Automotive Model Predictive Control

Recent years have witnessed an increased interest in model predictive control (MPC) for fast applications. At the same time, requirements on engines and vehicles in terms of emissions, consumption and safety have experienced a similar increase. MPC seems a suitable method to exploit the potentials of modern concepts and to fulfill the automotive requirements since most of them can be stated in the form of a constrained multi input multi output optimal control problem and MPC provides an approximate solution of this class of problems. In this introductory chapter, we analyze the rationale, the chances and the challenges of this approach. This chapter does not intend to review all the literature, but to give a flavor of the challenges and chances offered by this approach.

For pollutant emissions predictive physical modeling in diesel engines, three key points have to be taken into account:

This chapter takes the NOx formation/destruction process modeling as an example to illustrate the most important aspects of this kind of modeling.

The importance of simulation in power train and combustion engine development is undisputed today, but the search for the most efficient use of simulation in the development cycle is still ongoing. In parallel available computing power and the number of tools are increasing. The choice of the right tool has significant impact on the development process. In this chapter some insight will be given into a process relying on the mean value model approach for control strategy development and validation regarding the development of a 1.6-litres 4-cylindre direct injection diesel engine respecting Euro 5 legislation. To reach the performance and emission targets in a cost effective way, simulation was used early in the project to study different concepts of air loop control architecture and then was employed for the control function development of the air loop control. The present study addresses the main components of the mean value model, the air loop, the degree of refinement in regards to combustion and the actuator dynamics. Two variants of a new control concept have been studied with considerable refinement. To position the predicted performance they were compared to the conventional Euro 4 approach. To obtain significant simulation results the work was concentrated on specific engine life situations. The comparatively fast execution of mean value models has been exploited to realize statistical computations on the robustness of the controlled system. The main targets of control development are a high degree of precision but also a robust behavior in real life. To asses the performance of the control approaches under study, the main sources of disturbances du to series production were identified and than this knowledge was used to build a mean engine model that allowed to represent their impact on the engine. This allowed distinguishing the control approaches not only on their capacity to perform well on a nominal engine but their ability to cope with the large spectrum of engine behavior to be expected in series production. The obtained results allowed for an intelligent choice of the control strategy, choice which has been confirmed by experimental work.

The common theme in this chapter is physical modeling of engines and the subjects touch three topics in nonlinear engine models and parameter identification. First, a modeling methodology is described. It focuses on the gas and energy flows in engines and covers turbocharged engines. Examples are given where the methodology has been successfully applied, covering naturally aspirated engines and both single and dual stage turbocharged engines. Second, the modeling with the emphasis on models for EGR/VGT equipped diesel engine. The aim is to describe models that capture the essential dynamics and nonlinear behaviors and that are relatively small so that they can be utilized in model predictive control algorithms. Special emphasis is on the selection of the states. The third and last topic is related to parameter identification in gray-box models. A common issue is that parameters with physical interpretation often receive values that lie outside their admissible range during the identification. Regularization is discussed as a solution and methods for choosing the regularization parameter are described and highlighted.

The classical trade off between nitrogen oxides (NO _x) and particulate matters (PM) is still one of the key topics for Diesel engine developers. This article gives an overview about models for these emissions usable for online engine and exhaust after treatment control, offline optimization and virtual sensors for monitoring. Two different ways for obtaining such models are presented in detail: first we present a black-box data-based mean value model which estimates engine raw emissions from quantities available in the engine control unit. Second a gray-box model is shown in which also physical equations are used to describe emission formation over crank angle with the measured cylinder pressure as main input.

The Homogeneous Charge Compression Ignition (HCCI) principle holds promise to increase efficiency and to reduce emissions from internal combustion engines. As HCCI combustion lacks direct ignition timing control and auto-ignition depends on the operating condition, control of auto-ignition is necessary. Since auto-ignition of a homogeneous mixture is very sensitive to operating conditions, a fast combustion phasing control is necessary for reliable operation. To this purpose, HCCI modeling and model-based control with experimental validation were studied. A six-cylinder heavy-duty HCCI engine was controlled on a cycle-to-cycle basis in real time using a variety of sensors, actuators and control structures for control of the HCCI combustion in comparison. The controllers were based on linearizations of a previously presented physical, nonlinear, model of HCCI including cylinder wall temperature dynamics. The control signals were the inlet air temperature and the inlet valve closing. A system for fast thermal management was installed and controlled using mid-ranging control. The resulting control performance was experimentally evaluated in terms of response time and steady-state output variance. For a given operating point, a comparable decrease in steady-state output variance was obtained either by introducing a disturbance model or by changing linearization point. Additionally, the robustness towards disturbances was investigated.

This chapter reviews some of the main approaches, results and open problems in Nonlinear Model Predictive Control. The style of the presentation is maintained at a high level, reducing to the minimum the mathematical details.

This chapter describes the application of Pontryagin’s Maximum Principle and Dynamic Programming for vehicle drivingwith minimum fuel consumption. The focus is on minimum-fuel accelerations. For the fuel consumption modeling, a six-parameter polynomial approximation is proposed. With the Maximum Principle, this consumption model yields optimal accelerations with a linearly decreasing acceleration as a function of the velocity. This linear acceleration behavior is also observed in real traffic situations by other researchers. Dynamic Programming is implemented with a backward recursion on a specially chosen distance grid. This grid enables the calculation of realistic gear shifting behaviour during vehicle accelerations. Gear shifting dynamics are taken into account.

Automotive control applications are very challenging due to the presence of constraints, nonlinearities and the restricted amount of computation time and embedded facilities. Nevertheless, the need for optimal trade-off and efficient coupling between the available constrained actuators makes Nonlinear Model Predictive Control (NMPC) conceptually appealing. From a practical point of view however, this control strategy, at least in its basic form, involves heavy computations that are often incompatible with fast and embedded applications. Addressing this issue is becoming an active research topics in the worldwide NMPC community. The recent years witnessed an increasing amount of dedicated theories, implementation hints and software. The Control Parametrization Approach (CPA) is one option to address the problem. The present chapter positions this approach in the layout of existing alternatives, underlines its advantages and weaknesses. Moreover, its efficiency is shown through two real-world examples from the automotive industry, namely:

In the first example, the CPA is applied to the BMW M47TUE Diesel engine available at Johannes Kepler University, Linz while in the second, a real world Smart hybrid demo car available at IFP is used. It is shown that for both examples, a suitably designed CPA can be used to solve the corresponding constrained problem while requiring few milliseconds of computation time per sampling period.

Research Committee on Advanced Powertrain Control Theory in Society of Instrument and Control Engineers (SICE) provided a benchmark control design problem on a V6 automotive spark ignition engine simulation that has strong nonlinearity and discrete event features. Challengers have to start the engine and regulate the engine speed at 650rpm within 1.5s after the ignition by their designed controllers actuating the spark advance and the fuel injector of each cylinder as well as the throttle valve. The background is that systematic control design methodologies are challenged in Model-based Development (MBD) that has been expected to resolve the serious complexity issue of automotive control system developments. MPC is a candidate of the recommended control design methodologies. (Generalized Predictive Control) GPC was studied in this chapter because it can be embeddable on production ECUs with the limited execution speed and the ROM/RAM memory sizes. GPC with the optimized feedforward control succeeded to satisfy the benchmark problem requirements. However, lots of efforts were necessary to design the feedforward and the local linear models, therefore, it is necessary to further continue to investigate NMPC to get the potential benefits of MPC.

Partially premixed combustion is a compression ignited combustion strategy where high exhaust gas recirculation (EGR) levels in combination with early or late injection timing result in a prolonged ignition delay yielding a more premixed charge than with conventional diesel combustion. With this concept it is possible to get low smoke and NO_x emissions simultaneously. Accurate control of injection timing and injection duration is however necessary in order to achieve this favorable mode of combustion. This chapter presents a method for controlled PPC operation. The approach is to control the time between end of injection and start of combustion which if positive yields sufficient premixing. Model Predictive Control was used to control the engine which was modeled using System Identification. The results show that it is possible to assure PPC operation in the presence of both speed/load transients and EGR disturbances.

Model Predictive Control (MPC) can enable powertrain systems to satisfy more stringent vehicle requirements. To illustrate this, we consider an application of MPC to idle speed regulation in spark ignition engines. Improved idle speed regulation can translate into improved fuel economy, while improper control can lead to engine stalls. From a control point of view, idle speed regulation is challenging, since the plant is subject to time delay and constraints. In this chapter, we first obtain a control-oriented model where ancillary states are added to account for delay and performance specifications. Then the MPC optimization problem is defined. The MPC feedback law is synthesized as a piecewise affine function, suitable for implementation in automotive microcontrollers. The obtained design has been extensively tested in a vehicle under different operating conditions. Finally, we show how competing requirements can be met by a switched MPC controller.

In this chapter we present low complexity predictive approaches to the control of autonomous vehicles. A general hierarchical architecture for fully autonomous vehicle guidance systems is presented together with a review of two control design paradigms. Our review emphasizes the trade off between performance and computational complexity at different control levels of the architecture. In particular, experimental results are presented, showing that if the controller at the lower level is properly designed, then it can handle system nonlinearities and model uncertainties even if those are not taken into account at the higher level.

The efficient development of high performance control is becoming more important and more challenging with ever tightening emissions legislation and increasingly complex engines. Many traditional industrial control design techniques have difficulty in addressing multivariable interactions among subsystems and are becoming a bottleneck in terms of development time. In this article we explore the requirements imposed on control design from a variety of sources: the physics of the engine, the embedded software limitations, the existing software hierarchy, and standard industrial control development processes. Decisions regarding the introduction of any new control paradigm must consider balancing this diverse set of requirements. In this context we then provide an overview of our work in developing a systematic approach to the design of optimal multivariable control for air handling in turbocharged engines.

In this work we present the integration of a Linear-time-varying Model-predictive-control (LTV-MPC), designed to stabilize a vehicle during sudden lane change or excessive entry-speed in curve, with a slip controller that converts the desired longitudinal tire force variation to pressure variation in the brake system. The lateral controller is designed using a 3DOF vehicle model taking into account both yaw rate and side slip angle of vehicle while the slip controller is a gain scheduled proportional controller with feedforward action. The performances are validated through simulation: in particular, the authors use a proprietary simulator calibrated on an oversteering sport commercial car and commercial simulator calibrated on a standard light car. Simulation results show the benefits of the control methodology in that very effective steering manoeuvres can be obtained as a result of this feedback policy while satisfying input constraints and show the importance of the introduction of inputs constraints in the control strategy design.

The legal requirement of NO _x emission reduction from legacy gas engines used in compressor stations asks for an improved engine control. A gas engine is a MIMO system with strong coupling, the inputs and outputs being limited by physical constraints and customer requirements. The engines drive compressors that change the load at time instants known in advance and the load change pattern can be modeled. A MIMO online linear model predictive controller (MPC) with the objective of keeping the fuel/air ratio and the engine speed constant was applied and compared to the standard SISO PID controls. The tracking of the fuel/air ratio during the transients was improved up to 80% when using the MPC approach which is sufficient to meet the up-coming emission legislation.

The combination of different desirable characteristics and situation-dependent behavior cause the design of adaptive cruise control (ACC) systems to be time consuming and tedious. This chapter presents a systematic approach for the design and tuning of an ACC, based on model predictive control (MPC). A unique feature of the synthesized ACC is its parameterization in terms of the key characteristics safety, comfort and fuel economy. This makes it easy and intuitive to tune, even for nonexperts in (MPC) control, such as the driver. The effectiveness of the design approach is demonstrated using simulations for some relevant traffic scenarios.