curriculum vitae
english version
Tesi di Dottorato


High Performance Genetic Algorithm for Engine Optimization

CREA  Research Center for Energy and Environment

Center for Advanced Computational Technologies/ISUFI ,

An innovative methodology for the design of the combustion chamber based on a genetic algorithm and numerical simulations has been developed and implemented on a web portal called DESGrid to allow a trusted user to apply the optimization code called HiPeGEO.

The DESGrid architecture


The method consists of nine steps and was applied to optimize the design of a single-cylinder research engine by changing the shape of the combustion chamber. Emissions reduction was considered as the main goal of the design process at low speed and load while the increasing of IMEP was set as target at full load. Four separated fitness components were defined according to NOx, soot, HC and IMEP.

The optimization was performed by means of a micro genetic algorithm  (HiPeGEO) combined with a modified version of the KIVA3V code.

After running the genetic algorithm several configurations capable of improving engine behavior with respect to all the optimization goals were identified. One of these chambers was chosen as the best configuration thanks to its ability to reduce soot, NOx and HC emissions and increase Indicated Mean Pressure for all modes with respect to the reference chamber. These results were explained by analyzing the flow field and the oxygen distribution in the chamber.

The method was experimentally validated by building and testing the optimized chamber and a good correspondence between measured and calculated behavior was found. Thanks to the experimental results the capability of the design methodology in developing combustion chambers respectful of the more and more stringent legislation about pollutant emissions was proved.

 The design method consists of nine optimization steps:

 A.                  Identification of the geometrical parameters to be optimized, e.g. combustion chamber shape, compression ratio, bore to stroke ratio, injector position, inlet valve closing, etc.

B.                   Choice of a model to predict engine behavior when geometrical parameters are changed; if multi-dimensional simulation codes are used, this step requires the definition of a parametric computational mesh to divide the geometrical domain in a sufficiently high number of cells.

C.                   Choice of the strategic goals to reach, e.g. improvement of engine torque, reduction of pollutant emissions, control of combustion noise, etc. Note that only the most important output values have to be included in the fitness array, since the other ones can be used as penalty functions (see step E).

D.                  Selection of the operating conditions (modes) for the optimization and choice of their weight in the definition of the fitness array; since engine behavior is strongly influenced by load and speed, as many modes as possible should be used; the weight should be assigned according to the importance of each mode for each objective, e.g. if the optimization process aims at reducing emissions high-speed high-load modes are not important while low and medium values of load and speed could be weighted according to their occurrence in the driving tests;

E.                   Definition of penalty functions to penalize geometrical solutions that do not respect user-defined requirements, e.g. noise emissions, structural constrains on maximum pressure, maximum discharge temperature, etc.

F.                   Setup of an interface between the genetic algorithm and the engine model;

G.                   Identification of the reference configuration for the optimization; e.g. a commercial chamber that represent the baseline configuration to improve or a target solution that indicates the best expected results obtainable by changing engine geometry. This step requires the definition of the range of variation for all parameters chosen at step 1

H.                  Run of the genetic algorithm with a selected number of possible solutions to be contemporary analyzed in each GA step (number of individuals in both replaceable and non-replaceable portions) and a pre-determined number of external iterations. The highest the number of individuals to be tested, the highest the required computational time and the highest the confidence in the optimization results.

I.                     Representation of the results in the hyperspace defined by the fitness array components and choice of the optimal configurations to be built and tested.

Results of the optimization



To reduce soot and

NOx emissions




              To improve torque


Home | Hi.Pe.GEO | Tesi di Dottorato | H2-VOLKS | ITAN500 | pubblicazioni

Ultimo aggiornamento: 01-02-13