Two-stage model predictive control for the air path of a turbocharged gasoline engine with exhaust gas recirculation

• Zweistufige modellprädiktive Regelung für den Luftpfad eines aufgeladenen Ottomotors mit Abgasrückführung

Keller, Martin Gerhard; Abel, Dirk (Thesis advisor); Pischinger, Stefan (Thesis advisor)

Aachen : RWTH Aachen University (2021)
Dissertation / PhD Thesis

Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2021

Abstract

Turbochargers have been an industry standard for years to increase the specific power of internal combustion engines. Two-stage sequential turbocharging overcomes even the usual trade-off between a fast transient charging-pressure buildup and high specific power. However, increased knock probability and high exhaust gas temperatures at high-load operation lead to the need for enrichment and late ignition timings for component protection. These measures diminish the maximal reachable engine efficiency of gasoline engines. In order to decrease the high engine-out temperatures while maintaining the stoichiometric engine operation, exhaust gas recirculation (EGR) is in the focus of current research for spark-ignition (SI) engines. The more and more complex engine and air-path setups impose high demands on the process control. The thesis at hand proposes a development methodology for the implementation of a model predictive controller (MPC). The MPC developed controls both the desired charging pressure and the EGR rate of a two-stage turbocharged SI engine with low-pressure EGR. The two-stage turbocharging results in an overactuated system setup in which the additional degree of freedom can be used for an intelligent and efficient allocation of the charging-pressure buildup. In addition, the system has multiple inputs and multiple outputs with considerable cross-couplings and constraints. To tackle these characteristics, a two-stage MPC structure is proposed and implemented on a rapid-control-prototyping (RCP) hardware. For this purpose, the present thesis aims to investigate physics-based as well as data-driven reduced-order-modeling approaches for engine modeling and addresses the necessary adaptations towards real-time computation within online-optimization algorithms. Furthermore, different formulations of the resulting optimal control problems are analyzed and their computational burdens validated on the RCP hardware. Hence, the resulting nonlinear programs (NLPs) are solved either directly with interior-point NLP solvers or with sequential quadratic programming approaches. For the latter, sparse and condensed formulations are introduced. All the investigated solving techniques are evaluated with respect to the control performance and the achievable computing times. The two-stage control structure contains a target selector for the consideration of the system's overactuation. It calculates the optimal allocation for efficiently providing the desired charging pressure and EGR rate. In the dynamic controller, the actual optimal control inputs are calculated taking the transient system behavior into account. Promising control algorithms are tested experimentally in a prototype vehicle equipped with the aforementioned air and EGR path setup during real-driving maneuvers.