Researches On Performance Monitoring And Fault Diagnosis For Process Industry Based On Statistical Theory

How to ensure safety, enhance the quality of products and the overall economic benefit are the critical problem in the process of modern industrial. An effective of technical method to solve the problem is the application of effective process performance monitoring and fault diagnosis. The rapid development of the computer and information technology made that a large number of process data of industrial have been sampled and collected. Since the methods of statistical performance monitoring rely only on readily available process data and do not assume accurate mathematical model, they have been paid more attention.In this paper, some improvements of traditional methods have been made which based on the research of traditional statistical process monitoring method, and furthermore some new statistical monitoring methods are proposed too. The dissertation includes following main contents:1. A new method combined wavelet packet transform (WPT) with principal component analysis (PCA) for process monitoring is proposed. Firstly, the method use WPT technology to restrain the noise and disturbance that contained in the process data effectively. And then the PCA technology is used to reduce the dimension of the process data and establish principal component (PC) model for monitoring. The contribution plots which represent the contribution of each monitoring variable to the PC are used for fault diagnosis. The monitoring results of the application to the Tennessee Eastman (TE) chemical process confirm its effectiveness.2. A new fault detection and diagnosis method based on kernel principal component analysis (KPCA) is described. Firstly, it removes the noise from data set using WPT. The difference is that the new method here using KPCA to detect fault. And furthermore kernel function gradient algorithm is used to diagnosis fault. KPCA contribution plots are protracted which represent the contribution of each monitoring variable to the statistics. During the monitoring process, the feature vector selection method scheme is given to reduce the computation complexity of kernel matrix. The proposed method is applied to the simulation of Tennessee Eastman (TE) chemical process. The monitoring results confirm that the proposed method can effectively detect faults and diagnoses faults.3. Using the advantage of KPCA for nonlinear monitoring and introducing the accuracy of multiple kernel learning support vector machines (MKL-SVM) for fault diagnosis, a new method for nonlinear process monitoring based on KPCA and MKL-SVM is proposed. The data is analyzed using KPCA. In the feature space, through constructing the statistical index and control limit, performance monitoring is implemented. T2 and SPE are constructed in the future space. If statistical index exceed the predefined control limit, a fault may have occurred .Then the nonlinear score vectors are calculated and fed into the MKL-SVM to identify the faults. The proposed method is applied to the simulation of Tennessee Eastman(TE)chemical process .The simulation results show that the proposed method can identify various types of faults accurately and rapidly.4. To further improve the diagnosis speed and accuracy, a new method for nonlinear process monitoring based on KPCA and sparse SVM is proposed. The data is analyzed using KPCA. Through constructing the statistical index and control limit in the feature space, performance monitoring is implemented. If statistical index exceed the predefined control limit, a fault may have occurred. Then the nonlinear score vectors are calculated and fed into the sparse SVM to identify the faults. The proposed method is applied to the simulation of Tennessee Eastman(TE)chemical process. The simulation results show that the proposed method can identify various types of faults accurately and rapidly.5. On the analysis of the characteristics of complex industrial processes basis, making full use of advantage of kernel principal component analysis method which could dealing with nonlinear data and the ability of kernel independent component analysis (KICA) to extraction high-dimensional feature space information, a new nonlinear performance monitoring and fault diagnosis method based on kernel independent component analysis (KICA) and SVM is proposed. The key of this method is map the data into high-dimensional feature subspace. And then analysis and computation can be done using the KICA algorithm. In the feature space, with the help of constructing the statistical index and control limit, the performance monitoring is implemented. SVM with the capacity of data classification are used to diagnosis the process fault. The effectiveness of this new method is confirmed by the application to the Tennessee Eastman (TE) chemical process.6. A new statistical process monitoring and fault diagnosis method having the character of nonlinear which based on KICA and kernel FDA (KFDA) is proposed. Kernel learning theory is introduced into linear fisher discriminant analysis (FDA). KICA is used to establish the normal operating conditions and detect the fault. If a fault occurs, the nuclear fisher discriminant vector and feature vector F of the process data are extracted from the Fisher subspace. Thus, the batch normal or not can be detected by comparing distance with the predefined threshold. Comparing the present discriminant vector and the optimal discriminant vector of fault in historical data set, the similar degree can be detected. According to the similar degree, the perform fault can be diagnosed. The results of simulating TE process demonstrate that the proposed method can efficient in detecting and diagnosing the malfunctions,with more accurate result.