In the present paper the combustion process in a modern second- generation, common-rail diesel engine for light-duty application is experimentally and numerically investigated. An improved version of the KIVA3V-Release 2 code was used for the simulations. To model the combustion process, a detailed kinetic scheme involving 57 species and 290 equations, based on the n-heptane combustion, was used, interfacing the KIVA3V code with the CHEMKIN-II chemistry package. The full set of equations is concurrently solved in each computational cell by different solvers with the final aim of obtaining a locally adaptative code: local choices are undertaken in terms of time steps as well as in terms of the employed solvers. To reduce computational time, the code was parallelized: this parallelization is mainly focused on the chemical subroutines, considering that they are responsible for more than the 95% of the computing. Due to the spatial in-homogeneous characteristics of diesel combustion, the grid partitioning is a key point for efficient computation. Therefore, different grid partitioning criteria were used and analyzed in terms of "divide and conquer" advantages and load balancing issues. The performance analysis suggests that a random partitioning criterion is useful to smooth the grid inhomogeneities over the processes.
Multidimensional modeling of advanced Diesel combustion system by parallel chemistry
Belardini P;Bertoli C;Corsaro S;D'Ambra P
2005
Abstract
In the present paper the combustion process in a modern second- generation, common-rail diesel engine for light-duty application is experimentally and numerically investigated. An improved version of the KIVA3V-Release 2 code was used for the simulations. To model the combustion process, a detailed kinetic scheme involving 57 species and 290 equations, based on the n-heptane combustion, was used, interfacing the KIVA3V code with the CHEMKIN-II chemistry package. The full set of equations is concurrently solved in each computational cell by different solvers with the final aim of obtaining a locally adaptative code: local choices are undertaken in terms of time steps as well as in terms of the employed solvers. To reduce computational time, the code was parallelized: this parallelization is mainly focused on the chemical subroutines, considering that they are responsible for more than the 95% of the computing. Due to the spatial in-homogeneous characteristics of diesel combustion, the grid partitioning is a key point for efficient computation. Therefore, different grid partitioning criteria were used and analyzed in terms of "divide and conquer" advantages and load balancing issues. The performance analysis suggests that a random partitioning criterion is useful to smooth the grid inhomogeneities over the processes.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
