Guillaume Colin

Neural Control of Fast Nonlinear Systems— Application to a Turbocharged SI Engine With VCT

Today, (engine) downsizing using turbocharging appears as a major way in reducing fuel consumption and pollutant emissions of spark ignition (SI) engines. In this context, an efficient control of the air actuators [throttle, turbo wastegate, and variable camshaft timing (VCT)] is needed for engine torque control. This paper proposes a nonlinear model-based control scheme which combines separate, but coordinated, control modules. Theses modules are based on different control strategies: internal model control (IMC), model predictive control (MPC), and optimal control. It is shown how neural models can be used at different levels and included in the control modules to replace physical models, which are too complex to be online embedded, or to estimate nonmeasured variables. The results obtained from two different test benches show the real-time applicability and good control performance of the proposed methods.

Support vector regression from simulation data and few experimental samples

This paper considers nonlinear modeling based on a limited amount of experimental data and a simulator built from prior knowledge. The problem of how to best incorporate the data provided by the simulator, possibly biased, into the learning of the model is addressed. This problem, although particular, is very representative of numerous situations met in engine control, and more generally in engineering, where complex models, more or less accurate, exist and where the experimental data which can be used for calibration are difficult or expensive to obtain. The first proposed method constrains the function to fit to the values given by the simulator with a certain accuracy, allowing to take the bias of the simulator into account. The second method constrains the derivatives of the model to fit to the derivatives of a prior model previously estimated on the simulation data. The combination of these two forms of prior knowledge is also possible and considered. These approaches are implemented in the linear programming support vector regression (LP-SVR) framework by the addition, to the optimization problem, of constraints, which are linear with respect to the parameters. Tests are then performed on an engine control application, namely, the estimation of the in-cylinder residual gas fraction in Spark Ignition (SI) engine with Variable Camshaft Timing (VCT). Promising results are obtained on this application. The experiments have also shown the importance of adding potential support vectors in the model when using Gaussian RBF kernels with very few training samples.

Physical-Based Algorithms for Interpolation and Extrapolation of Turbocharger Data Maps

Data maps are easy to put in place and require very low calculation time. As a consequence they are often valued over fully physic-based models. This is particularly true when it is question of turbochargers. However, even if these maps are directly provided by the manufacturer, they usually do not cover the entire engine operating range and are poorly discretized. Thats why before implementing them into any model they need to be interpolated and extrapolated.

Optimal Energy Management Strategy including Battery Health through Thermal Management for Hybrid Vehicles

The paper proposes an energy management strategy (EMS) for hybrid electric vehicles (HEV) and plug-in hybrid electric vehicles (PHEV) taking into account battery health through an additional soft constraint on battery internal temperature, considered as one of the prime factors influencing battery aging. Battery cell temperature is modeled and considered as a second state constraint with the state of energy (SOE) in the optimization problem solved on-line using the equivalent consumption minimization strategy (ECMS). Simulation results are presented to highlight the contribution of the strategy including battery thermal management compared to the standard approach.

Residual gas fraction measurement and computation

Abstract To meet future pollutant emissions standards, it is crucial to be able to estimate not only the cycle-by-cycle mass but also the cycle-by-cycle composition of the combustion chamber charge. This charge consists of fresh air, fuel, and residual gas from the previous cycle. Unfortunately, the residual gas fraction cannot be directly measured. Two experimental methods have been designed to determine the residual gas fraction. The reference method is based on an in-cylinder sampling method. The second one is based on a hydrocarbon (HC) analysis of the exhaust gases. Two models have been compared to the experimental results. A one-dimensional computational fluid dynamics (CFD) code (WAVE) and a zero-dimensional model (AMESIM), which takes into account gas compressibility. The aim of the study was to compare the results of CFD codes (one- and zero-dimensional) to experimental results. If the code is validated by the experiments, it should be possible to determine residual gas fractions without needing a large experimental set-up.

Implementation of comfort constraints in dynamic programming for hybrid vehicle energy management

Optimisation algorithms for hybrid vehicles are used to evaluate embedded strategies and component sizing. Using a quasi-static vehicle model, dynamic programming is commonly used to solve the optimal control problem. Using this type of numerical algorithm yields results that are often incompatible with the purpose of smoothness and driver acceptance. Introducing a comfort criterion in the cost function increases the problem complexity and, therefore, the computational burden. In this paper, a novel suboptimal method is proposed which yields valuable results in terms of performance and computational time. This approach takes occupant comfort into account and thus comes close to a realistic solution.

Exact and Linearized Neural Predictive Control: A Turbocharged SI Engine Example

This paper describes a real-time control method for non-linear systems based on model predictive control. The model used for the prediction is a neural network because of its ability to represent non-linear systems, its ability to be differentiated, and its simplicity of use. The feasibility and the performance of the method, based on on-line linearization, are demonstrated on a turbocharged spark-ignited engine application, where the simulation models used are very accurate and complex. The results, first in simulation and then on a test bench, show the implementation of the proposed control scheme in real time.

Explicit nonlinear model predictive control of the air path of a turbocharged spark-ignited engine

Pollutant emissions and fuel economy objectives have led car manufacturers to develop innovative and more sophisticated engine layouts. In order to reduce time-to-market and development costs, recent research has investigated the idea of a quasi-systematic engine control development approach. Model based approaches might not be the only possibility but they are clearly predetermined to considerably reduce test bench tuning work requirements. In this paper, we present the synthesis of a physics-based nonlinear model predictive control law especially designed for powertrain control. A binary search tree is used to ensure real-time implementation of the explicit form of the control law, computed by solving the associated multi-parametric nonlinear problem.

On Learning Machines for Engine Control

The chapter deals with neural networks and learning machines for engine control applications, particularly in modeling for control. In the first section, basic features of engine control in a layered engine management architecture are reviewed. The use of neural networks for engine modeling, control and diagnosis is then briefly described. The need for descriptive models for model-based control and the link between physical models and black box models are emphasized by the grey box approach discussed in this chapter. The second section introduces the neural models frequently used in engine control, namely, MultiLayer Perceptrons (MLP) and Radial Basis Function (RBF) networks. A more recent approach, known as Support Vector Regression (SVR), to build models in kernel expansion form is also presented. The third section is devoted to examples of application of these models in the context of turbocharged Spark Ignition (SI) engines with Variable Camshaft Timing (VCT). This specific context is representative of modern engine control problems. In the first example, the airpath control is studied, where open loop neural estimators are combined with a dynamical polytopic observer. The second example considers modeling the in-cylinder residual gas fraction by Linear Programming SVR (LP-SVR) based on a limited amount of experimental data and a simulator built from prior knowledge. Each example demonstrates that models based on first principles and neural models must be joined together in a grey box approach to obtain effective and acceptable results.

Linearized Neural Predictive Control A Turbocharged SI Engine Application

Nowadays, (engine) downsizing using turbocharging appears as a major way for reducing fuel consumption. With this aim in view, the air actuators (throttle, Turbo WasteGate) control is needed for an efficient engine torque control particularly to reduce pumping losses and to increase efficiency. This work proposes Nonlinear Model Predictive Control (NMPC) of the air actuators for turbocharged SI engines where the predictions are achieved by a neural model. The results obtained from a test bench of a Smart MCC engine show the real time applicability of the proposed method based on on-line linearization and the good control performances (good tracking, no overshoot) for various engine speeds.