Introduction to Computational Fluid Dynamics

Instructor: Dmitri Kuzmin

Institute of Applied Mathematics
University of Dortmund
kuzmin@math.uni-dortmund.de
http://www.featflow.de

Fluid (gas and liquid) flows are governed by partial differential equations which
represent conservation laws for the mass, momentum, and energy.
Computational Fluid Dynamics (CFD) is the art of replacing such PDE systems
by a set of algebraic equations which can be solved using digital computers.
http://www.mathematik.uni-dortmund.de/kuzmin/cfdintro/cfd.html

What is fluid flow?

Fluid flows encountered in everyday life include
meteorological phenomena (rain, wind, hurricanes, floods, fires)
environmental hazards (air pollution, transport of contaminants)
heating, ventilation and air conditioning of buildings, cars etc.
combustion in automobile engines and other propulsion systems
interaction of various objects with the surrounding air/water
complex flows in furnaces, heat exchangers, chemical reactors etc.
processes in human body (blood flow, breathing, drinking . . . )
and so on and so forth

What is CFD?
Computational Fluid Dynamics (CFD) provides a qualitative (and
sometimes even quantitative) prediction of fluid flows by means of
mathematical modeling (partial differential equations)
numerical methods (discretization and solution techniques)
software tools (solvers, pre- and postprocessing utilities)
CFD enables scientists and engineers to perform numerical experiments
(i.e. computer simulations) in a virtual flow laboratory

real experiment

CFD simulation

Why use CFD?

Numerical simulations of fluid flow (will) enable
architects to design comfortable and safe living environments
designers of vehicles to improve the aerodynamic characteristics
chemical engineers to maximize the yield from their equipment
petroleum engineers to devise optimal oil recovery strategies
surgeons to cure arterial diseases (computational hemodynamics)
meteorologists to forecast the weather and warn of natural disasters
safety experts to reduce health risks from radiation and other hazards
military organizations to develop weapons and estimate the damage
CFD practitioners to make big bucks by selling colorful pictures :-)

Examples of CFD applications

Aerodynamic shape design

Examples of CFD applications

CFD simulations by L
ohner et al.

Examples of CFD applications

Smoke plume from an oil fire in Baghdad

CFD simulation by Patnaik et al.

Experiments vs. Simulations

CFD gives an insight into flow patterns that are difficult, expensive or impossible
to study using traditional (experimental) techniques
Experiments

Simulations

Quantitative description of flow

phenomena using measurements

Quantitative prediction of flow

phenomena using CFD software

for one quantity at a time

for all desired quantities

at a limited number of points

and time instants

with high resolution in

space and time

for a laboratory-scale model

for the actual flow domain

for a limited range of problems

and operating conditions

for virtually any problem and

realistic operating conditions

Error sources: measurement errors,

flow disturbances by the probes

Error sources: modeling, discretization, iteration, implementation

Experiments vs. Simulations

As a rule, CFD does not replace the measurements completely but the amount
of experimentation and the overall cost can be significantly reduced.
Experiments

expensive
slow
sequential
single-purpose

Simulations

cheap(er)
fast(er)
parallel
multiple-purpose

Equipment and personnel

are difficult to transport
CFD software is portable,
easy to use and modify

The results of a CFD simulation are never 100% reliable because

the input data may involve too much guessing or imprecision
the mathematical model of the problem at hand may be inadequate
the accuracy of the results is limited by the available computing power

Fluid characteristics
Macroscopic properties

Classification of fluid flows

density

viscous

inviscid

viscosity

compressible

incompressible

pressure

steady

unsteady

temperature

laminar

turbulent

velocity

single-phase

multiphase

The reliability of CFD simulations is greater

for laminar/slow flows than for turbulent/fast ones
for single-phase flows than for multi-phase flows
for chemically inert systems than for reactive flows

How does CFD make predictions?

CFD uses a computer to solve the mathematical equations for the problem
at hand. The main components of a CFD design cycle are as follows:
the human being (analyst) who states the problem to be solved
scientific knowledge (models, methods) expressed mathematically
the computer code (software) which embodies this knowledge and
provides detailed instructions (algorithms) for
the computer hardware which performs the actual calculations
the human being who inspects and interprets the simulation results
CFD is a highly interdisciplinary research area which lies at the interface of
physics, applied mathematics, and computer science

CFD analysis process

1.

Problem statement

information about the flow

2.

Mathematical model

IBVP = PDE + IC + BC

3.

Mesh generation

nodes/cells, time instants

4.

Space discretization

coupled ODE/DAE systems

5.

Time discretization

algebraic system Ax = b

6.

Iterative solver

discrete function values

7.

CFD software

implementation, debugging

8.

Simulation run

parameters, stopping criteria

9.

Postprocessing

visualization, analysis of data

Verification

model validation / adjustment

10.

Problem statement
What is known about the flow problem to be dealt with?
What physical phenomena need to be taken into account?
What is the geometry of the domain and operating conditions?
Are there any internal obstacles or free surfaces/interfaces?
What is the type of flow (laminar/turbulent, steady/unsteady)?
What is the objective of the CFD analysis to be performed?
computation of integral quantities (lift, drag, yield)
snapshots of field data for velocities, concentrations etc.
shape optimization aimed at an improved performance
What is the easiest/cheapest/fastest way to achieve the goal?

Mathematical model
1. Choose a suitable flow model (viewpoint) and reference frame.
2. Identify the forces which cause and influence the fluid motion.
3. Define the computational domain in which to solve the problem.
4. Formulate conservation laws for the mass, momentum, and energy.
5. Simplify the governing equations to reduce the computational effort:
use available information about the prevailing flow regime
check for symmetries and predominant flow directions (1D/2D)
neglect the terms which have little or no influence on the results
model the effect of small-scale fluctuations that cannot be captured
incorporate a priori knowledge (measurement data, CFD results)
6. Add constituitive relations and specify initial/boundary conditions.

Discretization process
The PDE system is transformed into a set of algebraic equations
1. Mesh generation (decomposition into cells/elements)
structured or unstructured, triangular or quadrilateral?
CAD tools + grid generators (Delaunay, advancing front)
mesh size, adaptive refinement in interesting flow regions
2. Space discretization (approximation of spatial derivatives)
finite differences/volumes/elements
high- vs. low-order approximations
3. Time discretization (approximation of temporal derivatives)
explicit vs. implicit schemes, stability constraints
local time-stepping, adaptive time step control

Iterative solution strategy

The coupled nonlinear algebraic equations must be solved iteratively
Outer iterations: the coefficients of the discrete problem are updated using
the solution values from the previous iteration so as to
get rid of the nonlinearities by a Newton-like method
solve the governing equations in a segregated fashion
Inner iterations: the resulting sequence of linear subproblems is typically
solved by an iterative method (conjugate gradients, multigrid) because
direct solvers (Gaussian elimination) are prohibitively expensive
Convergence criteria: it is necessary to check the residuals, relative solution
changes and other indicators to make sure that the iterations converge.
As a rule, the algebraic systems to be solved are very large (millions of unknowns)
but sparse, i.e., most of the matrix coefficients are equal to zero.

CFD simulations
The computing times for a flow simulation depend on
the choice of numerical algorithms and data structures
linear algebra tools, stopping criteria for iterative solvers
discretization parameters (mesh quality, mesh size, time step)
cost per time step and convergence rates for outer iterations
programming language (most CFD codes are written in Fortran)
many other things (hardware, vectorization, parallelization etc.)
The quality of simulation results depends on
the mathematical model and underlying assumptions
approximation type, stability of the numerical scheme
mesh, time step, error indicators, stopping criteria . . .

Postprocessing and analysis

Postprocessing of the simulation results is performed in order to
extract the desired information from the computed flow field
calculation of derived quantities (streamfunction, vorticity)
calculation of integral parameters (lift, drag, total mass)
visualization (representation of numbers as images)
1D data: function values connected by straight lines
2D data: streamlines, contour levels, color diagrams
3D data: cutlines, cutplanes, isosurfaces, isovolumes
arrow plots, particle tracing, animations . . .
Systematic data analysis by means of statistical tools
Debugging, verification, and validation of the CFD model

Uncertainty and error

Whether or not the results of a CFD simulation can be trusted depends on the
degree of uncertainty and on the cumulative effect of various errors
Uncertainty is defined as a potential deficiency due to the lack of knowledge
(turbulence modeling is a classical example)
Error is defined as a recognizable deficiency due to other reasons
Acknowledged errors have certain mechanisms for identifying, estimating
and possibly eliminating or at least alleviating them
Unacknowledged errors have no standard procedures for detecting them
and may remain undiscovered causing a lot of harm
Local errors refer to solution errors at a single grid point or cell
Global errors refer to solution errors over the entire flow domain
Local errors contribute to the global error and may move throughout the grid.

Classification of errors
Acknowledged errors
Physical modeling error due to uncertainty and deliberate simplifications
Discretization error approximation of PDEs by algebraic equations
spatial discretization error due to a finite grid resolution
temporal discretization error due to a finite time step size
Iterative convergence error which depends on the stopping criteria
Round-off errors due to the finite precision of computer arithmetic
Unacknowledged errors
Computer programming error: bugs in coding and logical mistakes
Usage error: wrong parameter values, models or boundary conditions
Awareness of these error sources and an ability to control or preclude the
error are important prerequisites for developing and using CFD software

Verification of CFD codes

Verification amounts to looking for errors in the implementation of the models
(loosely speaking, the question is: are we solving the equations right?)
Examine the computer programming by visually checking the source code,
documenting it and testing the underlying subprograms individually
Examine iterative convergence by monitoring the residuals, relative changes
of integral quantities and checking if the prescribed tolerance is attained
Examine consistency (check if relevant conservation principles are satisfied)
Examine grid convergence: as the mesh and/or and the time step are
refined, the spatial and temporal discretization errors, respectively, should
asymptotically approach zero (in the absence of round-off errors)
Compare the computational results with analytical and numerical solutions
for standard benchmark configurations (representative test cases)

Validation of CFD models

Validation amounts to checking if the model itself is adequate for practical purposes
(loosely speaking, the question is: are we solving the right equations?)
Verify the code to make sure that the numerical solutions are correct.
Compare the results with available experimental data (making a provision for
measurement errors) to check if the reality is represented accurately enough.
Perform sensitivity analysis and a parametric study to assess the inherent
uncertainty due to the insufficient understanding of physical processes.
Try using different models, geometry, and initial/boundary conditions.
Report the findings, document model limitations and parameter settings.
The goal of verification and validation is to ensure that the CFD code produces
reasonable results for a certain range of flow problems.

Available CFD software

ANSYS CFX

http://www.ansys.com

commercial

FLUENT

http://www.fluent.com

commercial

STAR-CD

http://www.cd-adapco.com

commercial

FEMLAB

http://www.comsol.com

commercial

FEATFLOW

http://www.featflow.de

open-source

As of now, CFD software is not yet at the level where it can be blindly used by
designers or analysts without a basic knowledge of the underlying numerics.
Experience with numerical solution of simple toy problems makes it easier to
analyze strange looking simulation results and identify the source of troubles.
New mathematical models (e.g., population balance equations for disperse
systems) require modification of existing / development of new CFD tools.

Structure of the course

1. Introduction, flow models.
2. Equations of fluid mechanics.
3. Finite Difference Method.
4. Finite Volume Method.
5. Finite Element Method.
6. Implementation of FEM.
7. Time-stepping techniques.
8. Properties of numerical methods.
9. Taylor-Galerkin schemes for pure convection.
10. Operator-splitting / fractional step methods.
11. MPSC techniques / Navier-Stokes equations.
12. Algebraic flux correction / Euler equations.

Literature
1. CFD-Wiki

http://www.cfd-online.com/Wiki/Main Page

2. J. H. Ferziger and M. Peric, Computational Methods for Fluid Dynamics.

Springer, 1996.
3. C. Hirsch, Numerical Computation of Internal and External Flows. Vol. I
and II. John Wiley & Sons, Chichester, 1990.
4. P. Wesseling, Principles of Computational Fluid Dynamics. Springer, 2001.
5. C. Cuvelier, A. Segal and A. A. van Steenhoven, Finite Element Methods and
Navier-Stokes Equations. Kluwer, 1986.
6. S. Turek, Efficient Solvers for Incompressible Flow Problems: An Algorithmic
and Computational Approach, LNCSE 6, Springer, 1999.
7. R. Lohner, Applied CFD Techniques: An Introduction Based on Finite Element
Methods. John Wiley & Sons, 2001.
8. J. Donea and A. Huerta, Finite Element Methods for Flow Problems. John
Wiley & Sons, 2003.