How To Run A Steady Case On A Creeper

(1)

What is OpenFOAM? Using OpenFOAM Practice - Cylinder in uniform flow Summary

Crash Course Introduction to OpenFOAM

Artur Lidtke

University of Southampton [email protected]

(2)

Content

What is OpenFOAM?

The Idea

Overview of the libraries

Using OpenFOAM

Cavity tutorial overview Creating own simulations

Practice - cylinder in uniform flow

A look at the test case Running the simulation

Summary

(3)

(4)

The Idea

In a nutshell

I Free, open source set of libraries used to solve differential equations

I Primarily focused on Computational Fluid Dynamics (CFD), but may also be applied to other problems

I Uses finite volume approach to discretise and solve the governing equations

I Distributed by the OpenFOAM Foundation and developed by

OpenCFD Ltd.

I Diclaimer: This series of presentations and tutorials is unofficial and is not supported or endorsed by the owners of OpenFOAM

(5)

The Idea

Open source

I Significant amount of development done by the users. See for instance:

I swak4foam

I pyFoam

I OpenFOAM Extend (e.g. OF-1.7.x)

I Individual contributions (solvers, utilities) uploaded by single users

I Written in C++ - efficient, versatile and well structured code

I Difficult to get started with development unless a proficient C++ user with experience

(6)

The Idea

Perfect?

I Little official support - training sessions organised but not free

I Not much coherent training materials - typically need to go through many presentations, reports and forum posts to find the answer to a particular problem

I User has full control over most of the parameters of any operation - overwhelming at the beginning

I BUT: Many tutorial cases prepared by the developers and users - demonstrate how different features of the code work (although without too much narration)

I Intensive unofficial support through the on-line user forum:

http://www.cfd-online.com/Forums/openfoam/

I Wiki pages dedicated to OpenFOAM:

http://openfoamwiki.net/index.php/Main_Page

(7)

The Idea

Perfect?

I Little official support - training sessions organised but not free

I Not much coherent training materials - typically need to go through many presentations, reports and forum posts to find the answer to a particular problem

I User has full control over most of the parameters of any operation - overwhelming at the beginning

I BUT: Many tutorial cases prepared by the developers and users - demonstrate how different features of the code work (although without too much narration)

I Intensive unofficial support through the on-line user forum:

http://www.cfd-online.com/Forums/openfoam/

I Wiki pages dedicated to OpenFOAM:

(8)

Overview of the libraries

OpenFOAM capabilities

I See the official page for full info:

http://www.openfoam.org/features/

I What we are interested in:

I Incompressible, compressible and multiphase flows

I Reynold’s Averaged Navier Stokes (RANS), Unsteady-RANS, and Large Eddy Simulation turbulence modelling

I Built-in meshing tools

I Easy to use (and modify) tools for extraction of data for post-processing

I Readilly available plugins to ParaView for flow visualisation

I Fully paralellised solvers (and some utilities)

(9)

(10)

Cavity tutorial overview

Connecting to the cluster

I We will work solely on Iridis to ensure everybody has access to what they need

I Connect to the cluster using ssh

I Now your terminal is able to do things in Southampton

I All the files you will need are already here

I As a reminder, in order to copy some directories to the supercomputer

(11)

Connecting to the cluster

(12)

Connecting to the cluster

Use

ssh -X $YOUR USERNAME@iridis4 a.soton.ac.uk

(13)

Connecting to the cluster

(14)

Connecting to the cluster

(15)

Connecting to the cluster

(16)

Connecting to the cluster

Use

scp -r $SOURCE FOLDER \

$YOUR USERNAME@iridis4 a.soton.ac.uk:/home/\ $TARGET PATH

(17)

Let’s have a look at how to make OpenFOAM work

I Cavity tutorial - covered as one of the three cases in the official user guide:

http://openfoam.org/docs/user/cavity.php

I Very simple, lid-driven cavity flow - basic CFD validation test case, much like the Hello World! code for programming

I Worth going through the official tutorial as well as these slides not to miss anything

I Each OpenFOAM case is defined by a particular file structure in a separate folder

(18)

Let’s have a look at how to make OpenFOAM work

Go to

cd $FOAM RUN/tutorials/incompressible/icoFoam/cavity

(19)

Let’s have a look at how to make OpenFOAM work

(20)

What defines a case

I The very basic elments of an OpenFOAM simulation are:

I System directory - used for controlling the utilities and solvers

I Constant directory - holds the information about the case which (usually) doesn’t change over time

I 0 folder - contains the boundary conditions (separate file for each field, or flow variable)

(21)

What defines a case

I _{controlDict - main solution control: time step, start/end}

times, data extraction, additional library linking, etc.

I fvSchemes - describes the discretisation adopted for all quantities solved

I fvSolution - provides information as to how each of the equations is solved, given its discretisation

(22)

What defines a case

I polyMesh - contains the files describing the mesh

I transportProperties - description of the fluid and other associated parameters (viscosity, density, phase change parameters, etc.)

I turbulenceProperties - how the turbulence is modelled (no surprise there!)

(23)

What defines a case

(24)

Running the case

I First, we need to create the structure mesh using a ready dictionary file

I Let us see what we have achieved

I The boundary conditions are already set, so are the solver parameters. Just run the relevant application:

I Now it’s time to check the solution

(25)

Running the case

Use

blockMesh

(26)

Running the case

(27)

Running the case

Use

I Typical command is: paraFoam

I But we haven’t compiled the reuired ParaView libraries

I Convert solution to ParaView format foamToVTK -useTimeName

I Start ParaView and open the VTK file paraview

I Select ”Surface with edges” appearance

(28)

Running the case

(29)

Running the case

(30)

Running the case

(31)

Running the case

Use

icoFoam

(32)

Running the case

(33)

Running the case

Use

I foamToVTK -latestTime -useTimeName

I paraview

I Load the data

I Select ”Surface” appearance

(34)

Running the case

(35)

(36)

Creating own simulations

How to create a bespoke case

I Tutorial cases cover many different flows, configurations and variants - something for everyone!

I To make life easier:

I _{Copy and existing tutorial case similar to what you want to}

simulate

I Change the mesh

I Alter the flow parameters

I Add any post-processing

I Done!

(37)

(38)

A look at the test case

Oveview

I Cylinder in uniform flow - relevant to most engineering problems

I Very well covered in the literature, countless sets of validation data

I Very complex flow despite simple geometry

I Several regimes of the flow depending on the Reynolds number: laminar, sub-critical, transitional and fully turbulent

I For the purpose of this workshop, let’s look at sub-ciritical regime - low-Re but not low enough to be boring

(39)

Oveview

I Reynolds number is the ratio of inertial to viscous forces

I Given by

Re =

UD_ν

I Tells us how turbulent the flow is likely to be (the higher the more turbulence)

(40)

Oveview

(41)

Conditions

I Chosen to repeat the experiments by Parnaudeau et al. (2008)

http://www.irisa.fr/fluminance/team/Carlier/ publications/ParnaudeauCarlierHeitzLamballais_ 2008_POF.pdf

I Rich source of information, detailed and well presented data

I Re = 3900

I Expect a largely laminar boundary layer separating approximately half-way around the cylinder

I This gives rise to vortices being shed and turbulent wake appearing

(42)

What we want to capture

|ω| = |∇ × U| = 10 s−1

(43)

Running the simulation

Getting the initial state - mesh

I It is a good idea to get an approximate steady-state solution of the flow beforehand

I This tries to capture the mean flow without resolving the turbulence explicitly

I Let us do just that:

(44)

Getting the initial state - mesh

I Let us create and view the mesh

(45)

Getting the initial state - mesh

Go to

cd $FOAM RUN/cylinder2D

(46)

Getting the initial state - mesh

(47)

Getting the initial state - mesh

Use

(48)

Getting the initial state - mesh

Do

Visualise the mesh as for the cavity

(49)

(50)

Getting the initial state - solution

I To run a steady case RANS it’s also good to run potential flow first

I The boundary conditions are set but are protected from overwriting

I Now, run the potential flow solver

I In system/controlDict uncomment functions part to

enable run-time postprocessing

I Decompose the case to run in parallel

I Run the steady-state solver in parallel

(51)

Getting the initial state - solution

I For F. Herbert fans:

Solutions, within solutions, within solutions...

(52)

Getting the initial state - solution

(53)

Getting the initial state - solution

Use

cp -r 0.org 0

(54)

Getting the initial state - solution

(55)

Getting the initial state - solution

Use

potentialFoam

(56)

Getting the initial state - solution

(57)

Getting the initial state - solution

(58)

Getting the initial state - solution

Use

decomposePar

(59)

Getting the initial state - solution

(60)

Getting the initial state - solution

Use

foamJob -s -p simpleFoam

(61)

Post-processing

I Reconstruct the parallel solution

I Clean-up the decomposed directories

(62)

Post-processing

Use

reconstructPar -latestTime

I View the solution

(63)

Post-processing

(64)

Post-processing

Use

rm -r processor*

I View the solution

(65)

Post-processing

(66)

Post-processing

I View the solution

Do

I Load the case to ParaView

I Visualise the x-component of the velocity field

(67)

(68)

Steady state solution - really?

(69)

Steady state solution - really?

I This is NOT a steady state solution!

I We ran it for long enough to expect at least partial convergence

I So, what could be the problem?

I We applied a steady-state solver to an inherently unsteady problem - cannot get real convergence

I Still OK for a rough guess of initial condition but meaningless physically

(70)

Steady state solution - really?

(71)

Steady state solution - really?

(72)

Preparing to run the LES case

I Navigate to the case

I Create the mesh

I Copy the initial conditions

I Map the steady-state solution onto the 3D grid

(73)

Preparing to run the LES case

Go to

cd $FOAM RUN/cylinder3D pimple LES

I Create the mesh

(74)

Preparing to run the LES case

I Create the mesh

(75)

Preparing to run the LES case

I Create the mesh

Use

blockMesh

(76)

Preparing to run the LES case

I Create the mesh

(77)

Preparing to run the LES case

I Create the mesh

Use

cp -r 0.org 0

(78)

Preparing to run the LES case

I Create the mesh

(79)

Preparing to run the LES case

I Create the mesh

Use

(80)

Basics of using a supercomputer

I In order to use the full potential of the cluster we need to submit a job

I Now we are using the login node, not the nodes that are acutally use computation

I What happens during the job execution is defined in a control script

I Let us see what that looks like

(81)

Basics of using a supercomputer

I In order to use the full potential of the cluster we need to submit a job

I Now we are using the login node, not the nodes that are acutally use computation

I What happens during the job execution is defined in a control script

I Let us see what that looks like

Edit

(82)

Job control script - controlling the cluster

# !/ bin / b a s h # PBS - S / bin / b a s h # M e m o r y n e e d e d by the job # PBS - l mem =25 GB # T i m e a l l o c a t e d # PBS - l w a l l t i m e = 6 0 : 0 0 : 0 0 # N u m b e r of n o d e s and p r o c e s s o r s per n o d e # PBS - l n o d e s =8: ppn =16 # G i v e the job a n a m e # PBS - N C y l _ L E S _ s m a g # M P I _ B U F F E R _ S I Z E = 1 2 0 0 0 0 0 0 0

(83)

Job control script - executing commands

# C h a n g e to o r i g i n a l w o r k i n g d i r e c t o r y cd $ P B S _ O _ W O R K D I R # p r e p a r e to run in p a r a l l e l d e c o m p o s e P a r > log . d e c o m p o s e P a r 2 >&1 # run the s o l v e r m p i r u n - np 128 p i m p l e F o a m - p a r a l l e l > log . i r i d i s _ r u n 2 >&1

(84)

Submitting and viewing the job

I Submit the job to the queue

I To view the jobs currently queuing, running and recently cancelled

I Now, all we need to do is wait and hope the simulation doesn’t crash!

(85)

Submitting and viewing the job

Use

qsub Allrun

(86)

Submitting and viewing the job

(87)

Submitting and viewing the job

Use

qstat -u $YOUR USERNAME

(88)

Submitting and viewing the job

(89)

Some useful commands

I qsub - submit a job to the queue (the script is stored in memory so any changes made to it after submission won’t be applied)

I qstat - view your jobs, option -u is particularly useful

I showstart $JOB ID - get an estimated time required for a job to start

I showq - view the status of the queue, useful to redirect the output to a text file: showq > log.queue

(90)

Summary

(91)

Conclusions

I We’ve learned how OpenFOAM case is structured, how to

mesh and run a simple case

I The idea of introducing an initial solution for a Large Eddy Simulation case has been discussed

I Brief introduction to using a supercomputer to do CFD has been delivered

I A potential pitfal of using an incorrect solver type for a particular problem has been indicated

(92)