PLS-based Statistical Quality Monitoring: Theories And Applications

The data-driven industrial statistics extract the inherent mechanism of the industrial process by making full use of the accumulated operation data. It is one of the main methods to ensure the safety of operation and increasing the quality of the process. While the PLS (Partial Least Squares) algorithm is just one of the key methods of the data-driven industrial statistics.The topic of this thesis focuses on the researches on the theories and applications of the PLS based modeling, prediction, and quality monitoring of the industrial processes. According to the characteristics of continuous process and batch procedures, proposed the methods as follows: a) The sufficient and necessary conditions of faults detectability under the PLS framework; b) PLS-based optimal quality control model; c) VPLS (VP-PLS) based quality and cost control; d) Sensor faults isolation methods; e) Q_v statistical index to monitoring sensor faults; f) Discounted-measurement RPLS algorithm; g) RPLS (Recursive PLS) based adaptive on-line quality monitoring strategy and so on. The theoretical findings are fully supported by the application performed on the TE process (one of the major chemical benchmarks) and the rubber mixing process. Furthermore, the methods such as: the on-line quality prediction and monitoring, rubber-discharging control methods etc., aiming at improving the quality of mixture, have been applied successfully on the quality control of the rubber mixing process in a large-scale tire plant located at east China.In case of the continuous industrial process, the TE (Tennessee Eastman) benchmark is the application background. The main researches including:a) Though the PLS algorithm is widely used for quality monitoring, some fundamental questions (such as the sufficient and necessary conditions of faults detectability) have not been resolved. Then they are discussed here.b) A new Partial Least Squares optimum detecting model (PLS-ODM) is proposed to improve the monitoring performance of the PLS method. This PLS-ODM is deduced according to the fault subspace and the V_P (Variable importance in Projection) index of the process variables that could be used to estimate the importance of the measurements to the quality. Compared with traditional PLS, the PLS-ODM has the stronger fault detecting power, and could detect much weaker important process faults exactly in time.c) A new ï¿¡)rpv statistical monitoring index is proposed to improve the detection power of the PLS method for on-line quality monitoring. In terms of the VP indices of monitored process variables, Qrpv is calculated only by the residuals of the Remarkable Process Variables (RPVs). Therefore, it is the dominant relation of the process not all process variables (as in the case of the conventional PLS) that is monitored by this new VPLS method. Hence the new Qkp\ index is significantly advanced over and different from the conventional Q statistic. The combination of ï¿¡)rpv and Hotelling T2 statistics is applied to the quality and cost control of the Tennessee Eastman (TE) process, then weak faults can be detected as quickly as possible. Consequently, the product quality of TE process is guaranteed and operation costs are reduced.d) At the basis of the Vp index of the process variables, a new Qv monitoring statistical index is developed to improve the sensor fault detection power of the data-driven monitoring methods. This index makes the monitoring system much sensitive to important sensors that could heavily affect on the quality of the process. Then using the Vp index to guide the design of the structured residuals matrix to identify the important faulty sensors.In case of the batch procedures, it is at the basis of the rubber-mixing process practice:a) To overcome the shortage of the traditional PLS algorithm, a Discounted-measurement Recursive PLS (DRPLS) algorithm is proposed to build the Mooney-viscosity model of the mixture. This improved RPLS model could be on-line updated with new data, and more adjustable parameters are available. On the other hand, the rub...