Implementation of optimization-based controllers in industrial processes

This thesis presents and discusses the possibility of the use of online model predictive control as the control strategy, embedded in an educational device from National Instruments. The work considers the use of different optimization solvers and code generators, which have come out in recent years by the academia, to control a four-tank pilot plant located in the Institut de Robòtica i Informàtica Industrial, CSIC-UPC. The mathematical model of the system, used to implement the optimization-based controllers, includes nonlinear dynamics, whereby two approaches have been discussed and implemented: linearizing the model around an operating point to design a linear predictive controller, and using the original nonlinear dynamics to design a nonlinear predictive controller. The experiments performed for the case studied shown the suitability of the obtained model to be used in the linear case, while less accuracy is obtained in the nonlinear case due to the approximation of the dynamics previously reported in the literature. Moreover, a possible implementation and embedding procedure for the designed controllers are both presented and discussed in detail. For the given implementation, an assessment between the solvers embedded in the real-time device is done, giving an evaluation of the suitability of each solver in the considered scenarios. Finally, this work opens different ways of future work not only in the adaptation of the dynamic model of the case studied, but also in the optimization of the software codes generated to design the real-time controllers.