Effects Of Several Autocorrelation Process Control Charts

If any company wants to keep advantage in competition, what they should do is to control and improve the quality of products. Statistical process control is the most commonly used method for quality control in industrial production process, which is mainly monitored quality characteristic by used control chart to ensure the production process is in control. Choosing appropriate control charts is very important to improve the quality of products and reduce the cost. The commonly used control charts, such as traditional control charts, CUSUM control charts and EWMA control charts, are constrained by the assumption that the quality characteristic data is independent and identical distributed. However, in the practical production process, the observed data are usually auto-correlative. For the production process that the observations are auto-correlative, if we still use the traditional control chart to monitor the process, there would produce a large number of false alarms, which lost the meaning of control chart itself. Therefore we need to eliminate auto-correlation of the quality characteristics data, then use traditional control charts to monitor these data; or we can adjust the control limits of the control chart, then directly monitor the auto-correlation data. They are two feasible way for monitoring auto-correlated processes.In this paper, we mainly discuss the problems of monitoring the auto-correlated processes, and try to complete the theory and method of statistical process control. Specific work is as follows:(1) We expound the principle of conventional control chart and the method of how to determine the parameters. The research is only aimed at the state auto-correlated processes. We use AR(1) model to fitting the observed data that have a shift at a certain time point, if use the traditional control charts to monitor the data directly, the control charts would give false alarms, and it shows that traditional control charts do not apply to monitoring auto-correlated processes.(2) The residual is obtained by fitting the time series model, to eliminate the auto-correlation, use residual as the statistic and get the control charts. The results of the numerical experiment show that the residual control chart, the residual CUSUM control chart and the residual EWMA control chart have some effects on the detection of the shift in the process, but calculating residual is more complicated.(3) To solve the above problems, this paper gives out the improved EWMA control chart, by estimating the variance of the statistics to change the control limit, then monitoring the data directly, and the effect of the detect the shift is improved.(4) Considering the possible mean shift of the noise terms, proposed an improved CUSUM control chart by adjusting the decision parameters of the control chart, and the detection of the process mean shift is more accurate. By comparing with the residual control chart in the numerical experiment, the results show that the improved control chart have better effect in monitoring the auto-correlated processes.