Research Area 2: Model adaptation and model quality monitoring

The quality of the results of model-based optimization strongly depends on the accuracy of the models that are employed. It is essential that the predictions of variables that are considered in the constraints and the cost function of the optimization, e.g. product quality parameters are accurate. Convergence to the optimum depends on the precise knowledge of the gradients of the cost function and of the constraints with respect to the degrees of freedom, and the speed of convergence depends on the accuracy of second order terms. In model-based control algorithms, the offset between the predictions and the measurement is usually taken into account to correct the target values in an ad-hoc manner, but no information about derivatives is used. The quality of the models can be increased by online adaptation of crucial parameters by robust state and parameter estimation schemes. This introduces the problem of “dual control”: gaining information about the process may make other control moves preferable than optimal operation. We will pursue two related research directions here: Parametric model improvement by online parameter estimation and model augmentation by estimated nonparametric response surface models. In both cases, the issue of dual control arises: some control moves may be advantageous for the improvement of the model accuracy because they generate information-rich data, but not optimal for the performance of the controlled plant as predicted by the nominal model on the prediction horizon. On a longer horizon, however, the improvement by a better model quality may outweigh the losses incurred by the “data collecting” moves. The research on model adaptation will contribute to the reduction of the cost and effort of modeling, as we expect that with parameter adaptation less complex rigorous models can be used in the control algorithms.