As typically practiced today, PID tuning is performed on a single-loop basis, via trial and error, or based on the results of an open-loop step test. For interacting loops, this is a time-consuming task that often results in compromised performance, or at least leaves open the

questions: "How much performance is being left behind?" and "How robust is the tuning?"

This contribution presents a multivariable approach to PID tuning, which takes advantage of modern identification methods and incorporates a control-relevant optimization to determine optimal PID tuning parameters. A key feature of the approach is that robustness margins to dead time and gain errors are explicitly addressed.

The approach is illustrated on an industrial example.