Development of a Novel Machine Learning Methodology for the Generation of a Gasoline Surrogate Laminar Flame Speed Database under Water Injection Engine Conditions

Features
Authors Abstract
Content
The water injection is one of the technologies assessed in the development of new internal combustion engines fulfilling new emission regulation and policy on Auxiliary Emission Strategy assessment. Besides all the positive aspects about the reduction of mixture temperature at top dead center and exhaust gases temperature at turbine inlet, it is well known that the water vapor acts as a mixture diluter, thus diminishing the reactants burning rate. A common methodology employed for the Reynolds-Averaged Navier-Stokes Computational Fluid Dynamics (RANS CFD) simulation of the reciprocating internal combustion engines’ turbulent combustion relies on the flamelet approach, which requires knowledge of the Laminar Flame Speed (LFS) and thickness. Typically, these properties are calculated by means of correlation laws, but they do not keep into account the presence of water mass fraction. A more precise methodology for the definition of both the LFS and thickness is thus required. The interrogation of a previously computed look-up table of such properties during run time seems to be a suitable and more accurate method than using correlations. In order to generate a database with all the possible combinations of chemical and physical properties that can be reached during the simulation of internal combustion engines, including the presence of a given mass fraction of water vapor and exhaust gases, a very high number of detailed chemical kinetics simulations need to be performed. The present work aims to introduce a new methodology for the fast generation of laminar flame characteristics look-up tables that account also for the presence of water vapor in the reacting mixture. By using this new approach, engine designers will have the possibility to generate look-up tables of laminar flame characteristics for different fuels with the same computational cost that is currently required to generate a single table.
Meta TagsDetails
DOI
https://doi.org/10.4271/04-13-01-0001
Pages
13
Citation
Pulga, L., Bianchi, G., Ricci, M., Cazzoli, G. et al., "Development of a Novel Machine Learning Methodology for the Generation of a Gasoline Surrogate Laminar Flame Speed Database under Water Injection Engine Conditions," SAE Int. J. Fuels Lubr. 13(1):5-17, 2020, https://doi.org/10.4271/04-13-01-0001.
Additional Details
Publisher
Published
Nov 19, 2019
Product Code
04-13-01-0001
Content Type
Journal Article
Language
English