Abstract
One key challenges in the process control industry is designing controllers to maintain process stability. The most common controller used in this industry is the Proportional-Integral-Derivative (PID) controller. This form of control involves the tuning of gain coefficients.
Much has been done in the study of PID controllers. Traditionally, these gain coefficients have been tuned manually. Manual tuning of these coefficients is time consuming and can lead to poor performance. Therefore, systematic tuning methods were developed to achieve specific desired performance criterion for the system. Automated tuning software has been used both in research studies, and in a wide range of industrial applications. Its main feature is to calculate the gain coefficients for a PID controller.
A model-based systematic tuning approach was employed in this study. The approach involves testing the physical system by controlling an actuator. The system's response to the test is used by a plant identification method to generate an approximation of the plant. This plant approximation is then mapped to PID gain coefficients using PID tuning rules. The modeling and tuning can require a significant amount of storage in memory. This research creates a low-resource auto-tuning program, using a new plant-identification technique, and was paired with an existing PID tuning rule to generate PID gain coefficients. The auto-tuning program was developed to obtain these gain coefficients using efficient, low-memory algorithms that do not sacrifice performance. The algorithms use significantly fewer variables than other methods. The auto-tuning program applies statistical process control by using individuals control charts, and uses a streaming algorithm to compute the corrected sum of squares for standard deviation.
This research was carried out as a case study of the Chilled Beam Water System at the University of Virginia's Rice Hall. Focus is on the interaction between a water valve, known as the Bypass Valve, and a downstream water temperature, the Chilled Beam Supply Water temperature. This system can be characterized as a first order system with dead time. Dead time, also known as time delay, is present in any system where a signal or physical variable originating in one part of the system becomes available in another part after a lapse of time. In this system, like many HVAC systems, dead time varies. This causes the need for the PID gain coefficients to be readjusted from any initial setting in response to changing dead time. Automatic tuning provides a way to readjust these gain coefficients as needed.
Both simulation and experimentation are carried out on the Chilled Beam Water System. The auto-tuning program's plant identification technique is compared with common plant identification techniques in the literature. Many PID tuning rules exist for different performance specifications, and for different plants. This thesis examines tuning rules specified for first order with dead time systems. The plant identification techniques and PID tuning rules are compared among each other by examining the generated PID gain coefficients. The PID gain coefficients are then compared in simulation by evaluating the ability of each to recover from a disturbance.
The comparisons indicate the auto-tuning program, although using significantly fewer variables, still achieves high fidelity performance like the other plant identification methods that use more resources. Among PID tuning techniques, the Internal-Model-Control (IMC) tuning rule produced the best performance.
