Date of Award
Master of Engineering (Research)
Dr. Tom O' Mahony
This thesis attempts to determine whether advanced control algorithms offer any real benefits to the process industry in terms of optimising single-input single output process units. To establish this, an ideal PID controller, Two Degree of Freedom (2 DOF) PID Controller and the Generalised Predictive Controller (GPC) are compared. A common design philosophy is applied based on optimising performance subject to constraints on robustness. Specifically, three different designs are examined; minimum subject to constraints on the gain and phase margin, minimum lAE subject to constraints on the modulus margin and minimum lAE subject to constraints on the input sensitivity function. Each of these optimisation problems are solved using a genetic algorithm.
The control algorithms and design methodologies axe evaluated in simulation using a range of thirteen benchmark .systems common in the process industry. Subsequently, the controllers and design strategies are evaluated on a real time laboratory scale process a flexible link. The results indicate that the GPC algorithm performs notably better than the other two controllers on the benchmark simulation study. However, when the GP was applied to the flexible link the benefits are not so obvious. The 2 DOF PID controller achieves a really good trade-off between performance, robustness and ease of-tuning. These issues are undoubtedly one of the reasons for the success of the structure in practice and the name Industrial PID controller is well deserved.
While each of the three proposed controller design techniques worked well, the design based on the input sensitivity function performed best. This is especially noticeable on the real-time system as the nature of the technique ensures that the high-frequency gain of the controller is adequately shaped and resulting in good immunity to high-frequency noise.
The novelty in this work primarily arises from the problem domain studied anti the proposed controller design methodologies. This is particularly true for the designs that minimised the lAE subject to constraints on the modulus margin and input .sensitivity function. Additional novelty arises from the nature of the penalty functions to avoid constraint violation and the modifications that were made to the canonical genetic algorithm to reduce the computational time.
Czarkowski, Daniel, "An Evaluation of Model-Based Predictive Control" (2005). Theses [online].
Available at: https://sword.cit.ie/allthe/162