Process Monitoring And Quality Control Based On Statistical Methods

In process industry, the effective process monitoring and quality control are the key to ensure safety, enhance product quality and economy benefit. For the complex chemical processes, it is difficult to achieve the exact mathematical model of industry process. Even if it could achieve, the equations that deduced from the original theory only describe the balance relationship of part of energy and material, this will limit the application of process monitoring methods based on systems theory or rigorous process models. On the other hand, with the application of the computer technology, distributed control system, all kinds of intelligent meters and control equipments, large amount of data are sampled and collected. It is one of the most active research areas in the field of process control that how to transform these collected data into valuable information, and mine deep-level information about process operation to improve the performance of process monitoring.Statistical process monitoring (SPM) is a data-driven method, which is based on multivariate statistical methodologies for online process malfunctions detection and diagnosis through analysis and interpretation of the collected measurements. The information extracted from process data could reflect the operating status of the process at any time and guarantee the efficient, safe and product quality of the process. In view of characteristics of the continuous and batch industry process, some improvements of tradition SPM methods have been made at different degree, and some new statistical monitoring algorithms are also proposed in this thesis.The main working results and contribution of this dissertation are stated as follows:1. A combination method of wavelet transform (WT) and principal component analysis (PCA) for process monitoring is proposed. WT could effectively restrain the noise and disturbance that contained in the process data; PCA are used to reduce the dimension of the process data and establish principal component (PC) model for monitoring. The proposed method is more exact and effective than the tradition PCA monitoring method. In view of insufficient number of fault samples in process industry, support vector machines (SVM) with the capacity of sufficient learning from a limited training set are used to diagnose the process fault. When fault occurs, the PC score vectors extracted from PC model input the well-training multiple SVMs, the outputs are used to identify the fault. The results from the application to the Tennessee Eastman (TE) challenge process demonstrate its effectiveness.2. In view of the characteristics of the batch process monitoring and the Fisher discriminant analysis (FDA) method with the advantage of data classification, a new batch process monitoring and fault diagnosis method based on feature extract in Fisher subspace is proposed. The approach only uses the present data and overcomes the need in multi-way PCA (MPCA) for pre-estimating the unknown part of the process variable trajectory from the current time until the end of the batch. Feature vector and discriminant vector of the process data that extracted from the Fisher subspace are used for process monitoring and fault diagnosis. Simulation results on a penicillin fermentation process can demonstrate that, in comparison to the MPCA method, the proposed method is more accurate and efficient to detect and diagnose the malfunctions.3. Kernel learning theory is introduced into linear FDA. A nonlinear statistical process monitoring and fault diagnosis method based on kernel FDA (KFDA) is proposed. The basic idea of KFDA is to first map the original space into high dimension feature space via nonlinear mapping and then extract the optimal Fisher feature vector and discriminant vector to achieve process monitoring and fault diagnosis. In comparison with Kernel PCA (KPCA), the proposed method can decrease the calculation and avoid determining the number of kernel PCs. It is easy to realize fault diagnosis through optimal FDA vector. The KPCA method is evaluated by the application to the fluid catalytic cracking unit (FCCU) model and its effectiveness is demonstrated.4. Traditional quality control method could not well restrain the process disturbance that will influence the final product quality when the quality measurements are not available on-line or they have long time delays. A multivariable statistical quality control method is presented to decrease variance in product quality by the influence of process disturbance. A data-based model achieved under normal operation condition is to predict quality variables through high-rate frequently available measurements of process variables. The difference between the value of prediction and setting is used to obtain the adjustment of the process manipulated variables. The process disturbance can be restrained and the variance of product quality can be reduced by regulating the process manipulated variables. The proposed method is demonstrated on the TE process. Simulation result shows that the variance of product quality used by proposed scheme is smaller than that utilized by conventional PID quality control.5. SPM methods successfully applied to methanol rectify process of Shanghai coking and Chemical Corporation. The characteristic and technics of the methanols rectify process were analyzed and the process variables were chosen. The data collected in normal operation condition were used to build statistical monitoring model. To demonstrate the effectiveness of the model, the data were collected on-line and in the condition of leakage of the pressure tower reboiler. What have done lay the foundations for the model of statistical process monitoring carrying into execution.