About the teaching
About the PhD project
Water and resource recovery processes are under enormous pressure to handle increasing uncertainties related to spatio-temporal evolutions of source quality, new technological developments, changing regulatory constraints etc. Process control strategies help utilities and industries to manage their processes under these different pressures. For water and resource recovery systems (and many other biotech systems), process control practice is ruled by PID controllers which are based on simple input-output relations. The use of more advanced control algorithms that allow simultaneous control of multiple inputs and outputs as well as the integration of quality and cost criteria could significantly help utilities and industries to become more resilient and sustainable.
Model-based predictive control (MPC) is a powerful control algorithm where the controller uses a model to anticipate the system’s reaction over a certain prediction horizon. However, the application of MPC in the water and resource recovery industry is still in its infancy due to the strong non-linearity of the underlying processes which is either not sufficiently captured by biokinetic process models or creates a high computational burden on the control algorithm. MPC algorithms based on data-driven models (for example artificial neural networks) are becoming more popular since they allow for significant flexibility to deal with non-linearities. However, data-driven MPC controllers come with important limitations with respect to their interpretability and their ability to handle unforeseen circumstances (shock load, first flush, changes in operating conditions…). Hence, the choice of the underlying model in MPC algorithms is very important for its successful application.
This PhD project aims to develop MPC algorithms for complex non-linear systems in the water and resource recovery sector in order to operate these systems at reduced energy and resource consumption without compromising the water or product quality. As a first milestone, MPC algorithms will be developed based on models of different complexity (biokinetic models, fully data-driven models and combinations of both) and a comparative benchmark analysis will be made with respect to disturbance rejection, set point tracking, feasibility of implementation, data requirements and performance assessment under unforeseen circumstances. In a next step of the project, the developed control strategies will be applied to a full-scale cases and more complex control targets.
Prof. dr. ir. Elena Torfs