Statistical Process Monitoring And Control For Autocorrelated Processes

Globalization of markets depend on the product quality. Generally speaking, Enterprises of high quality product will win in the market competition, while companies of ignoring the quality will eventually fail. In fact, companies weren't intent to make low quality product. Low quality products are often caused by process variation. Therefore the production process will impact the quality. This is also the source of statistical process control (SPC).SPC is an important tool of statistical quality control (SQC). It mainly use statistical methods (such as control charts etc.) to detect and warn abnormal process fluctuations. Qulity managers can take timely measures to eliminate abnormal process behavior and ensure product quality according to the alarm. However, traditional control charts are applied only to normal, independent and identically distributed processes. But in modern manufacturing environment, operators, machines, measurement methods and other factors will result in the autocorrelation of process observations. In this article, some contributions to monitoring autocorrelated process are as follows:First of all, the traditional EWMA control chart has been improved. For autocorrelated processes, the variance of EWMA control charts has been reestimated, and the control limits have been adjusted to the new level. A simulation case is given to analyze the performance of control charts.Then state space method is introduced to the autocorrelated process. Time series model of processes is transformed into state space model. The concept of integrating SPC and engineering process control (EPC) have been discussed. What's more, we consider the performance of EPC controllers and the relationship between the average run length (ARL) performance and poles selection. The method is separately applied to design the EPC feedback controller for the first-order and second-order dynamic process in order to analyze the effectiveness.Finally, wavelets decomposition is used to achieve decorrelation of process data. In addition, multi-resolution wavelets analysis can monitor signal at multiple scales. Through Monte-Carlo simulation method, we study the relationship between the performance of control charts and wavelets decomposition depth.