Tom Oomen


Data-Driven Control

As is argued on the identification for control page, typically the only purpose of our models is subsequent control design. One could argue that this is a lot of effort, and one could wonder whether we cannot skip this intermediate step of modeling. In our research, we have worked on several algorithms to address this, which are typically called “data-based control” or “data-driven control”.

Iterative Feedback Tuning (IFT)

Iterative Feedback Tuning (IFT) is an algorithm developed in the 1990s for data-driven controller tuning with pre-specified structure. Basically, the need for a model is replaced by a dedicated experiment on the true system, which essentially generates the gradient of the control criterion w.r.t. the controller parameters. In fact, this algorithm was one of my first encounters with the field of automatic control, and this research is published as the following abstract.

In later research, we have applied some of the insights obtained in this work (i.e., adding additional constraints based like the modulus margin) towards an industrial wafer stage system. The results are reported in the following paper.

Recently, we have continued the development of IFT algorithms towards so-called LPV systems. The results are reported in the following article.

Virtual Reference Feedback Tuning (VRFT)

An alternative method to design controllers in a data-driven manner is through an algorithm called Virtual Reference Feedback Tuning (VRFT). We recently developed a new multivariable algorithm for VRFT. The results are reported in the following paper.

Data-driven learning of the cal{H}_infty-norm

In the context of data-driven control, we have recently been investigating iterative data-driven algorithms for learning the mathcal{H}_infty norm of multivariable systems. The main motivation for doing this originates from uncertainty modeling for robust control, where an accurate bound on the model uncertainty is required. Results of this research are reported in

Iterative Learning Control (ILC)

Finally, Iterative Learning Control is one of our research topics. Standard ILC requires some model. Interestingly, we have been able to fully abandon this model. This is reported in

An alternative approach to eliminate this model knowledge is described in advanced feedforward page, under the IV-based technique. However, from our perspective this is really a model-based/identification-based approach: one basically directly estimates a system inverse!

Acknowledgement

The above results are in collaboration with many co-workers, both from TU/e-ME, industry, and international universities.

Note that all figures shown on this page can be found in the mentioned papers. Please follow the guidelines regarding copyright and references when citing these.