Best practices in CFD

©ZeusNumerix
Defense | Nuclear Power | Aerospace | Infrastructure | Industry
Practices to be followed to get
reasonable results
Abhishek Jain
abhishek@zeusnumerix.com
Best Practices

©ZeusNumerix
2
Overview
 Zeus Numerix: Introduction
 Assumption
 What is a Best Practice
 Surface Grid
 Volume Grid
 Sample Problems
 Pressure Drop
 Skin Friction
 Heat driven flows
 Initial and Boundary Conditions
 General Gyan

©ZeusNumerix
Assumption
 Participants know what CFD is about
 Know the basics of Compressible and Incompressible flow
 Know what are Initial Conditions and Boundary Conditions
 Know about Meshes
 Surface Mesh
 Volume Mesh
 Clustering
 Smoothening
 Have basic idea of Fluid Mechanics
3

©ZeusNumerix
Best Practices
 Best Practice is a misnomer
 It should be good practices
 Before touching computer know
 Problem statement
 Aim of the problem
 Accuracy required
 Computing power available
 No blind trust on the software
 Validation
 http://www.grc.nasa.gov/WWW/wind/valid/tutorial/glossar
y.html (a must read for all)
4

©ZeusNumerix
GRID GENERATION
Rules to be followed in grid generation
5

©ZeusNumerix
Grid Generation Essentials
 Knowledge of Physics is essential
 Grid for pressure drop estimation is very different from heat
flux estimation
 Accuracy required should be understood
 Preliminary design calculations or design improvement
 Very accurate may not be cost effective
 Will the grid be moving, if yes where
 Parameters that may be changed in design
 Good grid is half CFD done
 Bad grid is full CFD repeated
6

©ZeusNumerix
Grid contd…
 What size of mesh is good mesh size
 For any new problem or new software grid convergence has
to be repeated
 Recipe for failing
 This software is well established, it will work
 Physics may be different but this software works
 Automatic mesh will be good
 Grid read by solver so it must be OK
 Make a large mesh and coarsen it, till you get 2 meshes
with reasonably identical results
7

©ZeusNumerix
Grid contd…
 Always smoothen a grid before doing any other operation
 A grid parallel and perpendicular to the flow is the best grid
 Clustering must never precede smoothening
 More clustering makes convergence slower
 Less clustering does not resolve flow features
 5-10 grid points in boundary layer based on Reynolds
number
 Refinement of grid at sharp corners, bends and locations
where shocks are expected
 Check – Skewness, smoothness and aspect ratio
8

©ZeusNumerix
Grid contd…
 Coordination with solver
 Velocity regime
 Euler, Laminar or Turbulent flow
 Which turbulent model
 Proper labeling and boundary conditions
 What BCs are required
 Proper labels to various components
 Units must be specified when grid is made
 If not made in SI units, COMMUNICATE
9

©ZeusNumerix
SIMPLE PROBLEMS
Problems requiring less accuracy and attention
10

©ZeusNumerix
Problem Types
 Problem desiring less accuracy
 Pressure drop across regions
 Supersonic flow – lift estimation
 Preliminary design calculations
 Design of low cost mechanical equipment
 High rise buildings
 Ventilation of huge spaces
 These problems require only estimation of gross number
 Spatial resolution of physical properties may not be
accurate, however gross properties are reasonably accurate
11

©ZeusNumerix
Case Study: High Rise
 Aim 1: Estimation of wind loads on a high rise building
 Aim 2: Estimation of discomfort due to air flow on balconies
 Data required
 Building drawings – usually given in 2D format
 Wind data – Collection of data of metrology dept
 Data on nearby buildings
 Comfort data
 Step 1: Selection of number of simulation
 Select severest wind conditions for three seasons
 Also select any other worst case scenario as seen in the
metrology table
12

©ZeusNumerix
CFD of Buildings
 Input is received in format as shown below
 Convert the format to 3D using Revit® or other tools
 Remove features that are small compared to the building
size like grill, ventilators, balcony designs and make them
flat
13
Buildingisusuallyrepresentedin
AutoCAD®format

©ZeusNumerix
Meshes
 Shown are meshes and cleaned building model as seen by
CFD
 Mesh very coarse even near wall
 Good mesh = billions of cells 
14

©ZeusNumerix
Simulation Table
 Following Simulations are selected from the Metrology data
 Data measured near Colaba by Metrology dept
 Site near the sea hence relative humidity is high
 Ground Boundary layer is approximated when performing
simulation
15
North-WestWestNorth-WestEastWind Direction
43.264.825.227Wind Speed (Kmph)
70877570Relative Humidity
37.434.842.236.2Temperature (C)
NovemberJulyAprilJanuary

©ZeusNumerix
Results
 Load on the whole building is integrated by adding
individual forces on surface cells
 Major focus is given on the vortices formed in the buildings
16

©ZeusNumerix
Results
 Vortices formed near NW apartments will be uncomfortable
 Construction of small structures suggested to suppress
them
17
Vortices

©ZeusNumerix
ACCURACY MEDIUM
Problems requiring care in making mesh and simulation
18

©ZeusNumerix
Problem Types
 Problem desiring higher accuracy
 Flow around automobiles
 Flow around aircraft and missiles
 Flow in process industry – cyclone separator, piping, ducts
 These problems require only estimation of properties at
specific locations with reasonable accuracy
 Grid generation requires attention
19

©ZeusNumerix
Case Study: Flow Past Automobile
 Aim 1: Estimation of aerodynamic forces on an automobile
 Aim 2: Estimation of discomfort due to dust ingress
 Data required
 IGES file of the automobile
 Wind Conditions for the problem
 Step 1: Cleaning of CAD data
 Remove the components that are small in size
 Close the gaps of handle, doors
 Assumptions
 Tyre rotation effect is not modeled
 Dust as particles is not modeled
20

©ZeusNumerix
CFD of Automobile
 Input is received in format as shown below
 Remove features that are small compared to the vehicle
size like grill, door handle , headlight protrusions etc
21
CADDrawingofanSUV
Withallfeatures
15000components

©ZeusNumerix
Grid Generation
 CAD is cleaned to remove surfaces not exposed to
aerodynamics or insignificant surfaces
 Surface is divided into patches for generation of hex grid
 Clustering done to fraction of mm to capture boundary layer
22
Domain of Analysis for MUV
MUV Surface
MUV Block Surface

©ZeusNumerix
Grid
 View showing clustered grid at the surface
23
Grid density is
high near vehicle
surface

©ZeusNumerix
Results
 Streamlines are important; seeds are correct locations
 Dust ingress will be known by streamlines curving inside
24
Path lines seen to
point inward

©ZeusNumerix
Problem Types
 Problem desiring highest accuracy
 Aerothermal considerations in high speed flow
 Heat driven flows
 Reactive flows
 Turbomachinery (rotating flows)
 These problems require estimation of properties accurately
to serve the aim of simulation
 Grid generation requires high degree of care as the
simulations are very sensitive
27

©ZeusNumerix
Heat Flux
 Aim 1: Estimation of heat flux of a missile
 Aim 2: Estimation of aerodynamic coefficients
 Data required
 IGES file of the missile
 Atmospheric Conditions for the problem
 Isothermal temperature of the wall
 Step 1: Cleaning of CAD data
 Remove the components that are small in size
 Close the gaps of missile and fins etc
 Assumptions
 Air remains calorifically perfect
28

©ZeusNumerix
Grid Generation
 Surface grid chosen after extensive analysis
 Requirement of y+ is within bounds of 1-5 at all places
 Since the flow is very high speed, first cell distance has to
be 1 micron from the surface for the given mach number
 VALIDATION is a must before attempting these problems
29
Comparisonofresultsfor
variousmeshes.151and201
havesameresults

©ZeusNumerix
Results
 Variation of y+ on the surface
 Variation of heat flux on the surface
 * Heating more in wing than nose
30
Mach Number
variation
High
Low
High heat flux regions
Heat Flux
Spalart-Allmaras Turbulence
Model, HLLC Scheme
y+ range 1-5 at all places

©ZeusNumerix
Initial Conditions
 It is a wrong assumption that whatever conditions you give
flow will finally converge to a correct solution
 Sometimes wrong initial condition may have a physical
meaning
 Constant Initial Conditions throughout the domain may not
work always
 Most obvious initial conditions may not be the fit for certain
solvers
32

©ZeusNumerix
Example
 Consider a case of supersonic CD nozzle with the three ICs
 Flow at zero velocity inside domain
 Flow supersonic everywhere
 Flow supersonic in converging section and gradually decreased
to subsonic till the exit
 The representation physically
 Flow shock will enter inside and form a normal shock and
subsequent flow will be subsonic throughout
 Flow has been made supersonic everywhere by use for
external machines
 Flow has been made supersonic and now exiting in
atmosphere
33

©ZeusNumerix
Boundary Conditions
 A pressure and velocity boundary is not the same as mass
flow boundary for incompressible flow
 Signal traveling backward may change mass flow rate
 Farfield and outlet are not the same for compressible flow
 Farfield means waves are not affected by flow
 Isothermal wall is significantly different than adiabatic,
default expression is adiabatic wall
34

©ZeusNumerix
General
 Better to use standard format like CGNS for interoperability
 Research says lot of time wasted in converting file formats
 Aim not written at the beginning is a sure recipe for
disaster
 Solution should be doable in reasonable cost
 Try to arrange a computer that can handle the problem
instead of taking shortcuts and solving on existing resource
 CDAC gives good access to computing power for academics
 Make a checklist of activities if same type of problem may
come all the time
 Usually engineers make a mistake in routine simulations due
to over confidence
 More engineers doesn’t mean faster results; reverse
possible
35

Best practices in CFD

More Related Content

What's hot (20)

Viewers also liked (10)

Similar to Best practices in CFD (20)

Recently uploaded (20)

Best practices in CFD