A sensitivity-based approach for the control of repetitive processes

For a large number of technical processes, it is desirable to design control strategies which allow for tracking desired state or output profiles which are repeated periodically. Such tasks are commonly solved by means of iterative learning control strategies as well as by the concept of repetitive control. Most of these before-mentioned techniques are designed in such a way that the linearity of the underlying system model is exploited. If a dynamic system is nonlinear, techniques for gain scheduling, corresponding to an online adaptation of a quasi-linear system model, are commonly applied. However, such adaptation strategies, depending on state measurements or estimated variables, have to be derived specifically for each problem at hand. Therefore, a sensitivity-based control approach is presented in this paper that can be employed for tracking control of both linear and nonlinear dynamic systems in spite of non-modeled disturbances. This control strategy makes use of a real-time capable sensitivity analysis of dynamic system models and comprises aspects of model-predictive and iterative learning control. The applicability of the corresponding algorithm is demonstrated in simulation and experiment for a distributed heating system.