Design of an Adaptive Self-Tuning Smith Predictor for a Time Varying Water Treatment Process

Abstract Abstract This paper presents the simulation and real time implementation of an adaptive predictive PI controller for the control of Chlorine dosing in secondary water disinfection rigs. This trial is part of a project that looks at the optimisation of process control specifically in the water industry. As one of the main treatment processes, Chlorine dosing is one of the processes that naturally imposes very long dead times for the controller to deal with. Although PI controllers are still commonly used for controlling this process, previous literature as well as trials carried as part of this project proved that the performance of PI controllers, no matter how tuned they are, is very sluggish and unreliable. A pilot rig was used instead of a live secondary disinfection rig, and a number of open loop step tests were performed for system identification. Once the process dynamics became known, complementary functions that estimate the process transfer function based on the water flow were introduced. A standard tuned PI controller configuration was simulated for the process, and then a Smith predictor was used in order to be able to compare the performance of the predictive PI with a standard PI controller. Tuning functions were derived for the PI to make it a self-tuning predictive controller, and parameter estimation functions were also used so that the final outcome is an adaptive self-tuning system. This system was then implemented on the same pilot rig, and real time implementation proved the findings obtained from the simulation. Both simulation and pilot rig tests show a very good dynamic response with excellent accuracy.