CFD SIMULATION OF PRIMARY ATOMIZATION FOR ENGINE APPLICATIONS.

NIGRO NORBERTO MARCELO; AGUERRE HORACIO; GIMENEZ JUAN; MARQUEZ DAMIAN SANTIAGO; DAMIAN RAMAJO; IDENSHON SERGIO

Buenos Aires

Congreso; Pan-American Congress on Computational Mechanics - PANACM 2015; 2015

CIMNE-ANCA

Liquid atomization is an important process which found interest in several engineeringapplications such as aerospace propulsion systems, automotive engines, food processing,aerosols and ink-jet printing. Its numerical simulation allows to investigate physical processesof the atomization because our understanding on physical mechanisms of suchphenomena is still not sufficient.Our investigation group is doing its first steps in this research area and we report inthis work our early results using two different approaches. On one hand a typical Euleriancollocated finite volume and on the other hand a Lagrangian based Particle Finite ElementMethod (PFEM) using both a volume of fluid (VOF) two phase flow model to capturethe atomization phenomenon.The final target of this research is the improvement of less detailed methodologies tobe applicable in the internal combustion engine simulation using the results obtained herewith an almost direct numerical simulation.In order to validate these codes a very simple test has been chosen. This benchmarkrests on the work of Menard and co-workers [?, ?], which employ the Level Set Method(LSM) to track the interface added to the Ghost Fluid Method (GFM) to describe theinterface discontinuites and manage the pressure, density and viscosity jumps. Also, theLevel Set method is coupled with the Volume of Fluid method (VoF) to ensure massconservation. The mesh used by Menard in [?] is a 2048 × 256 × 256 Cartesian grid withregularly spaced nodes (∆ = 1.17[µm]). The size of the domain is (2.1[mm], 0.3[mm],0.3[mm]), where the first dimension is the streamwise direction and the other two, the1N. Nigro, J. Gimenez, S. Marquez, H. Aguerre and D. Ramajospanwise directions. At the injection level, the jet diameter D is equal to 0.1[mm], whilethe liquid jet Reynolds number is equal to Re = 4659. Liquid surface instabilities close tothe injector are visible. Their deformation leads to the formation of ligaments and dropletsof various sizes. At the end of the domain, the liquid core has almost disappeared anda dense spray of droplets leaves the computational domain. The key of the quickly dropproduction is the use of a space-time correlated turbulent flow at the inlet: Menard usesa syntetized correlated turbulence with a method proposed by Klein [?]. In the workof Desjardins [?], the author uses a forerunner simulation to impose the inlet turbulentboundary condition, obtaining similar results to the above mentioned strategy. Bothworks have a relevant conclusion: by the end of the computational domain, the liquidcore has been fully disintegrated. Another approach in the numerical characterization ofjet atomization is reported by Shinjo in [?, ?]. In his work, the author reports that thegrid resolution used by Menard was coarse for the chosen Reynolds and Weber numbers,so this was not a direct numerical simulation in a true sense: the produced ligaments anddroplets did not exhibit smooth shapes or wave dynamics driven by surface tension, butthe overall liquid jet motion was captured in that simulation. Shinjo et.al had obtainedtheir solution using different uniform meshes, with the finest mesh with around 6000million of cells solving scales up to (∆ = 0.35[µm]). The ligament instabilities are achievedfar from the inlet: the main responsible is the plain velocity front at the inlet imposed byShinjo instead of using a turbulent-induced flow [?].Our initial simulations using the algorithms above mentioned (PFEM and interFOAM)show some similarities with both results, depending on the inlet condition imposed. Itmust be taken into account that in the most refined case simulated with OpenFOAMR(atfull hardware capacity), the geometry was meshed with an uniform cartesian grid of128 × 128 × 1024 (∆x ≈ 2.3[µm]), while the PFEM simulations had a ∆x ≈ 7.5[µm] (7millons of tetrahedral), far from the refinement degree used in the reference works.In order to improve these results an adaptive refinement strategy with OpenFOAMRcalledinterDyMFoam was firstly used with a base mesh of 16×16×128 using 4 refinement levelsreaching around 13 millons cells and a scale resolution of ∆x ≈ 1.17[µm]. These auspiciousresults serve not only for an initial comparison against the references but to understandthe physical phenomena involved and their impact on the engine operation. The role ofthe turbulent velocity profile at the inlet was assessed using the so called vortex methodimplemented in the two codes used in this work. The following step was the addition ofa new refinement level (5 instead of 4) reaching a scale resolution around ∆x ≈ 0.6[µm].The droplet distribution convergence is tested in order to decide when the mesh is fineenough to accomplish this problem. The results show that the droplet formation and thelike-mushroom shape are comparable with Shinjo but the minimum drop size is betterdescribed using a finer mesh reachable with an extra refinement level (6 instead of 5),but currently unaffordable with our available computation resources. Finally it may beconcluded that both methodologies have the potential to solve this type of problems