Towards automatic real-time controller tuning and robustness

In some industrial applications, real-time controller tuning is a time-consuming exercise; it can be costly and difficult when there are a multitude of control loops acting on a number of processes and where the complexity of the control problem requires robust and effective controller design. An automated tuning procedure could speed up this activity, but the tuning method should have good properties. This paper investigates whether certain families or groups of cost functions can provide an enhanced degree of robust controller design. An optimisation algorithm was developed to provide the optimal controller parameters using a number of integral-error cost functions. The robustness of each cost function was assessed using standard and parametric stability margins. For the class of systems investigated, it seems that exponentially- or time-weighted cost functions can consistently give more robust controller design.