People Education Research Industrial Agenda  
Overview MSc info MSc program MSc topics ET TN WBMT  


Towards data-driven autonomous control systems in the process industry

Subject: Towards data-driven autonomous control systems in the process industry
Staff Mentor: X.J.A. Bombois
Other Mentor(s): Max Potters (DCSC) and Paul Van den Hof (DCSC)
Keywords: System identification and estimation; Industrial processes; Fundamentals; Optimal and model predictive control
Description: The process industry - businesses that deal with extracting, transporting and processing raw materials, including their physical, chemical or mechanical transformation - plays an important role in the European economy. However the industry in Europe is facing many crucial challenges that are forcing it to improve quality, reduce operating costs and raise environmental standards all at the same time. To achieve this, model-based control systems are expected to play a huge role.

The main limitation currently preventing model-based control from ensuring plants operate optimally is the reliability of the models themselves. However, current approaches for developing accurate models involve conducting experiments on the plants, which is hugely expensive. As a result, control laws are not often updated - if ever - even though plant dynamics change over time. This decreases the quality of the model and, in turn, of the control laws.

To solve this problem, we propose to develop a self-regulating, autonomous control system. This system will continuously monitor the control loop, detect any drop in performance, and use the least costly experiments to update the model and adapt the control laws. To control costs even more tightly, the system will only start this updating procedure if it is assessed as being economically attractive.

We plan to develop procedures for designing experiments that generate high-quality data at low cost. The experiments concerned are 1) an experiment allowing one to develop a sufficiently accurate model for the design of a satisfactory control law, and 2) an experiment for reliably detecting whether changes in the system have caused the drop in performance. In both cases, the cost of the experiment is dependent on its duration and how much it disturbs the normal operations of the plant.

We are looking for MSc students who would be interested to contribute to one of the aspects in this project. This project is an European KP7 STREP project in collaboration with academic partners (KTH, Aachen, Eindhoven) and industrial partners (ABB, Boliden and Sasol).

Back to list

Last modified: 14 March 2012, 12:25 UTC
Search   Site map