Numerical Study of Liquid Sheet Atomization Using Two-Phase Flow Solver on GPU Architecture

Reddy, S Rajesh and Banerjee, Raja (2016) Numerical Study of Liquid Sheet Atomization Using Two-Phase Flow Solver on GPU Architecture. PhD thesis, Indian Institute of Technology Hyderabad.

Text
ME10P006.pdf - Submitted Version
Restricted to Repository staff only until August 2024.
Download (12MB) | Request a copy

Abstract

Sprays are encountered in a wide variety of engineering applications such as diesel engines, gas turbine engines, coating and painting, inkjet printing etc. To generate sprays in these applications, a liquid must be atomized. Atomization usually refers to the disintegration of a bulk liquid material into small droplets in the ambient gas. The main objective of atomization and spray systems is to generate a spray with desired droplet size and velocity distribution. In classical atomization process, the initial liquid jet/sheet emanating from the injector starts disintegrating into ligaments or droplets when the aerodynamic forces due to shearing interaction at the liquid-gas interface is greater than the surface tension of the liquid. This is called primary breakup or primary atomization. Majority of the droplets generated during primary atomization are unstable and may further disintegrate in to smaller droplets if the parent droplet is larger than a critical size. This is called secondary breakup or secondary atomization. Mechanism of primary atomization is still not well understood and is an active subject of scienti�c research. The ideal approach to resolve the primary atomization process is by direct numerical simulations (DNS). The objective of the current study is to predict the primary atomization process by means of high �delity numerical simulations (close to DNS). Two basic requirements of DNS of multiphase ows are � an Eulerian based method capable of e�cient representation of evolving ow features of widely di�erent characteristic spatial scales. � e�cient parallelization of the solver to handle computational requirement. Volume of uid (VOF) method has been implemented to capture the gas liquid interface. One uid formulation of Navier-Stokes equations was used to describe the motion of the two uids present in the domain. Numerical solution of the 2-D Navier-Stokes equations in non-conservative form is obtained using simpli�ed marker and cell (SMAC) algorithm. Surface tension was included as a source term and evaluated using continuum surface force (CSF) method. Pressure Poisson equation was solved using two di�erent approaches namely multigrid method and symmetric Gauss Siedel preconditioned conjugate gradient (SGSPCG) method. Solving pressure Poisson equation is the most time consuming part of the entire solver. To e�ciently handle the computational requirement pressure Poisson solvers were parallelized on graphics processing unit (GPU) architecture. GPU is a many-core multithreaded multiprocessor that can perform both graphics and computing and can be used in conjunction with a computer. GPU programming was done using compute uni�ed device architecture (CUDA) platform by NVIDIA. The GPU based two phase ow solver was validated against standard benchmark test cases and approximate DNS studies were performed to study primary atomization process under Diesel jet, gas blast and pre�lming gas blast atomization like conditions. vi Dependence of pre�lming gasblast atomization spray characteristics on di�erent parameters like inner core gas velocity, outer gas velocity and liquid sheet thickness was investigated. Pre�lming gas blast atomization simulations are done on a grid size of 2048 � 1024 and a speedup of approximately 12 � is obtained with GPU based solver using Tesla K20 GPU accelerator. In actual pre�lming gas blast atomization (occurring typically in gas turbines), the gases exiting the injector ori�ce are subjected to velocity modulations. Therefore liquid sheet atomization with the e�ect of forcing conditions imposed on discharging gas streams was also studied.

[error in script]

IITH Creators: