INAUGURAL–DISSERTATION zur Erlangung der Doktorwürde der ...
INAUGURAL–DISSERTATION zur Erlangung der Doktorwürde der ...
INAUGURAL–DISSERTATION zur Erlangung der Doktorwürde der ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
3.2. Spray Modeling 51<br />
application and numerical accuracy, a second or<strong>der</strong> explicit finite difference scheme is<br />
applied to solve steady state form of Eqs. (2.26) – (2.28) [192], which are represented<br />
by Eq. (3.21). Thus the solution formula may be written as [193]<br />
U j+1<br />
n,i<br />
= U j n,i − ∆x [<br />
1.5E<br />
j<br />
i<br />
∆z<br />
− 2Ej i−1 + ] 0.5Ej i−2 + ∆xS<br />
j<br />
i , (3.23)<br />
where i and j are grid nodes in radial and axial directions, respectively.<br />
The above formulation is applied to an equidistant rectangular grid, where the size<br />
of each grid cell is 1.5 × 10 −3 m in radial direction and 1.0 × 10 −4 m in axial<br />
direction, resulting in a maximum of 80 × 1000 grid nodes. The initial data to start<br />
simulations in both the configurations, i.e, one and two-dimensional cases is generated<br />
from the experimental data provided by Dr. R. Wengeler, BASF Ludwigshafen (onedimensional<br />
water spray in nitrogen) and Prof. G. Brenn, TU Graz (two-dimensional<br />
water and PVP/water spray in air) using Wheeler algorithm (see Subsection 3.2.3). The<br />
experimental data closest to the nozzle exit is taken for generating the initial data and<br />
the procedure for calculating this initial data from experiment is explained in Chapter<br />
4 along with brief description about the experimental setup. The boundary conditions<br />
in solving DQMOM include (1) if droplets hit the axis of symmetry, they are reflected,<br />
and (2) Neumann boundary is applied for the lateral sides of the computational domain<br />
and exit plane. The experimental data available at other cross sections away from the<br />
nozzle exit is used to validate the simulation results. The flowchart of the computational<br />
code is illustrated in Fig. 3.1.<br />
3.2.3 Wheeler Algorithm<br />
The Wheeler algorithm developed by Sack and Donovan [136], requires 2N +1 moments<br />
to compute N weights (number density) and N abscissas (droplet radii or velocities).<br />
The moment set is represented as M = [M(0), M(1), ...M(2N + 1)] T . This algorithm<br />
is used to generate the initial data in DQMOM whereas in QMOM it is used to compute<br />
the unknown moments. The first step in Wheeler algorithm is to compute the<br />
coefficients π α based on these 2N + 1 moments of the distribution function n(ξ), given<br />
as<br />
π α+1 (ξ) = ξπ α (ξ). (3.24)<br />
The above recursive relation has the properties of π −1 (ξ) = 0 and π 0 (ξ) = 1. Here, α<br />
is a subset of number of moments 2N + 1, i.e., α ∈ 0, 1, 2..N − 1. From these coefficients<br />
π α (ξ), a symmetric tridiagonal matrix is computed through some intermediate<br />
quantities:<br />
∫<br />
σ α,β = n(ξ)π α (ξ)π β (ξ)dξ, (3.25)