Convergent iterative model unfalsification for control

Iterative controller design by model unfalsification and robust controller design is reexamined and developed. Two frequency dependent uncertainty functions are used to ease the process of modelling for control. One is the maximum allowed gain errors (AGE) of a nominal plant which depends on the controller, nominal model and the stability robustness requirement. The other functions used are frequency dependent unfalsified lower bounds of plant dynamical uncertainty around nominal models. The controller is called feasible if the allowed plant perturbation function bounds an estimated plant uncertainty for each frequency. If there is no feasible controller then more experiments are carried out with the same controller which excite the plant at the "offending frequencies". Otherwise the feasible controller with the largest stability margin and "most promising" performance is used in the next experiment with the testing input spectrum focusing on the frequencies where the stability margin is small and the performance is sensitive.