Application of Statistical Process Control to Continuous Processes
Control charts represent an efficient and easy tool to assure the state of statistical quality control in a manufacturing process. These tools are also implemented in continuous processes, where the critical parameters are often monitored by on line senso
- PDF / 1,034,790 Bytes
- 8 Pages / 481.89 x 691.654 pts Page_size
- 80 Downloads / 212 Views
2
Dipartimento di Tecnologia e Produzione Meccanica, University ofPalermo, Italy Dipartimento di Ingegneria Iudustriale e Meccanica, University of Catania, Italy
KEYWORDS Control charts, Autocorrelation, Average Run Length. ABSTRACT. Control charts represent an efficient and easy tool to assure the state of statistical quality control in a manufacturing process. These tools are also implemented in continuous processes, where the critical parameters are often monitored by on line sensors measuring data with short time intervals. In this paper a continuous process is monitored by using control charts and its dynamic is modeled through linear time series that allow the effects ofthe autocorrelation tobe eliminated. In this way, the control charts can operate on residuals that result identically and independently distributed. A statistical analysis on EWMA, CUSUM and control charts for individual measurements has been carried out to select the most performing tool for process monitoring.
1 INTRODUCTION Statistical quality control tools are widespread in industry to achieve the control of critical process parameters. Nowadays, several industries implement these instruments to prevent sudden process failures, which can affect the quality of the output of their processes. In particular, control charts are a simply and powerful tool to plot the behaviour over time of some critical process parameters. Among them, the Shewhart control charts, the cumulative sum control charts (CUSUM) and the exponentially moving average (EWMA) control charts are easy to be implemented and interpreted. One of the basic assumptions for a correct application of these charts is that the collected data are independently and identically distributed (liD), [1]. Butthis hypothesis cannot be always verified. Unfortunately, for continuous processes, i.e. chemical processes and all manufacturing processes driven by inertial elements, the observations on the process are correlated over time. Therefore, the application of conventional control charts on these data can Iead to misleading results in the form of excessive false alarms due to the data correlation. Alwan and Roberts [2] have developed an extension of traditional process control procedures that allows the effect of the systematic behaviour of the process to be separated from the variability depending on the special causes. The time series modeHing of the collected data is based on the ARIMA models. Once a correct model has been implemented, it is possible to determine the residuals of the measurements, evaluated as difference between the real data and the fitted model. Then, a common cause chart on the fitted data is used to give a view of the level of the process and of its evolution through time. The state of statistical control of the process is evaluated by plotting the residuals on a Shewhart control chart or eventually on a chart for individual measurements. EWMA and CUSUM control charts applied on the residuals have also been proposed by [3,4,5,6] for the monitoring of a manufacturing
Data Loading...
