This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

Anirban Guha April 28, 2009

∗

Abstract

The present paper deals with the modeling and validation of a 2-D laminar, incompressible, viscous lid-driven cavity ow. Proper choices of articial compressibility, articial viscosity and over-relaxation parameters were made. Both square cavity and a cavity with aspect ratio of 3:1 were analyzed. Vortex formation and its dependence on Reynolds number as well as geometry were studied. Quantitative estimations of the location of the eye of vortex and its strength were made.

Introduction

Lid driven cavity ow is a class of internal, bounded ow of an incompressible, viscous, Newtonian uid in which the motion is generated by a portion of the containing boundary. If L and U be the associated length and velocity scales respectively, then the normalized continuity and Navier Stokes equations can be written as

**Figure 1: Schematic of a 3-D lid driven cavity ow [3] here is
**

2-D Cartesian.

•

Time Dependence: Fluid Properties:

The ow properties are dependent on time, hence the problem needs to be modeled as an unsteady problem. The uid is incompressible, visand Newtonian. The uid properties like density and viscosity are assumed to be constant.

cous

.u = 0

(1)

•

∂u + (u. )u = − p + Re−1 2 u (2) ∂t where Re = U L/ν is the Reynolds number. The boundaries have no-slip conditions and the initial condition corresponds to a quiescent uid. The usefulness of studying driven cavity ows is that they exhibit almost all phenomena that can possibly occur in incompressible ows, for example eddies, secondary ows, complex patterns, instabilities, transition and turbulence. Figure 1 shows the schematic diagram of a driven cavity ow. The gure shows that in the plane of symmetry, there is a primary eddy and also secondary eddies at the corners, two at upstream and one at downstream. Our interest is to develop a CFD code which can correctly simulate the ow physics of a 2-D lid driven cavity ow.

•

Turbulence:

The ow can become transitional or turbulent depending on the Reynolds number. Since we are interested in the laminar characteristics of the ow, we work at a much lower Reynolds number so that laminar ow can be safely assumed.

•

Energy:

The uid is treated as an isothermal uid and there are no sources or sinks involved. Hence energy equation can be safely neglected.

•

Body Forces:

Body forces like gravity are not important in understanding the ow physics, hence it can be safely neglected.

Modeling

•

We are interested in modeling a 2-D lid driven cavity ow, hence the geometry considered

**Code Verication and Validation
**

1. General Validation

1.1 Correctness of the residual

A square geometry of [0,1]×[0,1] with no slip boundary conditions was considered. The ow variables (P,u,v) 1

Geometry:

∗ Mechanical Engineering Department, University of British Columbia

u = A x y 2 P (5) machine round-o errors. The lid was set to zero velocity.0500260 0.2 Correctness of the Flux Jacobian pressure in the alternate lines of the mesh.0007490 u-velocity 0.999e − 12 v-Velocity 4.1000000 0. The ow variables (P. Also from Table 1b.u. This problem was ity=0 tacked by introducing an articial compressibility factor 2. multiplied by a smoothing factor −10 are less than 10 . i. The oscillaaccurate as expected for a second order scheme. After incorporating it in the code and verifying functions were the results with that posted in Vista. The Vista. Lid Driven Cavity Flow . it was observed that the result of rst pass of approximate factorization P0 cos (πx) cos (πy) P was dierent. 200 time steps ( number of time-steps for convergence it is evident that Flux Integral evaluation is second order was 520 with convergence criteria of 10−9 ).00 1.v velocities 2.999e − 12 Figure 2: Convergence history for the rst 200 time steps Table 2: Error in Implicit Flux Integral were initialized to some known smooth functions of x and y which satised the boundary conditions and this The Block Thomas Algorithm was supplied in UBC enabled us to calculate the ux integrals exactly.0125860 0.Lid Velocmomentum and continuity equations.0031510 (a) L2 norm of error continuity u.0500260 0.0479470 0.0125000 continuity 0. The code was able to take time-steps much larger than 0. Figure 2 shows the convergence history for the rst and v velocities had same residuals. Since the continuity equation for incompressible uid (Equation 1) doesn't have pressure term(s) in it. As expected.0125860 0.0500000 0. This may be because of the fact that the errors were close to ∂P 1 + .v) were initialized according to ∂P 1 + .1 Validation Case: Stability (β ) in the continuity equation.0250000 0.05 and 20 × 20 mesh. Because of geometric symmetry. tions in pressure (visible in Figure 3a) is a common issue with collocated grids because of de-coupling between 1.98 (b) Order of Accuracy Table 1: Flux Integral verication Pressure 8.3 Block Thomas Algorithm .e L2 norm of ux the boundaries were zero. there is no coupling between pressure and velocity between the 2. i. The pa∂t β rameters chosen were P0 = u0 = v0 = β = 1 .u = 0 (4) Equation 3. Since the velocities at all The correctness of the residuals. the initial condition was much integrals was calculated using a second order accurate like a disturbance in a conned uid which eventually centered ux evaluation scheme for Re = 10 and P0 = decayed with time to zero velocity everywhere in the dou0 = v0 = β = 1. 2 1. The maximum error even cells of the solution domain could be handled by obtained for Re = 10 and P0 = u0 = v0 = β = 1 and a introducing an articial viscosity term in the form of the 20×20 mesh is given in Table 3a. size and the errors were found to increase slightly.1950580 0.No.0031510 v-velocity 0. the errors Laplacian of pressure. the u main. tested by incorporating a small perturbation at the cenThe pressure oscillations.Re = 100 . t = 0.e decoupling of the odd and ter of the computational domain. The code was run for 40 × 40 mesh into the continuity equation.05 and for The function which evaluated the Flux Jacobian was the same conditions the biggest time-step was t = 1. hence physical justication of ∂t β this behavior might be incorrect.0029960 0.174e − 20 u-Velocity 4. of meshes 10X10 20X20 40X40 80X80 mesh size 0. This is because of the reason that the data u = u0 sin (πx) sin (2πy) (3) posted in Vista didn't include the factor of time step v0 sin (2πx) sin (πy) v which needs to be multiplied with the Flux Integrals to get accurate results.1950580 0.0119830 0.

0 4.5 3. the code converges to a wrong solution.2 Choice of Pressure Smoothing Factor(A) The value of A should not be very high since it incorporates a second order error in the system.0 40.A 0. The values are tabulated in (c) v-Velocity Table 3b and Figure 5b gives the diagramatic represenFigure 3: Contours of dierent ow properties with tion of the same.05.0 # of Time Steps 614 520 520 519 512 (a) Choice of A β 0.05.0 # Time Steps A=1 A=4 659 659 539 539 520 519 538 523 600 580 651 627 690 664 800 765 955 905 (b) Choice of β Table 3: Convergence history under dierent conditions (a) Pressure 2.0 0.0 10. the dependence of convergence rate on β was evaluated. From the results obtained. Intuitively this seems right and similarly.0 1. Also it was found that at higher values of β .5 2. For A = 1 and 4. for a 40 × 40 mesh with P0 = u0 = v0 = β = 1 .1. 2.4 Choice of Over-Relaxation Factor(ω) It is well known that over-relaxation accelarates convergence upto an optimal value.5 1. 3 . Since coupling is very important for numerical stability. For example. signicant coupling was achieved and the oscillations were essentially damped out. It was observed that increasing A increases the convergence rate too.0 5. it can be concluded that β = 1 is a reasonable choice for t=0. It is numerically implemented as Qn+1 = Qn + ωδQ.Re = 100 . From Figure 5a it was found that for a mesh of 40 × 40 and A = 1 the optimal value of β shifts to the right with increasing t. it is obvious that higher values of β provides additional stability while lowering it increases the incompressibility and thereby accelerate convergence.5 1. the reason of introducing the articial compressibility factor was stated. t = 0. since the oscillations are damped resulting into faster convergence. For rest of the code A = 4 was chosen as the optimal value. For A = 4. Finding the optimal value of β is critical.3 Choice of Articial Compressibility Factor(β ) In Sub-section 1. where Q is the solution.1 0. the convergence with convergence criteria of 10−9 is shown in Table 3a. Utop = 0 (b) u-Velocity 2. increasing the number of meshes will make the optimal value of β to shift to the left.0 2.β → 0 gives a singular continuity equation ( Equation 4) and the coupling is broken giving rise to instability.

0 (b) β for dierent A Figure 5: Choice of β from convergence history (d) A=40.(a) A=0.5 (b) A=1.0 (a) β for dierent t (c) A=4.0 Figure 4: Pressure Contours for dierent values of A 4 .

(b) v-Velocity 3.5 was used for Re=100 and ω = 1. ω = 1.1 Solution For β = 1. Thus a value of ω > 1 helps a faster update of the solution. The parameters which were kept constant were β = 1 . A = 4. Lid Driven Cavity Flow .For Re=1000.3 while at Re=100. the dierent ow properties and the convergence history is plotted in Figure 7 . the solution divereges at ω = 1. The optimum choice of ω being dependent on Re. The results were checked for mesh size of 40 × 40 and no signicant change in the convergence characteristics was observed.A = 4 . the solution will become unstable for ω > 2.Lid Velocity=1. The optimal value of ω was explored for two dierent Reynolds number.(a) Re=100 (a) u-Velocity (b) Re=1000 Figure 6: Choice of ω from convergence history n is the time-step number and δQ is the change in solution. t=0. ω = 1. 5 (c) Pressure with velocity vectors Figure 7: Contours of ow variables for Utop = 1 and 20 × 20 mesh . the divergence was detected for ω > 2.0 3. Re=100.5.1 was used for Re=1000. 100 and 1000.1.05 and mesh size of 20 × 20. Theoritically.mesh size of 20 × 20 and t = 0.Convergence criteria was set to 10−9 .

the curvature is maximum and it can be a test for dierent meshes to capture that curvature properly.2123895 Error(%) 30. the dierent ow properties are shown in Figure 10. This was checked numerically and the result is shown in Figure 9. the velocity at the point y = 0. Assuming that 160 × 160 has no error. the dierences in solution with mesh renement is clearly visible.2092910 -0. ω = 1.57 4. 3.3 Sanity Check-Symmetry Because of symmetry in the geometry.05 and mesh size of 80 × 80.2 Grid Convergence Study (b) u-velocity along mid-plane for 20 × 20 mesh The grid independence study is shown in Figure 8c. 6 . y) = 0.Mesh Size 10 × 10 20 × 20 40 × 40 80 × 80 160 × 160 Value -0. It is clearly visible that for 80 × 80 mesh. y)+uUtop=−1 (1−x.32 1. Comparing Figure 10 with Figure 7.46 - Table 4: Grid convergence study (a) Convergence history up to 10−6 Figure 9: Sanity Check-Symmetry 3.2032175 -0. the whole eld is zero for a calculation up to 10 places of decimal.1835620 -0.5. t=0. the value of the expression uUtop=1 (x. The dierent values of u-velocities for dierent meshes at y = 0. As expected. Convergence criteria was set to 10−9 . the error is 1.4 Results for 80x80 mesh (c) grid convergence study Figure 8: Dierent studies for Utop = 1 For β = 1.08 13.1485003 -0. At that point.45 is evaluated for dierent mesh sizes. 3. Re=100.45 is presented in Table 4.46 % and thus this mesh size is sucient for grid convergence. A = 4.

Assuming that 160 × 480 has no error. the primary vortex becomes unstable and gives rise to a series of vortices below it. simulations were performed for dierent Reynolds number. the velocity at the point y = 2.1 Grid Convergence Study The grid independence study is shown in Figure 12 for Re=100. Exploratory Case: Eect of height With an increase in the cavity height. (c) Pressure with velocity vectors 4. (b) v-Velocity 4. the curvature is maximum and it can be a test for dierent meshes to capture that 7 Figure 10: Contours of ow variables for Utop = 1 and 80 × 80 mesh . Increasing the Reynolds number causes increase in momentum transfer from the primary vortex to the secondary vortices and thus the latter grows in size. It was observed that increasing the Reynolds number causes the primary vortex to move towards the center of the domain which is well in agreement with the theoretical results of Batchelor [1]. [2] and that obtained from the solver. It can be concluded that the physics is well captured by the code.(a) Results of Ghia et al. With a height to width ratio of 3:1.45 was evaluated for dierent mesh sizes. The results are very similar. At that point.5 Identication of Vortices for dierent Reynolds Number Figure 11 compares the streamlines of Ghia et al.[2] for Re=100 and 1000 (a) u-Velocity (b) Re=100 (c) Re=1000 Figure 11: Comparison of stream-lines with published results 3.

the vortices with subsequent numbering. the ow is turbulent and performing laminar ticity was evaluated.2 Eect of Reynolds Number on Vortex Forma. corners. Figure 14 shows that the study was done to see whether the code was able to at Re=10. Also two more very small vortices were found at the lower corners. From the results it is obvious that there is almost no Figure 14: Eect of Reynolds number on Vortex formadierence between 80 × 240 and 160 × 480 mesh sizes. It is clearly visible that for 80 × 240 mesh. The eect of Reynolds number on analysis was not a true depiction of the ow physics. A case study was performed with ∂x ∂y Re = 5000 and that vortex was identied.Re = 5000.19642846 -0. there exists two big primary vortices and two identify UUE. tion Thus increasing the Reynolds number increases momenVorticity is the tendency of uid to spin and is mathe.18915177 -0. Two secondary vortices were found at the were obtained from Tecplot. The location and strength 8 . vor. 6 vortices were observed.24 % and which makes it sucient for grid convergence.24 - Table 5: Grid convergence study Figure 13: Streamlines at dierent mesh-sizes(Re=100) Figure 12: Grid Convergence Study curvature properly. yet Vortex formation is interesting.number from 10 to 100 caused the primary eddies to decrease in size while the corner eddies merged and growed. Figure 16 shows lower two vortices merged together to produce a big pri.16 4.45 is tabulated for dierent meshes in Table 5. tion Hence the subsequent analysis is performed on 80 × 240 mesh. matically dened as The Upper Upstream Eddy (UUE) was not visible in any of these Reynolds number because it is visible at higher ∂v ∂u Γ= ×V = − (6) values (Re > 3200) [3]. Although at Using a second order central dierence scheme.4.Mesh Size 20 × 60 40 × 120 80 × 240 160 × 480 Value -0.tum transfer from the primary to the secondary eddies. The results mary vortex. It was observed that increasing the Reynolds 4.90 1. The dierent values of u-velocities at y = 2. The Figure 13 shows the streamlines at dierent mesh sizes. Increasing Re to 1000 showed that the For Re=100. the error is 1.16874568 -0. The vortices were found to be very symmetric.3 Location of Vortex Center and its strength gether. secondary ones at the lower corners. Increasing Re to 100 showed the tendency of the two secondary vortices to merge to.19889857 Error(%) 15.

583 × 107 0.599 × 105 −1. The way Tecplot calculates the Vorticity at the eye is probably by interpolating the neighboring cell-center values.59501 0.260 × 10−7 1. T. 1(02):177190. Batchelor. it was needed to probe at a sub-grid scale resolution.01869 0. Journal of Fluid Mechanics Digital Archive.4 Error Band In order to nd the vortex center.01705 0.693 × 102 10−7 4.345 1.680 × 102 −1. December 1982. Ghia.652 × 10−5 10−5 Γrel −0. K.000 Table 6: Location and strength of vortices Vortex No. The location of a vortex is the point where the velocity eld has a local minimum. where Γmax and Γmin are the maximum and minimum vorticities in the neighborhood respectively.00625 in both directions (∆x = ∆y ). In our case. References [1] G. Ghia.53015 0.122 × 10−5 6.40275 0. considering the absolute value.57764 0. N. 1 2 3 4 5 6 |Γmax | |Γmin | 2. K.203 × 10−7 1.388 × 10−5 7. and C. Table 7 shows the error band in Vortex strength estimation.61480 0.Vortex 1 2 3 4 5 6 x 0. i.724 × 10−7 5.98354 y 2. Since the data was only at the cell centers.683 0.23133 0. Figure 16: Location of Vortices at Re=100 [2] U.e Vortex-1 : Vortex-2 is 97.623 × 10−5 0 0 Table 7: Vortex Estimation Error Band of each of these are tabulated in Table 6.23154 0. On steady laminar ow with closed streamlines at large reynolds number.060 × −7. the accuracy of determining the exact location should be within ±∆x in x direction and ±∆y in y direction. Figure 15: Upper Upstream Eddy at Re=5000 4. 9 . the uncertainty in vortex location is ±0.0257 8. Thus the error band in calculating vorticity can be written as Γmax ≥ Γ ≥ Γmin .821 × 10−5 6. 1956.73297 1.516 × 10−2 −7.451 2. The ratio of strength of the two primary vortices.417.020 6. Shin.01921 Γ −2. High-re solutions for incompressible ow using the navier-stokes equations and a multigrid method.116 × 5. 48:387411. Journal of Computational Physics.924 0.

Deshpande. Annual Review of Fluid Mechanics. 2000. N. 32(1):93136. Shankar and M. 10 . D. Fluid mechanics in the driven cavity.[3] P.

- K. Bajer and H.K. Moffatt- Theory of Non-Axisymmetric Burgers Vortex with Arbitrary Reynolds Number
- Stokes Law Presentation
- 1204.1103v7
- Flow Through Packed Lab report for chemical engineering
- 10.1.1.174
- AVL
- Ch. 10 Uniform Flow
- An Experimental Observation of the Mach-And Reynolds Number Independence of Cylinders in Hyper Sonic Flow
- RANS vs LES
- Applied Mathematical Sciences Peter J. Schmid, Dan S. Henningson (Auth.-Stability and Transition in Shear Flows
- tzthfh.txt
- DV-II+PRO-Op
- Fm 1
- EdwardsHartPhotosObjects.pdf
- 104-660-2-PB pk
- HighViscosity_Rewrite1_Sample
- Viscosity and Density (Metric SI Units)
- Otc 22951
- Couette Flow Analytical and Numerical Solution.docx
- Flow Analyses Inside Jet Pumps Used for Oil Wells
- Flow Over Immersed Bodies
- undersaturated oil viscosity correlations.pdf
- 5. Particle Slip Velocity Cuttings Transport_NEW
- Zhou Dissertation
- Pitot Tube Fluids Essay
- AE221_16_Tut_04
- 106BRE
- Supercritical Natural Laminar Flow Airfoil Optimization for Regional Aircraft Wing Design
- Flow in pipes
- Chapter 6 Good

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd