32
The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL Intro to Computational Fluid Dynamics Brandon Lloyd COMP 259 April 16, 2003 Image courtesy of Prof. A. Davidhazy at RIT. Used without permission.

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL Intro to Computational Fluid Dynamics Brandon Lloyd COMP 259 April 16, 2003 Image courtesy of Prof. A

  • View
    212

  • Download
    0

Embed Size (px)

Citation preview

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Intro to Computational Fluid Dynamics

Brandon LloydCOMP 259April 16, 2003

Image courtesy of Prof. A. Davidhazy at RIT. Used without permission.

2

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Overview

• Understanding the Navier-Stokes equations­ Derivation (following [Griebel 1998])­ Intuition

• Solving the Navier-Stokes equations­ Basic approaches­ Boundary conditions

• Tracking the free surface

3

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Operators - gradient div - divergence2- Laplacian… - Hand waving / Lengthy

math compression

Foundations

yu

xuuu

div

2

2

2

22

yu

xuu

y

ux

uu ∂∂,∂

4

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Transport Theorem

xdtxufft

xdtxfdt

dt t

),()div(),(

),(),( tct

txu

),( tcx

c

0

),( tc

t

5

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Conservation of Mass

density­is­;­­­­­­­­),()0,(mass0

t

xdtxxdx

0),()div(),(

xdtxu

txdtx

dt

dt t

0)div(

ut

Transport theorem

0div u

Integrand vanishes

is constantfor incompressiblefluids

Continuity equation

6

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Conservation of Momentum

t

xdtxutx

),(),(momentum

forces­actingmomentumin­­change

t

xdtxftx

),(),(­­:forcesbody­

t

dsntx),(­­:forces­surface σ

ttt

dsntxxdtxftxxdtxutxdt

d ),(),(),(),(),( σ

normal­:­­tensor­stress­: n

σ

7

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Conservation of Momentum

0divdiv)())(()( σguuuuudt

d

Transport theorem Divergencetheorem

fupuudt

ud

2­1

)(

Momentum equation

ttt

dsntxxdtxftxxdtxutxdt

d ),(),(),(),(),( σ

8

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Navier-Stokes Equations

fupuudt

ud

u

2­1

)(

0

convection viscosityexternalforcespressure

9

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Solving the equations

Basic Approach1. Create a tentative velocity

field.a. Finite differencesb. Semi-Lagrangian method (Stable

Fluids [Stam 1999])

2. Ensure that the velocity field is divergence free:

a. Adjust pressure and update velocitiesb. Projection method

10

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Tentative Velocity Field

Finite differences – mechanical translation of equations.

n

11

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Tentative Velocity Field

Limits on time step• CFL conditions – don’t move

more than a single cell in one time step

• Diffusion term

ytvxtu maxmax ­­­­­­,

1

22

112

yx

t

12

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Tentative Velocity Field

Stable Fluids Method1. Add forces: 2. Advection3. Diffusion

)()()(~01 xftxuxu

13

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Tentative Velocity Field

AdvectionFinite differences is unstable for large Δt.Solution: trace velocities back in time.

Guarantees that the velocities will never blow up.

)),((~)(~12 txpuxu

14

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Tentative Velocity Field

DiffusionDiscretizing the viscosity term spreads velocity

among immediate neighbors. Unstable when time step too small, grid spacing too large, or viscosity is high.

Solution: Instead of using an explicit time step use an implicit one.

This leads to a large but sparse linear system.

)(~)(~)( 232 xuxut

I

15

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Satisfying the Continuity Eq.The tentative velocity field is not necessarily

divergence free and thus does not satisfy the continuity equation.

Three methods for satisfying the continuity equation:

1. Explicitly satisfy the continuity equation by iteratively adjusting the pressures and velocities in each cell.

2. Find a pressure correction term that will make the velocity field divergence free.

3. Project the velocities onto their divergence free part.

16

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Explicitly Enforcing u=0

Since we have not yet added the pressure term, we can use pressure to ensure that the velocities are divergence free.

u>0 increased pressure and subsequent outfluxu<0 decreased pressure and subsequent influx

Relaxation algorithm1. Correct the pressure in a cell2. Update velocities 3. Repeat for all cells until each has u<ε

17

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Solving for pressure

Another approach involves solving for a pressure correction term over the whole field such that the velocities will be divergence free and then update the velocities at the end.

)1()()1( 1~

nnn

pt

uu

dt

ud

)1()()1( ~

nnn pt

uu

0~ )1(2)()1(

nnn pt

uu

Discretize in time

Rearrange terms

Satisfy continuity eq.

18

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Solving for pressure

We end up with the Poisson equation for pressure.

This is another sparse linear system. These types of equations can be solved using iterative methods.

Use pressures to update final velocities.

)()1(2 ~ nn ut

p

)1()()1( ~

nnn pt

uu

19

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Projection Method

The Helmholtz-Hodge Decomposition Theorem states that any vector field can be decomposed as:

where u is divergence free and q is a scalar field defined implicitly as:

We can define an operator P that projects a vector field onto its divergence free part:

­­quw

­­2qw

­­qwwu

20

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Projection Method

Applying P to both sides of the momentum equation yields a single equation only in terms of u:

Thus for the last step :

Look familiar? The scalar field q is actually related to pressure!

quuqu ~­­­­­­~ 2

)­)(P( 2 fuuudt

ud

)1()()1( ~

nnn pt

uu

)()1(2 ~ nn ut

p

21

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

The Bottom Line

All three methods are equivalent!

22

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Boundary Conditions

• No slip: Set velocity to 0 on the boundary. Good for obstacles.

• Free slip: Set only the velocity in the direction normal to the boundary to zero. Good for setting up a plane of symmetry.

• Inflow: Specified positive normal velocity. Good for sources.

• Outflow: Specified negative normal velocity. Good for sinks.

• Periodic: Copy the last row and column of cells to first row and column. Good for simulating an infinite domain.

23

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Staggered GridThe staggered grid provides

velocities immediately at cell boundaries, is convenient for finite differences, and avoids oscillations.

Consider problem of a 2D fluid at rest with no external forces. The continuous solution is:

On a discretized non-staggered grid you can have:

jip , jip ,1jip ,1

1, jip

1, jip

21, jiv

21, jiv

jiu ,21 jiu ,2

1

constant­­­­­­­­0­­­­­­0 pvu

odd­for­­­even,­for­­­­

­­­­­­0­­­­­­0

21,

,,

jiPjiPp

vu

ji

jiji

24

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Tracking the Free SurfaceThe movement of the free surface

is not explicit in the Navier-Stokes equations.

Three methods for tracking the free surface:

1. Marker and cell (MAC) method2. Front tracking3. Particle level set method

25

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

MAC

Due to [Harlow and Welch 1965].Track massless marker particles to

determine where the free surface is located.

Markers are transported according to the velocity field.

Cells with markers are fluid cells. Fluid cells bordering empty cells are surface cells.

There are boundary conditions that must be satisfied at the surface.

Extended by [Chen et al. 1997] to track particles only near the surface.

26

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

MAC

Problems:• Can lead to mass dissipation, especially with

stable fluid style advection.• No straight forward way to extract a smooth

surface.

Image from [Griebel 1998].

27

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Front Tracking

Proposed by [Foster and Fedkiw 2001]Front tracking uses a combination of a level set

and particles to track the surface. The particles are used to define an implicit

function. An isocontour of this function represents the liquid surface.

The isocontour yields a smoother surface than particles alone.

28

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Front Tracking

Using the level set method, the isocontour can be evolved directly over time by using the fluid velocities.

Particles and level set evolution have complementary strengths and weaknesses­ Level set evolution suffers volume loss ­ Particles can cause visual artifacts­ Level sets are always smooth.­ Particles retain details.

29

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Front Tracking

Combine the two techniques by giving particles more weight in areas of high curvature. Particles escaping the level set are rendered directly as splashing droplets.

30

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Particle Level Set MethodPresented by [Enright et

al 2002].Implicit surface loses

detail on coarse grids.Particles keep the

surface from crossing them but can’t keep it from drifting away.

Add particles to both side of the implicit surface.

Escaped particles indicate the location of errors in the implicit surface so it can be rebuilt.

31

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

Particle Level Set MethodExtrapolated velocities at the surface

give more realistic motion.

32

The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL

References

CHEN, J., AND LOBO, N. 1994. Toward interactive-rate simulation of fluids with moving obstacles using the navier-stokes equations. Computer Graphics and Image Processing, 107–116.

CHEN, S., JOHNSON, D., RAAD, P. AND FADDA, D. 1997. The surface marker and micro cell method. International Journal of Numerical Methods in Fluids, 25, 749-778.

FOSTER, N., AND METAXAS, D. 1996. Realistic animation of liquids. Graphical Models and Image Processing, 471–483.

FOSTER, N., AND FEDKIW, R. 2001. Practical animation of liquids. In Proceedings of SIGGRAPH 2001, 23–30.

GRIEBEL, M., DORNSEIFER, T., AND NEUNHOEFFER, T. 1998. Numerical Simulation in Fluid Dynamics: A Practical Introduction. SIAM Monographs on Mathematical Modeling and Computation. SIAM

KASS, M., AND MILLER, G. 1990. Rapid, stable fluid dynamics for computer graphics. In Computer Graphics (Proceedings of SIGGRAPH 90), vol. 24, 49–57.

O’BRIEN, J., AND HODGINS, J. 1995. Dynamic simulation of splashing fluids. In Proceedings of Computer Animation 95, 198–205.

STAM, J. 1999. Stable fluids. In Proceedings of SIGGRAPH 99, 121-128.