A flexible algorithm for structuring and reducing model error bounds

We propose a way of bounding model uncertainty. The method incorporates noisy measurements of dynamical systems. Model uncertainty bounds are formalized in terms of both a structured and an unstructured part. No prior structuring of the uncertainty is needed. The uncertainty description is compatible with μ controller design. Our approach is based on the calculation of dominating components of the model uncertainty for a process operating in multiple working points. Experimental data is required for each of the working points. These dominating components are bounded separately. It is shown that the remaining unstructured uncertainty becomes much smaller in this way. Several kinds of prior knowledge can be handled and are briefly discussed