Controller design from data under UBB noise

In this paper, a controller tuning methodology for unknown linear systems is proposed. The approach requires a set of experimental data generated by the plant for a single experiment. For a controller structure parametrized as a linear combination of basis functions, the procedure allows to find a feasible set of parameters compatible with a given performance criterion expressed as a desired complementary sensitivity function. The algorithm deals with unknown but bounded noises and does not require any statistical hypotheses on disturbances nor open loop operation to capture the experimental data. The performance of the procedure is illustrated by numerical examples. © 2014 American Automatic Control Council.