The main theme of this thesis is the design of optimization-based iterative learning control (ILC) with guaranteed stability for constrained batch processes. By combining information from previous cycles together with knowledge of a system model, it is known that optimization-based ILC is a good tool for the control of batch processes. It has a fast convergent rate and good ability to handle system constraints. However, stability of the systems controlled by optimization-based ILC has been a critical issue. For many years, the system stability could only be guaranteed by tuning the controller parameters based on trial and error. Experiments which are expensive and time-consuming need to be conducted many cycles over to verify the system stability.

The objective of this thesis i...[ Read more ]

The main theme of this thesis is the design of optimization-based iterative learning control (ILC) with guaranteed stability for constrained batch processes. By combining information from previous cycles together with knowledge of a system model, it is known that optimization-based ILC is a good tool for the control of batch processes. It has a fast convergent rate and good ability to handle system constraints. However, stability of the systems controlled by optimization-based ILC has been a critical issue. For many years, the system stability could only be guaranteed by tuning the controller parameters based on trial and error. Experiments which are expensive and time-consuming need to be conducted many cycles over to verify the system stability.

The objective of this thesis is to provide systematic ways for controller design, such that the system stability is theoretically guaranteed and the parameter tuning is avoided. The thesis is composed of two parts. The first part studies how to combine a model predictive control with an iterative learning control in the local time domain. Here, a batch process controlled by ILC is formulated as a two-time-dimensional (2D) system. Stable model predictive control algorithms are designed based on the 2D system. Three types of invariant sets are developed to ensure recursive feasibility.

The second part looks into how to design optimization-based ILC in the trial domain. For stable systems, an adaptive robust ILC is designed based on set-membership identification. For unstable systems with non-repetitive disturbances, a new framework to combine ILC with a state feedback controller is proposed. This framework is then applied to control batch processes with stochastic disturbances and unknown nonlinear terms. All of the proposed methods can theoretically guarantee the system stability and constraint fulfillments. Numerical simulations are conducted to verify these theoretical results.