Study On Robust Parameter Control Based On Statistical Regression Model For Complex Manufacturing Processes

When the manufacturing processes are getting more and more integrated and intelligent, there are a lot of complex manufacturing processes in various manufacturing companies. With the increasing demand on quality of products, manufacturers care more of the control performance of their manufacturing processes since the quality is produced. Generally, complex manufacturing processes are featured with a large number of process variables, complex interactions among those variables, multiple stages and platforms. Hence, it is pretty hard or even impossible to find an appropriate differential or difference equation to describe the complex manufacturing process, which makes controlling of complex manufacturing processes much harder. This article aims at the problem of process variable control in complex manufacturing processes. Based on the previous work on"regression model based robust parameter control", this article further perfects its framework and system; discusses the new classification of process variables and the new effect hierarchy principle in experimental design; proposes the design approach of cautious control law; develops an adaptive control charts with variable parameters; proposes an economical control law with reduced adjustment frequency; solves the online observation of attributes data. By implementing the proposed methods in steel industry, semiconductor industry, forming industry, the methods are verified to be effective and practical. There are five main sections in this article.1. Based on the existing method, we summarize and perfect its framework. Additionally, modeling approaches regarding experimental design are researched to improve and revise. The concrete solving procedures of the proposed method are also given.2. It is explicitly pointed out that the estimation error of regression model coefficients and the observation error of observable noises should be considered. A cautious control strategy is developed after the influence of those two types of errors on control performance is studied. The proposed cautious control law is capable of improving the robustness of control performance to uncertainties.3. An adaptive control charts with variable parameters is designed for a robust parameter control method applied process. Integrated with the applied control law, the proposed control charts can automatically adjust control chart parameters based on the prediction of system responses. There are two types of adaptive control charts developed. One enhances the tracking performance of process changes. The other one reduces the SPC run cost. The proposed SPC monitoring strategy, together with the robust parameter control method, effectively improves the process quality of complex manufacturing processes.4. An economical control law with reduced adjustment frequency is proposed based on the robust parameter control method. New concepts quality margin and self-compensation of noise change are proposed. We analyze the chances for adjustment frequency reduction that those two concepts could create. The innovative design of control law fully utilizes those two concepts which is quite different from the existing control law design. With a comparison study, we show that the proposed reduced control law requires much less in-line adjustments while guarantees the specified process quality.5. It is explicitly to consider the observation problem and implementing problem of attributes data process responses. Generally, attributes data process responses are hard to be measured online. However, one of the prerequisites for conducting regression modeling in complex manufacturing processes is process response should be quantified and observed in real time. In order to solve this problem, an image processing algorithm based on sensing camera is developed, which is able to realize the online observation of attributes data. By implementing this technique into continuous casting process, a core algorithm capable of automatically detecting surface defects of casting billets is developed. Then, a full form of data structure can be expected. In order to solve the modeling of attributes process response and variables process data, a logistical regression is built for casting process modeling. Based on this logistical regression, the proposed control method is implemented and optimal control law is obtained. The suggestions on control are accepted by steel plant, which verifies a successful implementation case of the proposed method in casting processes.Through integrating statistical methods, automatic process control, cautious control, control charts, optimal control, and image processing method, the article makes the proposed method more robust, adaptive, economical, and practical.