Author:

Abstract:

Repetitive processes are widespread in our world. Through repetition performance can be improved. A pianist practices a piece of music until the desired performance is obtained. Gymnasts practice the same exercise many times in training. In manufacturing industries, repetitive processes are ubiquitous as well because of the repetitive nature of mass production. For these industrial systems, performance can be improved through repetition by means of iterative learning control (ILC). ILC is a control technique that aims at iteratively improving the performance of systems by exploiting data from previous repetitions. This thesis develops ILC analysis and design methodologies for linear time invariant systems that allow the user to account for model uncertainty, and trade off multiple objectives such as learning speed, input constraints and accuracy. First, a robust constrained norm-optimal ILC design in time domain is proposed. Contrary to conventional norm-optimal ILC, this approach incorporates model uncertainty and input constraints into the cost function. The worst-case cost function is then optimized. Second, a novel multi-objective ILC design in frequency domain is presented. This approach is novel because it realizes an optimal trade-off between different ILC objectives: robustness, tracking performance, learning speed, and input constraints. In addition, both the Q-filter and the learning function are optimized simultaneously hereby eliminating the need for the common two-stages ILC design procedures in frequency domain. Third, a study of ILC analysis and synthesis in time-domain using the lifted system representation is discussed. The presented method approaches the ILC analyses in a more systematic way than existing approaches since it accounts for both robust convergence and robust performance, and with both unstructured and structured system uncertainty. The proposed controller designs in this thesis are formulated as convex optimization problems, guaranteeing global and efficient optimization solutions. The proposed design methods are extensively validated experimentally on a lab scale overhead crane system. Experimental comparisons with existing iterative learning control designs are given. The results show that the proposed methodologies outperform the major current robust ILC approaches. They guarantee not only robustness but also high tracking performance while satisfying input constraints. Moreover, the desired control performance can be selected accounting for multiple ILC objectives and their trade-offs. Practical guidelines are provided to aid control engineers in selecting the suitable ILC approach.