A framework for robust control based model invalidation

In this paper, the authors establish a connection between robust control design and model validation. In particular, they present a new framework for invalidating a set of models which has been assumed to "adequately describe" the unknown underlying process. The technique uses robust control synthesis and observations of a new set of variables, rather than the usual inputs and outputs of the process, to invalidate models. In contrast to existing techniques for model validation where the input/output experiments are specified a priori, an important contribution of this scheme is that it automatically generates plant inputs which are necessary to test the model with respect to the goal of trying to achieve some desired performance for the unknown process. Another important byproduct is that if the set is not invalidated (meaning it may adequately describe the process) the authors have a controller that, so far, achieves the desired closed loop performance for the unknown process.