On the parameterisation of simple process models for the autotuning of industrial regulators

This manuscript deals with the problem of identifying process models in the particular application context of model-based autotuning. First the work illustrates some pitfalls caused by the use of conventional identification methods in that context, where far from ideal identification conditions are experienced. Then, the effects of the interplay between identification and control are examined, and it is shown that unsatisfactory control results may be obtained, if the two tasks are treated independently. In view of this a specific formulation for the identification problem in the autotuning context is given, which requires that the identification objective be formulated in terms of the control objective. Some guidelines are provided for the design of identification techniques suited for autotuning applications, and an integrated identification/tuning technique is presented, based on these guidelines. Some examples are finally reported, to show the achievable improvements.