This contribution presents a method for modeling and controlling a heavy-duty biturbocharged diesel engine. The modeling scheme can incorporate expert knowledge of the control relevant combustion quantities into Gaussian process models. A nonlinear model predictive controller (MPC) is used to control the engine outputs subject to the gas path dynamics and nonlinear constraints for the emissions and for the sake of engine protection. In addition, an online learning scheme based on Gaussian process regression is used to compensate for model uncertainties due to aging effects and manufacturing tolerances. A consistent model smoothing strategy is derived to preserve the given expert knowledge and to avoid abrupt reactions of the MPC due to the online learning of the models. All parts of the controller are implemented with respect to real-time feasibility and small memory footprint. Experimental results for a real-world heavy-duty engine demonstrate the performance and the online learning ability of the presented nonlinear MPC scheme that may be transferred to various diesel engine applications.

In this article, the problem of minimizing the overall operational cost of a heavy-duty off-highway diesel engine combined with a selective catalytic reduction (SCR) catalyst is considered. Here, we propose a hierarchical model-based scheme described as an optimal control problem. The sequence of resulting optimal control values are setpoints for the underlying engine controller, applied in a model predictive control (MPC) fashion. The presented scheme meets several demands. While minimizing the overall costs, it handles box constraints for the control variables as well as a nonlinear NO R -conversion rate constraint ensuring that a given emission target is met. The approach makes use of Gaussian process models for the input-output behavior of the underlying components and a technique for online adaptation. Thus, the presented hierarchical scheme is able to compensate model uncertainties and aging effects of engine, air path, and SCR catalyst. Moreover, in comparison to the literature, our approach doesn't require detailed models of the underlying components, and the hierarchical, modular design allows the applicability to different engines and SCR controllers. We illustrate the proposed approach by several simulation results.

This paper presents a safe optimization method for minimizing an unknown cost function subject to unknown inequality and equality constraints. The cost function as well as the constraints evaluation may be corrupted with Gaussian measurement noise with known uncertainty. The focus especially lies in the safe exploration of the cost function, which means that an evaluation of the cost function far away of previous evaluations is not favoured. Also evaluations in regions where the constraints are violated are undesired. This is the case, for example, in technical applications, where systems may become unstable or damaged, if constraints are violated. To this end, a combination of the expected improvement of the cost and its mean is minimized, while bounding the variance of the cost. The inequality and equality constraints are reformulated as constraints for the probability of the constraint violation. The optimization method is evaluated on numerical examples.

Zusammenfassung Datenbasierte Modellbildungsverfahren haben auch für komplexe und hochdimensionale Systeme an Bedeutung gewonnen. Sie bieten den Vorteil einer deutlich verkürzten Entwicklungszeit im Vergleich zur physikalischen Modellierung. Der Nachteil vieler herkömmlicher Verfahren ist allerdings, dass selbst einfaches Wissen, wie etwa Extrapolationseigenschaften, nicht in der Modellgenerierung berücksichtigt werden können. Deshalb soll in diesem Beitrag eine Methodik vorgestellt werden, wie auf Basis der Gaußprozessregression solches Wissen auch für datenbasierte Verfahren genutzt werden kann. Desweiteren ist es in vielen physikalischen Systemen notwendig, dass die Modelle während der Laufzeit an individuell unterschiedliche Systeme angepasst werden können. Dazu wird ein Online-Adaptionsverfahren vorgestellt, welches die Unsicherheitsinformation des Gaußprozesses nutzt, um Änderungen im System zu detektieren und das Modell anzupassen.

In this work we present a hierarchical optimization scheme to minimize the overall operational costs of a heavy-duty diesel engine combined with a selective catalytic reduction (SCR) catalyst. The optimization variables of the hierarchical optimization scheme are setpoints of the underlying engine controller. The presented hierarchical optimization scheme follows a modular approach with the capability of constraint integration and inclusion of additional subsystems. Moreover, to reduce the computational load, a learning method is given to adapt the solution of the hierarchical optimization scheme. Simulation results illustrate the proposed approach.

This paper presents a method for modelling and controlling charge air pressure of a heavy-duty Diesel engine. The modelling method combines expert information about the system with a training dataset to provide a reasonable extrapolation behaviour as well as high accuracy. The dynamics of the gas path are modelled as a Wiener process with separate dynamics for each input. After identifying these dynamics, a controller based on input-output linearisation can be designed. The controller is shown to be able to track a set point of the gas path even after a sensor failure. To this end, the system behaviour is learned online to increase the accuracy. An evaluation against a reference model shows the performance of the method.

This paper presents a method for nonlinear system identification with Gaussian process regression. The unsupervised method is able to generate an approximation of the system with correct extrapolation behaviour, that is refined with input/output-data in the typical working area and sampled online data. Therefore, an offline model is generated, which consists of a nominal model set up by the extrapolation behaviour and a detailed model for the refinement. The method is able to keep track of time-varying systems by using the confidence information to incorporate new measurements into the online model. The performance of the proposed method is tested on different numerical examples.