Control relevant model selection for uncertain processes

In process identification, it is common practice to seek the model that minimizes the variance of the data from the model. If the process parameters actually vary over a range, the authors show that choosing a model in the middle of the range is not generally best from a control perspective. The authors use M_p-synthesis (Laiseca and Brosilow, 1992) to obtain models for optimal control system performance for three uncertain processes: 1) a second order process with uncertainty in time constant and damping ratio, 2) an overdamped second order plus dead time process and, 3) a non-collocated spring mass system.