Choice of model structure for controller design: A case study

A parametric model for a laboratory scale ventilation system is identified. Based on the model, digital controllers are developed using a standard structure for pole-placement design. The performance of the controllers is shown to be crucially dependent on the number of parameters in the identified model: Even though no significant difference can be perceived in the ability of the identified models to describe the underlying system, similarly designed controllers will either perform well based on a statistically worse model or fail based on a better model. Some implications of this surprising observation for the entire control design process are discussed.