Venkatesan, Gopalachary (1997) A Statistical Approach to Automatic Process Control (regulation schemes). PhD thesis, Victoria University.

Abstract

Automatic process control (APC) techniques have been applied to process variables such as feed rate, temperature, pressure, viscosity, and to product quality variables as well. Conventional practices of engineering control use the potential for step changes to justify an integral term in the controller algorithm to give (long-run) compensation for a shift in the mean of a product quality variable. Application of techniques from the fields of time series analysis and stochastic control to tackle product quality control problems is also common. The focus of this thesis is on the issues of process delay ('dead time') and dynamics ('inertia') which provides opportunity to utilise technologies from both statistical process control (SPC) and APC. A presentation of the application of techniques from both SPC and APC is made in an approach to control the quality of a product (product variability) at the output. The thesis considers the issues of process control in situations where some form of feedback control is necessary and yet where stability in the feedback control loop cannot be easily attained. 'Disturbances' afflict a process control system which together with issues of dynamics and dead time (time delay), compound the control problem. An explanation of proportional, integral and derivative (PID) controllers, time series controllers, minimum variance (mean square error) control and MMSE (minimum mean square error) controllers is given after a literature review of stochastic process control and 'dead-time compensation' methods. The dynamic relationship between (output) controlled and (input) manipulative variables is described by a second-order dynamic model (transfer function) as also is the process dead time. The use of an ARIMA (0,l,l) stochastic time series model characterizes and forecasts the drifting behaviour of process disturbances. A feedback control algorithm is developed which minimizes the variance of the output controlled variable by making an adjustment at every sample point that exactly compensates for the forecasted disturbance. An expression is derived for the input control adjustment required that will exactly cancel the output deviation by imposing feed back control stability conditions. The (dead-time) simulation of the stochastic feedback control algorithm and EWMA process control are critiqued. The feedback control algorithm is simulated to find the CESTDDVN (control error standard deviation) or control error sigma (product variability) and the adjustment frequency of the time series controller. An analysis of the time series controller performance results and discussion follow the simulation. Time series controller performance is discussed and an outline of a process regulation scheme given. The thesis enhances some of the methodologies that have been recently suggested in the literature on integrating SPC and APC and concludes with details of some suggestions for further research. Solutions to the problems of statistical process monitoring and feedback control adjustment connected with feedback (closed loop) stability, controller limitations and adequate compensation of dead time in achieving minimum variance control. are found by the application of both process control techniques. By considering the dynamic behaviour of the process and by manipulating the inputs during non-stationary conditions, dynamic optimization is achieved. The IMA parameter, suggested as an on-line tuning parameter to compensate dead time, leads to adaptive (self-tuning) control. It is demonstrated that the performance of the time series controller is superior to that of the EWMA and CUSUM controllers and provides minimum variance control even in the face of dead time and dynamics. Some articles/papers have appeared in Technometrics, Volume 34, No.3, 1992, in relation to statistical process monitoring and feedback adjustment (25l-267), ASPC (286-297), and discourse given on integrating SPC and APC (268-285). By exploiting the time series controller's one-step ahead forecasting feature and considering closed-loop (feedback) stability and dead-time compensation, this thesis adds further to these contributions.

Search Google Scholar

Repository staff login