Optimal control of diesel engines

This research aims to develop a high performance feedback control strategy that minimizes fuel consumption of diesel engines, while satisfying stringent emission legislation, by using in-cylinder pressure information. To meet this goal both in test-cycles as well as under real-world conditions, adaptive control and online optimization are explored. 

PhD Candidate: ir. Robert van der Weijst
Supervisor: prof.dr.ir. Frank Willems, dr.ir. Thijs van Keulen
Promotor: prof.dr.ir. Maarten Steinbuch
Project Financing: Impuls 2
Project Period: April 2015 – April 2019

Control of diesel engines deals with the challenge of achieving minimal operational cost while satisfying a certain torque demand and emission legislation. Typically, the solution to this optimal control problem consists out of (1) a supervisory control system which prescribes engine-out emission levels for (2) the engine control system, and (3) control of the exhaust gas after-treatment system (EAS), which reduces engine-out emission levels to tailpipe limits. Both in the engine and the EAS, emission reduction leads to increased operational costs.

This project focusses on the engine control system, to be precise, accurate tracking of torque and engine-out emission levels are desired, with minimal fuel consumption. In-cylinder pressure sensors, which are not yet common in industry, provide information about the combustion process, including actual heat release and high bandwidth estimation of engine out emissions. This enables improved closed-loop control and an accurate online estimate of the performance to-be-optimized.

Our objective is to optimize measured engine performance online. Opposed to the state-of-the-art solutions, online optimization is robust to disturbances and varying operating conditions, e.g., ambient conditions or fuel quality. As such, this is a key factor in realizing minimal fuel consumption in real-world operation. Due to the complex dynamics of the considered system, model-free optimization using extremum-seeking is an appealing starting point. However, the potential of adding (parametric) system knowledge to the optimization algorithms is also considered.