Nowadays, quality is one of the very basic and crucial factors for any products or services in business world. It determines the success of an organization or product to a large extent. Large variability is believed to be the biggest enemy of excellent quality, especially, the variability caused by certain assignable causes. In practice, Statistical Process Control (SPC) techniques have been widely applied in industries to detect assignable causes and reduce variability. Most SPC charts are designed to detect assignable causes in single stage processes encountered in industrial practice. Due the complexity of modern techniques, more and more tasks and products require more than one stages/operations to accomplish. This is what we call multistage manufacturing processes (MMP). However, a...[ Read more ]

Nowadays, quality is one of the very basic and crucial factors for any products or services in business world. It determines the success of an organization or product to a large extent. Large variability is believed to be the biggest enemy of excellent quality, especially, the variability caused by certain assignable causes. In practice, Statistical Process Control (SPC) techniques have been widely applied in industries to detect assignable causes and reduce variability. Most SPC charts are designed to detect assignable causes in single stage processes encountered in industrial practice. Due the complexity of modern techniques, more and more tasks and products require more than one stages/operations to accomplish. This is what we call multistage manufacturing processes (MMP). However, applications of conventional charts to such multistage processes usually result in an unsatisfactory performance due to two main obstacles.

1. Limited resources, such as measurement systems and manpower, it is usually too costly to place SPC charts at every stage, especially for the processes which have a large number of stages.

2. Incomplete information. Even if we have the capability to set up control charts for all the stages, it is still possible that the whole monitoring performance (in terms of Average Run Length and Average Time to Signal) is not good enough, because each stage is treated as a single stage and inherent correlations among stages are ignored.

Therefore, we try to improve the monitoring performance and efficiency by investigating the inherent characteristics of multistage processes and introducing new monitoring strategy.

In this thesis, we first propose a strategy to properly allocate control charts in a serial multistage manufacturing process (S-MMP) that can enhance the fast detection of out of control behaviors of conventional SPC. Based on our chart allocation strategy, information of inherent interrelationship among stages is involved in decision making to achieve quicker detection of a potential fault. Two automotive assembly examples are used to demonstrate the applications of the chart allocation strategy. It shows that the rational strategy may differ from industrial common sense in some cases. The impact of uncertainty in the structural parameters is also considered, which may allow practitioners to make more realistic decisions in serial multistage manufacturing processes when they face such uncertainty or error. Guidelines of application procedures are also provided.

After investigating the chart allocation strategy in S-MMP, we carried on our research on extending the strategy to a more complicated and realistic situation-serial parallel multistage manufacturing processes (SP-MMP). The three special scenarios of SP-MMP are analyzed and modeled respectively. One of most important features of SP-MMP is that the mean shift in upstream stages may cause not only the mean shift but also the change of variance in downstream stages. Corresponding modifications in process modeling and optimization formulation are made in this work. Furthermore, the limited resource problem is also considered in the optimization formulation so that it can help us to determine appropriate amount of monitoring resource for a processes. Besides the chart allocation strategy in a MMP, it is also important for people to decide which kind of control charts is appropriate for a MMP. We tackle this problem together with the chart allocation in order to form a monitoring strategy for a MMP. Control charts are divided into two classes due to the data form: 1. Output monitoring chart; 2. Residual monitoring chart. The whole monitoring strategy is formulated to a nonlinear integer programming problem. However, when the number of stages and charts become large, it is usually hard for people to obtain the optimal solution for such problem. Therefore, we further simplify the problem to a max-min problem to achieve good solution if not the best with less effort.

Whenever a signal appears in the monitoring system of a process, people need to take corresponding actions to pull the process to the right track again. In the final part of the thesis, we also tackle this problem by studying the control policy for run-to-run processes which are common in semiconductor manufacturing and chemical engineering. Smith predictor structure is introduced together with the EWMA controllers to deal with the inherent metrology delays in run-to-run processes.