Research On The Theory And Methods Of Statistical Process Monitoring And Adjustment

Statistical Process Control (SPC) and Engineering Process Control (EPC), as two kinds of process control methods to variability reduction, have been successfully applied in manufacturing industry and process separately. SPC utilizes the control chart to detect and eliminate process variations caused by some special causes; EPC can adjust the variables to compensate the errors between process target value and output value based on feed-back control principle. Recently, the integration of SPC with EPC to monitor and adjust the process has become one of important content in the quality control field. In the tradition, SPC techniques have been popularly applied in the manufacturing industry to monitor the process, but little systemic research is found in how to improve the quality through the process adjustment. And furthermore, the traditional monitoring control chart will become meaningless to autocorrelated process output data. Therefore the integration of EPC with SPC has the potential to provide a new way to solve the problems in autocorrelated process.To solve the process initial error problems caused by machine-set occurring frequently in those small-batch of short-run production, we attempted to introduce Grubbs harmonic rule to adjust them; for those prevalent process auto-correlated questions, integration of MMSE adjustment with control chart was tried to monitor the process. A series of process adjustment methods were evaluated and compared; the effect of MMSE adjustment on the shift or drift of process mean and performance of control chart were both investigated systemically. The main content and innovational points in this thesis are as following:1. Evaluate and analyze the process adjustment methods originated from the fields of EPC and SPC, including the MMSE adjustment and EWMA adjustment which can eliminate autocorrelation of data, and Grubbs harmonic rules which can decrease the process set error. First, we analyze and simulate Deming's funnel so as to illuminate the need for process adjustment under the circumstance of modem manufacturing industry. As for the AR(2) and IMA(1,1)process, we study the performance of MMSE and EWMA adjustment methods in process under control. As for the AR(2) process, we study the relationship of related correlation coefficient, the value ofλ, with variability reduction after the adjustment of EWMA. These results have shown that although the optimizedλvalue could help achieving the minimized MSE, the effect of EWMA adjustment only was dependent on auto-correlated process. Our simulating data proved the adjustment advantage of Grubbs harmonic rule, and defined a relative performance ratio to compare the adjustment effects of EWMA method and Grubbs rules. When the process initial error is larger, Grubbs rule was better in application than EWMA method; but when the process initial error is relatively smaller, no obvious difference was observed between them. The differences and similarities between MMSE adjustment and Grubbs rules are compared from the model, manner, and background of adjustment methods based on state space model.2. As for the AR(2) process, we study the performance of control charts with MMSE adjustment on output and input. For auto-correlated process, the integration of MMSE and control chart to monitor the changes of process mean is an effective method for quality improvement. We discussed the reason for direct failure of control chart to monitor AR(2) process. And the ARL results simulating the control chart has shown the auto-correlated process has yielded more strong effect on EWMA chart and CUNSUN chart than Shewhart control chart. The shift model of process mean in MMSE adjustment was studied; we also investigated the signal-to-noises ratio, the alarm rates, ARL, in order to analyze the relationship of operating length, process mean shift model, and the auto-correlated. These results showed that as for the highly positive correlated AR(2) process, the performance of control chart becomes worse for the small, medium size of mean shift. Therefore we brought forward the view of combination monitoring the output and the input, i.e, after analyzing and comparing a transient and a steady state signal-to-noise ratios and simulation of ARL, the performance of monitoring charts of the input and the output can compensate each other.3. As for IMA(1,1) process, we studied the performance of adjusting output of MMSE method monitored by control chart. Firstly, we analyzed the performance of Shewhart chart monitoring shift and draft of process mean based on the parameters of signal-to-noise ratio, ARL, and alarm rates. We established the CUSCORE chart, and demonstrated the ARL simulation results of it and other three commonly-used control charts. Our results showed that for CUSCORE chart maybe be a better choice for the mean drift of IMA(1,1) process.In summary, our study expanded the research scope of SPC, and the results provided a new idea and reference evidence for the academy research and production practice. We hoped these conclusions could be valuable for the quality improvement of our manufacturing industry.