Introduction to the CFD Process

Computational Fluid Dynamics or CFD is the use of mathematics,

physics and computational software to solve problems of fluid flow
over body surfaces and through complex geometries. The process
involves five steps – Geometry creation, mesh generation, pre-
processing, solving, post-processing.
Geometry Creation: This is first step to solve any problem in fluid
dynamics. The domain and geometry of the body are created in this
step. Software like CATIA, SolidWorks, etc. are used to create the
geometry which are then imported into other software which generate
the required mesh. In some cases, the geometry can be created within
an all-in-one package like ANSYS, Su2, etc.
Mesh Generation: This is an important part of the CFD analysis, which
involves discretizing the domain into small parts for the solver to easily
analyze the fluid flow. The grids generated can be structured or
unstructured, each with its own advantages and disadvantages. The
structured mesh is rigid and cannot be sometimes used to properly
divide the domain, while the unstructured meshes are flexible and are
most favored. Software like ICEM, ANSYS, etc. are the most commonly
used software. The mesh can also be fine or coarse depending on the
application, and user. The number of grids is a key consideration
because the higher their number, longer the computational time. The
mesh flow should ideally should be made in such a way that it
intuitively follows the flow pattern and the areas with large gradients
are more refined. The mesh generation generally ends with the
specification of boundaries for the domain and body under study.
Pre-processing: This is the most crucial step in the entire CFD process,
as a mistake here can lead to erroneous calculations in the solver,
which gives a completely different result than expected. It involves
setting the boundary conditions for the body, free-stream conditions
for the domain, and is used to specify the number iterations, the
fineness of the calculations (given by the Courant-Fredrichs-Lewy or CFL
number), wall data, etc.
Solving: This step is at the heart of the entire CFD process. The solver is
a complex set of rules that solve Navier-Stokes’ equations or a different
of the same, such as RANS or Reynolds’ Averaged Navier-Stokes’. Other
approaches are Large Eddy Simulation (LES), or Direct Numerical
Simulation (DNS). Methods or models used to solve these equations are
the Spalart-Allmaras, k-omega, k-epsilon, Transition SST, etc. For
simplicity one or two equations models such as the Spalart-Allmaras or
k-omega models are commonly used. Since the solver solves these
differential equations using numerical methods, the main challenge is
to get a faster result with minimum error (as numerical methods are
just an approximation). The numerical methods used are Newton-
Rapson method, Runge-Kutta method, generalized minimal residual
method (GMRES), Backward Differential Formulae 1, 2, 3. (BDF1, BDF2,
or BDF3). The solver runs for the number or iterations specified or until
the solution converges. The solver generally has a provision to view
convergence or residue plots, integrated data plots, etc. During the
course of this internship, we were trained to use the home-grown High-
resolution Flow solver on UNstructured meshes or HiFUN.
Post-processing: This is the final step in the CFD process. Here, we can
visualize the flow pattern, physical properties at different places,
variation in the them as time progresses, and plot the different graphs
required the problem under analysis. In a nutshell, it provides the user
with a tool with which he/she can relate the solved problem with real
scenarios.
 Geometry Creation

Grid Generation

Pre-processing

Solving

Post Processing

Features of the HiFUN solver:

HiFUN or High resolution Flow solver on UNstructured meshes or
HiFUN, is a general purpose flow solver that uses unstructured data
based algorithms. The code is fine tuned to solve problems pertaining
to the aerospace industry and can be extended to the automobile
industry. The solver can handle calculations for flows ranging from low
to high supersonic Mach numbers.
 Hexahedron, tetrahedron, prism and
Mesh elements
pyramid
Achieved through Green-Gauss procedure
Spatial Accuracy
or method of least squares
Supports a number of numerical flux
Numerical flux formula formulae but allows the user to determine,
based on his or her choice
Allows for robustness of the solver even in
Viscous discretization
highly skewed cells
Handles non conformal grids at the
Non-conformal Block Interface algorithm
interface of two blocks
Possible to simulate both translating and
Moving wall boundary condition
rotating walls
Multiple rotating frames of reference Handles multiple rotating frames of
algorithm reference for any problem
The flow solver is capable of solving
Algorithm to simulate porous media
problems involving porous media
Supports Spalart-Allmaras and k-omega
Turbulence Model
turbulence models
The standard equilibrium wall function
gets automatically activated for turbulent
Wall function
calculations if grid density is not adequate
near the wall
Rapid convergence through a matrix-free
Convergence acceleration
implicit procedure
A four layered approach to data handling
Parallel Processing ensures there is no degeneration of data
handling
A dual time stepping procedure is used to
Higher order time accuracy achieve second order time accuracy both
on stationary and moving grids
Sensors based on error estimator and
Grid Adaptation
error indicator is used as a hybrid strategy.
HiFUN is suited for serial applications,
Migratory Memory Algorithm which implies a machine’s solving
capability is substantially enhanced