information technology,the production technology of process industry becomes more and more complex,and the control process of the factory generally has the characteristics of nonlinear,strong coupling and high latency.The establishment of accurate mathematical models of system is difficult,thus fault detection and diagnosis method based on the data in complex industrial process gains more and more attention.Independent component analysis(ICA)is a classical method of data driven method,which can be used for fault detection by extracting independent component information from observation data.However,there are still some problems in the research of fault detection based on ICA algorithm:(1)Traditional ICA algorithms extract independent components by processing off-line data,there is still no good solution to realize on-line,stable,effective and fast detection of monitoring data.(2)In the kernel independent component analysis(KICA)algorithm for non-linear systems,kernel parameters are generally selected by empirical methods,and the detection results are dependent on kernel parameters,which is somewhat accidental and lacks the basis for the selection of kernel parameters.(3)Considering part of the variable data of the system transient process is nearly uniform distribution,the traditional ICA algorithm is only suitable for fault detection of non-Gauss distribution(super-Gauss distribution or sub-Gauss distribution)data,how to detect faults during transient process still remains challengable.Main results are as follows:A moving window ICA method with adaptive thresholds(MWAT-ICA)is proposed to detect faults online,and the detection effect of MWAT-ICA and ICA is compared by setting four types of failures in the three-tank simulation system.The results show that MWAT-ICA can not only realize online detection,but also has higher detection rate,lower false alarm rate and detection delay.The algorithm based on Chaotic Particle Swarm Optimization(CPSO)and KICA is proposed.CPSO-KICA is presented to obtain the best detection results with parameters selected by CPSO.And 20 faults collected on TEP are used to test the algorithm.The results show that this method can determine different kernel parameters for different types of data,and has higher detection rate and lower false alarm rate than the existing WKICA and PSO-KICA methods.A new ICA algorithm based on segment coordinate transformation(SCT-ICA)is proposed to detect faults during transient process.Segment coordinate transformation is used to stabilized the unstable variables in the transient process,and it solves the problem that the traditional ICA method can't diagnose the fault of the transient process of the system.4 faults collected on the physical experiment platform(IPC-TF)are used to verify the algorithm.It is demonstrated that the method can be applied to fault detection in transient processes.