rail system is not optimized to reduce engine emissions. Herzog [1] defined "ideal" shapes of the rate of injection versus crank angle for different values of engine ...
AN APPLICATION OF MULTI-CRITERIA GENETIC ALGORITHMS TO THE OPTIMIZATION OF A COMMON-RAIL INJECTOR Arturo de Risi
Teresa Donateo
Domenico Laforgia
Università di Lecce Dipartimento di Ingegneria dell’Innovazione, 73100 via Arnesano, Lecce - Italy
ISUFI - Innovative Materials and Technologies - Università di Lecce, 73100 via Monteroni, Lecce – Italy
Università di Lecce -Dipartimento di Ingegneria dell’Innovazione, 73100 via Arnesano, Lecce - Italy
ABSTRACT
INTRODUCTION
The aim of the present investigation is to optimize the injection profile and the time response of a high pressure common rail injection system by adjusting the geometric and dynamic characteristics of an electronically controlled injector.
The influence of injection parameters, like pressure, timing, duration and injection on the performance and emission levels of DI diesel engine has been widely assessed. The introduction of common rail fuel injection system allows a great flexibility in the control of the injection pressure. Moreover, because of the separation between pressure generation and fuel delivery, complex injection strategies including pilot and post injections can be exploited. On the other hand, the rate of injection versus time obtained with the injector used for common rail system is not optimized to reduce engine emissions. Herzog [1] defined "ideal" shapes of the rate of injection versus crank angle for different values of engine load and speed. These injection profiles are characterized by two stages of injection. In the first one, the injection velocity is sharply risen and kept at an intermediate value. Then, the velocity is increased up to the peak value and kept until the end of injection. The duration of the first stage and the rate of transition from the intermediate value to the peak value increases as engine speed is increased. Erlach et al. [2] reproduced this ideal shape of the rate of injection by using pressure modulated injectors.
The optimization method is based on the use of genetic algorithms which include the operators of crossover, mutation and elitist reproduction. As evaluation function for the GA a 1D simulation code of the injection systems, developed and extensively tested by the authors, has been used. The 1D model is based on the concentrated volume method and includes the effect of friction on the dynamics of the movable parts. Cavitation is also taken into account by the code as well as the effect on pressure wave propagation of the air and vapor mass fraction in the fuel. Conservation equations are integrated by using the characteristic method. The electromagnetic force generated by the solenoid on the head of the injector is simulated with an empirical function obtained by fitting experimental data. The optimized injection profile was defined by evaluating the predicted performance of a small bore DI diesel engine in terms of Brake Mean Effective Pressure (BMEP), soot and NOx emissions using a modified version of the KIVA-3V code. For the definition of the best injection profile injected quantity and start of injection were kept constants and only single-injection strategies were considered. The geometric and dynamic parameters (i.e. spring stiffness) of a commercial five holes VCO injector were used as baseline case. An optimized configuration of the selected parameters has been found. The optimized combinations of the investigated parameters are compared with the original values of the commercial injector as well as their predicted performance on the engine application.
The injection profile obtained with common-rail electronically controlled injectors is governed by both the fluid dynamics of the flow in the injector and the dynamic response of the system. Thus, due to the constant feeding pressure it is very difficult to modulate fuel injection according to the engine operating conditions. In a previous study the authors [3] proved that different injection profiles can be obtained by changing the geometric data of a high pressure common rail electroinjector. In the referred study, a square injection profile was set as goal for the optimization process to test the possibility to modify the injection velocity rate of change. The objective of the present investigation is twofold: the determination of optimized injection profiles by means of
251
Copyright © 2002 by ASME
multidimensional engine modeling, and the optimization of the dynamic response of the injector, simulated with a 1-D code, to reproduce the selected injection profile. Both optimization process has been performed with a multiobjective genetic algorithm presented in a previous investigation [3]. Multidimensional modeling as evaluation method for genetic algorithms optimization has been widely applied by Senecal and his coworkers[4-6]. Senecal developed a KIVA-GA computer code which allows the contemporary minimization of NOx, unburned HC, soot emissions as well as fuel consumption to be performed. The code was applied to heavy-duty and light duty diesel engines and several parameters affecting engine performances were taken into account in the optimization process: the combustion chamber shape, the injection timing, the percent of EGR, the swirl ratio at intake valve closing, the duration of injection, etc. The multi-objective optimization was performed by Senecal by combining three merit functions in a single overall fitness function.
THE MULTI-OBJECTIVE GENETIC ALGORITHM The multi-objective genetic algorithm has been presented and tested in a previous study [3]. For the reader's convenience, the most important phases of the optimization process are recalled here. The algorithm starts from a random initial population of binary strings. Each string, named individual, codifies a combination of parameters to be evaluated. Once a first random generation is selected, the algorithm calculates the fitness functions of each individual according to each single objective to optimize. Then, the individuals are ranked according to Pareto optimality criterion. According to Pareto optimality, a vector x is partially less (
