Ensemble average method for runtime saving in Large Eddy Simulation of free and Ducted Fuel Injection (DFI) sprays

Computational Fluid Dynamics (CFD) with Large Eddy Simulation (LES) turbulence model is a valuable tool to investigate complex problems. However, for high Reynolds number problems, the associated huge computational cost often leads researchers to the use of more simplified and less accurate approaches, especially if statistics is needed for the generalization of the results and comparison against experimental data. Therefore, the intro-duction of innovative methodologies to reduce the computational cost maintaining results reliability would be of paramount importance for LES-based investigation. In this context, the aim of this work is to assess a runtime saving methodology to ensemble average several axial symmetric spray simulations obtained with LES. In particular, the number of independent samples for the average procedure has been increased by exploiting the axial symmetry characteristics of a diesel spray case study, extracting more realizations from a single simulation. This ensemble average approach was compared with the standard one, based on one realization per simulation, at equal statistical sample size. Main spray physical quantities and turbulence characteristics were examined, both globally and locally. The same procedure was also applied to a different diesel-relevant spray configuration, known as ducted fuel injection. The reliability of this ensemble average methodology has been herein proven for both spray configurations, highlighting a dramatic runtime saving without any worsening of the accuracy level. In particular, this approach, as applied in this work, guaranteed a computational cost reduction of 50-75%. Thereby, the present method-ological assessment could motivate researchers involved in the investigation of spray processes to undertake the path of statistically significant LES analysis.

Automated summary

Computational Fluid Dynamics (CFD) with Large Eddy Simulation (LES) turbulence model is a valuable tool for investigating complex problems. However, high Reynolds number problems often lead to the use of simplified and less accurate approaches due to the huge computational cost. In this study, a runtime saving methodology was developed to increase the number of independent samples for ensemble average, resulting in a significant reduction in computational cost without sacrificing accuracy.