Fault Diagnosis Algorithms Of Chemical Process Based On Independent Component Analysis

With the expansion of the scale of the chemical process and the complication ofthe process flow, the probability of occurrence about all kinds of chemical industryaccident is increasing day by day. On the other hand, during the production processany misoperation is likely to cause failure, which affect the quality of products andcause economic loss. Therefore, correct and effective fault diagnosis has become thekey factor to guarantee the high quality of the product and the stable operation of thefactory. In addition, with the wide application of computer system, the monitoringsystem significantly improves the ability of saving data.These data implied a largeamount of information. Through data analysis we can monitor running state of theprocess, then judge the possible fault and put forward corresponding measures. As aresult we can greatly reduce the possibility of accident. It forces the development ofmultivariate statistical methods.This paper mainly makes a study about independent component analysis methodof the multivariate statistical methods, and proceed deepgoing research in the light ofthe problems about applying the method in fault diagnosis, the main work as follows:1,Although fuzzy neural network (FNN) has higher accuracy in fault diagnosismethod of chemical process, it still exists problems about slow operating rate and lowsensitivity. Therefore a fuzzy neural network based on kurtosis (KFNN) method ispresented and applied to ketone-benzene dewaxing fault diagnosis of chemicalprocess. The method using the information of ICA algorithm feature extraction formindependent component, for these independent component with the kurtosis fordimensional reduction, extraction refine data, remove redundant data in ensuring thathas high accuracy of circumstances, can improve chemical process of fault diagnosisof the operation speed and sensitivity.2, Production data of chemical process usually have characteristics ofnonlinearity and distributives. Based on Kernel Principal Component Analysis (KPCA)which can effectively deal with nonlinear data and Kernel Independent ComponentAnalysis(KICA)which can effectively deal with distributed data, a fault diagnosisalgorithm called Dual-Kernel Independent Component Analysis of chemical processis proposed. Firstly, this algorithm uses KPCA to whiten data by mapping intohigh-dimension feature subspace with nonlinear kernel function. Then the whiteningpreprocess data are applied to KICA and the results of kernel independent componentanalysis are got. Finally the monitoring of nonlinear chemical process is implementedby constructing statistical indices and control limit in the feature space. The algorithm is applied to CSTR process. The results show that DKICA algorithm can effectivelyincrease the accuracy and sensitivity and reduce false negative rate and false positiverate of the nonlinear chemical process fault diagnosis.3,Because of the problems about inaccurate of data estimate when The kernelindependent component analysis is applied to batch process data and multimodalcalculation, this paper proposes a modeling based on FCM clustering dual-coreindependent component analysis (DKICA) batch process fault diagnosis algorithm.The algorithm firstly batch process data batch development and launch, and then willopen variables of high dimensional data according to clustering module classification,in classification of each data module application DKICA extracts independentcomponent in modeling, and calculated I2and SPE statistics and the correspondingcontrol limit. The simulation results show the feasibility and effectiveness of theproposed methods, and show better diagnosis effect than KICA.