In this paper, unfalsified control theory (a data-driven model-free control theory) is applied to determine which control parameter sets in a specified control structure are able to meet a given performance specification, using merely measured input/output data. The need for a finite, often large, amount of parameter sets (gridding) is overcome by applying an ellipsoidal description of the region containing all unfalsified control parameter sets (unfalsified region). It is shown that by using an appropriate performance specification, the optimal update of the ellipsoidal unfalsified region, initiated by new data, can be computed analytically.With the two properties mentioned, improved convergence and reduction of computational effort are combined to derive the unfalsified control parameter set. Real-time implementation is demonstrated by experimental results obtained on a motion system.
|Title of host publication||Proceedings of the 16th IFAC World Congress, 4-8 July 2005, Prague, Czech|
|Publication status||Published - 2005|