You are on page 1of 44

MM4CFD Assignment 1 Malcolm Chan Hao Xian UNIMKL-003434

ASSIGNMENT 1a Comparison of CFD Simulation to Experimental Data for Backward Facing Step Test
Malcolm Chan Hao Xian Model Domain The domain used in computation is a two-dimensional setup with a stream wise length 30 meter with the entry length of 10 meter before the fluid flow encounters the step of 1 meter height. The total vertical height on the other hand is 6 meter. The model can be view in Figure 1, where h=1meter. The benchmark for this case study is taken from [1] and [2], utilizing the experimental results from Jovic and Drivers NASA experiment. The domain setup is extracted from Totas computational setup.

Figure 1: Model Domain of Backward Facing Step Viscous Model Many models can actually be utilized to perform computation. But often, to achieve the most accurate computation, a suitable viscous model must be selected where the selection made must be based upon wall y+ and meshing configuration of the Backward Facing Step model. From [5], it can be concluded that different turbulent models are often used based on the near wall treatment approach. Among all the viscous models, five models can be chosen; Spalart-Allmaras, Reynolds Stress, K-Epsilon, K-Omega, as well as Large Eddy Simulation Model. For example, if a model has high near-wall mesh resolution, the wall y+ value will be lower and located within the viscous sub layer region, therefore is a perfect case for K-Omega and Spalart-Allmaras models; which is integrated to the wall itself. The perfect opposite can be commented on K-Epsilon and RSM as well. These two models are more dominant in turbulent core flow and their wall y+ values are higher than 30, located in the log law region. Therefore, to perform computation, the centroid of adjacent-to-wall mesh must be located within the log law region, eliminating the need for near wall meshing. However, a wall function must be attached to RSM and K-Epsilon to bridge the model to viscosity affected region on the wall. The two models chosen for computation in this assignment are the Spalart-Allmaras and K-Omega-SST Model. The reason behind using K-Omega SST model is because of elimination of wall functions; hence the model is able to integrate to the wall as well. Furthermore, the SST formulation combines the K-Epsilon solution model in free stream region thus enable the model to reduce sensitivity towards turbulent properties of free stream inlet flow. This enables K-Omega to produce a high quality simulation of flow separation. Spalart-Allmaras, on the other hand is being chosen for comparison of the best model to use within the case of backward facing step, in a case of solution within the viscous sub layer. Another apparent reason for these two selections is due to its low

Reynolds Number at 5100. From [7], it can be noted that K-Onega SST and Spalart-Allmaras is suitable for low Reynolds Number problem. Meshing Mesh Economy The mesh in Gambit is modeled where the step model is divided into three main faces. The reason behind such modeling is to facilitate the formation of meshes.

Figure 2: Model Split Into 3 Separate Domain From [5], it can be concluded that the meshing performed must pertain to the viscous model used to simulate the turbulent fluid flow in the test model and at the same time should consist of low skewness and sufficient definition as well in order to obtain a good computation result. However, the computational cost required must be considered as well. Therefore, the definition of mesh is not only prioritized according to the viscous model but also to the purpose of computation. In this case of backward facing step which is a wall bounded flow case study, the purpose is to study the characteristic of separation of fluid flow at the sudden step descend, as well as monitoring the behavior of fluid at different locations on the wall. Therefore, it can be said that it is more important to define the step and wall zones of the model with finer mesh, while other locations can be applied with bigger mesh size. This will allow the model to free up an amount of computational cost and focus them on areas that are more important. Another purpose of the zone division is to allow aspect ratio meshing on the model and prevent skewed mesh elements. As mentioned before, the step region and wall zone are the most important, and the importance of fine meshing will decrease for locations further away from zone 3. For example, in zone 1, the size of mesh will be finer as it progresses along into the step region. Also, the mesh is finer on both side of walls in zone 1, but the mesh will grow bigger as it progresses along the centerline of the channel. In zone 2 as well, the mesh is configured to grow coarser as it leaves the step region. These arrangements will help release unnecessary meshing in unimportant areas. On the other hand, there must also be a balance between mesh economy and mesh quality. It is acceptable to compute meshes that are considerably bigger in size at area that are not important. However, there must be a limit to the size as not to affect the quality of mesh.

Figure 3: Example of Aspect Ratio Meshing Wall Treatment For the K-Omega and Spalart-Allmaras case, the adjacent-to-wall mesh has to be fine enough to be in the viscous sub layer region. The meshing for this model is done accordingly to facilitate K-Omega computation, keeping a fine layer of mesh on the wall, as well as freeing up unnecessary meshing. To obtain a suitable viscous sub layer meshing, a first round computation must first be done in order to find out the initial wall y+. The first round of computation uses an amount of 9460 mesh elements producing a peak wall y+ value of 28. Therefore the recommended following step is to refine the mesh. The mesh is now being refined to 11850 mesh elements. The wall y+ peak value recorded is at 15. This shows a general improvement of wall y+ value. However, the mesh can still be refined further to produce a better wall y+ value. For cycle 3 of meshing, the model is meshed to an amount of 15800. For this model, the wall y+ value is at 7. Note that the mesh before the step needs to be refined but for the wall after the step, the mesh is fine enough to produce wall y+ of 4. The model is then further refined to 20400 mesh elements, this time producing a wall y+ of 3.5. This signifies that the adjacent wall mesh element is within the viscous sub layer range and is optimum for K-Omega SST computation. Uniform Meshing Another variation of mesh model is computed to compare the results to models that were applied with wall treatment. Two additional mesh configurations with uniform grid size were used for computation and the results (wall y+, Ux / Uo) are directly compared to the experimental results to validate the usage of wall treatment. Mesh economy Grid Size 0.0625 0.03125 Total No. of Cells 64000 256000 Cells For Step 16 32

Figure 4: Wall Mesh for 9460 Elements

Figure 5: Wall Mesh for 11850 Elements

Figure 6: Wall Mesh for 15800 Elements

Figure 7: Wall Mesh for 20400 Elements

Figure 8: Wall y+ Value for 9460 Elements

Figure 9: Wall y+ Value for 11850 Elements

Figure 10: Wall y+ Value for 18500 Elements

Figure 10: Wall y+ Value for 20400 Elements

Computational Methodology The case simulated for the K-Omega SST and Spalart-Allmaras model is mainly divided into three cases. Case 1 Wall y+ model of less than 5, Fine and Medium Uniform Mesh model, and untreated wall model computed for Re=5100 with K-Omega SST model and compared to experimental data from Jovic and Driver. Case 2 Wall y+ model of less than 5 computed for Re = 5,100, Re=5,500, Re=7,000, Re=10,000 with K-Omega SST model. The five variations of Reynolds Number in Case 2 are to compare and analyze the effect of Reynolds Number on the reattachment length, turbulent kinetic energy, static pressure and velocity magnitude after the step down. Another round of simulation is done for Spalart-Allmaras model. As explained in before, both K-Omega SST and Spalart-Allmaras solves for the same viscous sub layer region. Therefore, a comparison can be made between these two model and the experimental data, to find the suitability of the models to solve for a wall bounded passage problem. Case 3 Wall y+ model of less than 5 computed for Re = 5,100 using Spalart-Allmaras and compared to experimental data and K-Omega SST model Parameters Fluid Material Water 5100 5500 7000 10000 (0.0051153 m/s) (0.0055165 m/s) (0.0070210 m/s) (0.0100300 m/s)

Reynolds Number

Using information from the model domain in which the step height (h=1) and the kinematic viscosity of water ( =0.001003m2/s), the velocity magnitude in x-direction can be obtained. = (1)

in which is the height of the step. For all cases, excluding Velocity Magnitude and Viscous Model, all other parameters are kept constant. The turbulent kinetic energy for the channel inlet is changed to 0.0001m2/s2 Results For comparisons between different meshings and viscous model, the properties used are wall y+, normalized x-velocity, where x-velocity is divided by maximum free stream velocity (Ux/Uo), Turbulent Kinetic Energy (TKE), and Static Pressure. The comparisons of wall y+ value and Ux/Uo are shown in Figure 11 and 12. From the figure, it can be seen that the uniform medium meshing and fine wall meshing

has a wall y+ value that is significantly higher than the value of 5, at a peak of 22.5 and 12.2 respectively. From Figure 13, at x/h=4, all three meshing model shows a similar trend across the height of the step and the channel from y=0 up until y=1, at which every meshing model produces different trend. It can be concluded that between these meshing models, the wall treated model has a closer similar trend to the experimental data. As the trend line stretches to the middle section of the channel, the normalized velocity observed illustrates that the wall treated mesh has computed a much closer approximation of the experimental data, while both the medium mesh and fine mesh model has a less accurate approximate of normalized x-velocity. Note the backflow of fluid from y=0 to y=0.5 is caused by the streams of separation bubble on the step itself which can be observed in Figure 16 and 17. For x/h=6, the trend line for all three model is significantly different from the experimental data. The possible reason behind this is due to the large wall y+ value for both the fine and medium uniform mesh, which will produce inaccurate computations. However, this should not affect the treated wall model. Another possible flaw is such that the benchmark experimental data utilizes a function of turbulent mixing length which is a limitation in this study due to the unknown of the defined function. Therefore, this limitation may produce a generally larger Ux/Uo magnitude than the experimental data, since [1] implies the use of turbulent mixing length to produce better approximation. The Ux/Uo magnitude recorded at x/h=10 shows the same outcome x/h=6 producing a poor computation. The same reason should apply as well for case x/h=6. All in all, from Figure, it can be summed up that the wall treated mesh model possesses a better overall approximation of the experimental data and can be used as a basis for meshing strategy in this exercise. For the Spalart-Allmaras model, the comparison for Ux/Uo is as shown in Figure. From results of computation for x/h=4, 6, 10, it is very apparent that although both Spalart-Allmaras and K-Omega SST produced the basic trend similar to the experimental data, K-Omega SST has the better approximation. For example, in Figure, it can be seen that the computation of Spalart-Allmaras does not mimic the backflow of fluid in the step region from x=0 to x=0.5. Furthermore, for all three location (x=4, 6, 10), KOmega SST has a closer approximate of Ux/Uo from y=2 to y=3. Conclusively, K-Omega SST should be prioritized over Spalart-Allmaras in solving for wall bounded flows when solving in viscous sub layer region.

Figure 11: Wall y+ Value for Medium Mesh

Figure 12: Wall y+ Value for Fine Mesh

3.5 3 2.5 2 Experimental Data 1.5 1 0.5 0 -0.4 -0.2 0 0.2 0.4 Ux/Uo 0.6 0.8 1 1.2 Wall Treated Model Fine Mesh Model Medium Mesh Model

Figure 13: Comparison of Different Meshing Model of Ux/Uo using K-Omega SST Model For Position x=4
3.5 3 2.5 2 Wall Treated Model 1.5 1 0.5 0 0 0.2 0.4 Ux/Uo 0.6 0.8 1 Medium Mesh Model Fine Mesh Model Experimental Data

Figure 14: Comparison of Different Meshing Model of Ux/Uo using K-Omega SST Model For Position x=6

Distance Across Channel Passage (m)

Distance Across Channel Passage (m)

3.5 3 2.5 2 Wall Treated Model 1.5 1 0.5 0 0 0.2 0.4 Ux/Uo 0.6 0.8 1 Fine Mesh Model Medium Mesh Model Experimental Data

Figure 15: Comparison of Different Meshing Model of Ux/Uo using K-Omega SST Model For Position x=10

Distance Across Channel Passage (m)

Figure 16: Contours of Velocity Magnitude for Wall Treated Mesh with K-Omega SST Model

Figure 17: Contours of Stream Function for Wall Treated Mesh with K-Omega SST Model
3.5

3 Distance Across Channel Passage (m)

2.5

2 Exp Data 1.5 K-Omega Spalart-Allmaras 1

0.5

0 -0.4 -0.2 0 0.2 0.4 Ux/Uo 0.6 0.8 1 1.2

Figure 18: Comparison of Different Meshing Model of Ux/Uo For Position x=4

3.5 3 2.5 2 Exp Data 1.5 1 0.5 0 0 0.2 0.4 Ux/Uo 0.6 0.8 1 K-Omega Model Splarat-Allmaras

Distance Across Channel Passage (m)

Figure 19: Comparison of Different Meshing Model of Ux/Uo For Position x=6
3.5 3 Distance Across Channel Passage (m) 2.5 2 Exp Data 1.5 1 0.5 0 0 0.2 0.4 Ux/Uo 0.6 0.8 1 Spalart-Allmaras K-Omega

Figure 20: Comparison of Different Meshing Model of Ux/Uo For Position x=10

The comparison of Velocity Magnitude, Turbulent Kinetic Energy, and Static Pressure for different Reynolds Numbers for K-Omega SST are shown in figures in the following page. It can be seen from Figure 21, 22, 23 that the Velocity Magnitude plot recorded an increasing magnitude with the increase of Reynolds Number for all three locations in the channel(x=4, 6, 10). For x=4, it can also be observed that the backflow in the step region decreases when Reynolds Number is increased. This simply describes the flow as resistance to retardation from the wall due to the increase in inertial forces. The Turbulent Kinetic Energy plot for x=4, 6, 10, on the other hand recorded increase in Turbulent Kinetic Energy on the wall bounded region but remained constant and minimal in the channel centroid. Note that as the fluid flows past into the channel, the Turbulent Kinetic Energy for each Reynolds Number respectively decreases. This is due to the characteristics of fluid boundary layer. Logically, at x=10, the fluid should have a lower energy magnitude as compared to x=4 because the fluid is developing a turbulent flow and mixing in the step region. As the fluid passed by the step and recover the calm flow, experiencing a decrease in turbulence. The increase in Turbulent Kinetic Energy due to Reynolds Number can logically be explained with (1). Once the Reynolds Number is increased, the velocity will be increased as well, rendering a high inertial force. This will in turn create a higher turbulent kinetic energy. From Figure 27, the static pressure of the bottom wall surface after the step presents a very similar trend for all Reynolds Number starting with a drop in pressure from x=0 to x=4, at which the pressure experience an increase again until it stabilizes to a constant pressure. With every increase of Reynolds Number, it can be observed that the trend will experience an increase in its magnitude. The reattachment length after the separation bubble in the step region observed for all four different Reynolds number does not present any significant change. Referring to [1], Tota V. suggests that the increase of turbulent mixing length will render the reattachment length of the separation bubble smaller. Again due to the limitations of the unknowns in the benchmark exercise, the behavior of reattachment length is not observed in this exercise. However, it can be concluded that the increase of Reynolds Number does not alter the reattachment length. Another explanation for this is that the increase of Reynolds Number is not significant enough, and that the total increase in fluid velocity is still minimal. Limitations and Difficulties One main difficulty in the assignment is converging the computation in the software, Fluent. The main problem that happened is the indication of reversed flow and the limitation of turbulent viscosity ratio of the fluid flow. The reason behind this is the turbulent kinetic energy, which at default is 1, is too high for the input of velocity. Therefore, to acquire a converging solution, the turbulent kinetic energy has to be scaled down, in this case to a mere 0.00001m2/s2. Reversed flow in Fluent should be made clear, doesnt affect the final result of computation. However, it does affect the converging process, rendering it more time consuming to compute. Therefore, to perform a good computation, reversed flow should be eliminated. Another limitation of the exercise is the benchmark work used. Due to the fact that Fluent is just a computing software, to input a certain unique characteristic of fluid, functions are normally written and inserted into the software. In this case, it is the turbulent mixing length in the benchmark exervise, which the user defined function is unknown. Therefore the investigation of reattachment length due to turbulent mixing length is unachievable.

Figure 21: Velocity Magnitude Plot for x=4 for Different Reynolds Number

Figure 22: Velocity Magnitude Plot for x=6 for Different Reynolds Number

Figure 23: Velocity Magnitude Plot for x=6 for Different Reynolds Number

Figure 24: Turbulent Kinetic Energy Plot for x=4 for Different Reynolds Number

Figure 25: Turbulent Kinetic Energy Plot for x=6 for Different Reynolds Number

Figure 26: Turbulent Kinetic Energy Plot for x=10 for Different Reynolds Number

Figure 27: Static Pressure Plot for Wall Surface 10m to 30m for Different Reynolds Number

Figure 28: Contours of Stream Function for Reynolds Number 5500 with K-Omega SST Model using Wall Treated Mesh

Figure 28: Contours of Stream Function for Reynolds Number 7000 with K-Omega SST Model using Wall Treated Mesh

Figure 28: Contours of Stream Function for Reynolds Number 10000 with K-Omega SST Model using Wall Treated Mesh

Conclusion 1. Wall Treated Mesh is an optimum strategy to compute fluid flow for low Reynolds Number in wall bounded flows. 2. In a wall bounded channel passage flow, K-Omega SST presents a more accurate computation as compared to Spalart-Allmaras. 3. With increasing Reynolds Number, Turbulent Kinetic Energy, Static Pressure and Velocity Magnitude experience increment. 4. Reattachment Length characteristics is not affected by Reynolds Number in a case of Backward Facing Step.

ASSIGNMENT 1b Comparison of CFD Simulation to Experimental Data for T-Junction and Backward Facing Step Test
Malcolm Chan Hao Xian Model Domain The domain used in computation is a two-dimensional setup as expressed in Figure 29, where h=1meter. Referred domain is combined of two channel passage of 3m and 6m respectively, each with a 0.3m width. The benchmark for this case study is taken from [3] and [4], utilizing the experimental results from Shettar and Murthys experiment. The domain setup is extracted from Shamloos computational setup.

Figure 1: Model Domain of T-Junction Viscous Model The two models chosen for computation in this assignment is the Reynolds Stress Model and K-Epsilon RNG. The reason behind using Reynolds Stress Model is because of the number of equations in the model as well as the increased accuracy. This is because the eddy viscosity approach has been taken out of the model and the Reynolds Stress is being directly computed. This altogether enables a high quality simulation of flow separation, which in this case is the purpose of T-Junction test. K-Epsilon RNG, on the other hand is being chosen due to its simplicity and fast converging attributes. Another apparent reason for these two selections is due to its high Reynolds Number at approximately 255000. Also, from [7], it can be noted that K-Epsilon RNG and Reynolds Stress Model is suitable for swirling characteristics in fluid flow.

Meshing Mesh Economy The mesh in Gambit is modeled where the T Junction is divided into four main faces as can be seen below. The reason behind such modeling is to facilitate the formation of meshes.

Figure 2: Model Split Into 4 Separate Domain From the findings in the first part of exercise, only wall treated mesh will be used for computation purposes. In this case of T-Junction which is a wall bounded flow case study, the purpose is to study the characteristic of separation of fluid flow along the main channel and branch channel, as well as monitoring the stagnation area of the T-Junction corner.. Therefore, it can be said that it is more important to define the step and wall zones of the model with finer mesh, while other locations can be applied with bigger mesh size. This will allow the model to free up an amount of computational cost and focus them on areas that are more important.

Figure 3: Expected Fluid Behavior in T-Junction Another purpose of the zone division is to allow aspect ratio meshing on the model and prevent skewed mesh elements. As mentioned before, the junction and wall zone are the most important, and the importance of fine meshing will decrease for locations further away from these areas. For example, in zone 1 and 3, the size of mesh will be finer as it progresses along into the junction. Also, the mesh is finer on both side of walls in zone 1 and 3, but the mesh will grow bigger as it progresses along the centerline of the channel. The same applies for zone 4 as well. These arrangements will help release unnecessary

meshing in unimportant areas. On the other hand, there must also be a balance between mesh economy and mesh quality. It is acceptable to compute meshes that are considerably bigger in size at area that are not important. However, there must be a limit to the size as not to affect the quality of mesh. Wall Treatment For the Reynolds Stress and K-Epsilon case, the adjacent-to-wall mesh does not need to be fine to ensure an improved computation. As mentioned, these two models solve in the log law region and these models are coupled with wall functions, hence replacing the need of a fine adjacent to wall mesh. The meshing for this model is done sufficiently to provide a wall y+ within 30 to 60, as well as freeing up unnecessary meshing to ensure that the adjacent wall mesh is within log law region. To obtain a suitable log law meshing, a first round computation must first be done in order to find out the initial wall y+. The first round of computation uses an amount of 22420 mesh elements producing a peak wall y+ value of 100. Therefore the recommended following step is to refine the mesh. The mesh is now being refined to 23530 mesh elements. The wall y+ peak value recorded is at 85. This shows a general improvement of wall y+ value. However, the mesh can still be refined further to produce a better wall y+ value. The third meshing results in an amount of 25750 mesh elements. For this model, the wall y+ value is at 65. The model is then further refined to 26490 mesh elements, this time producing a wall y+ of 60. This signifies that the adjacent wall mesh element is within the log law range and is optimum for any log law region formula; Reynolds Stress Model and K-Epsilon RNG. Computational Methodology The case simulated for the Reynolds Stress Model and K-Epsilon RNG model is mainly divided into two cases. Case 1 Wall y+ model of less than 60, computed for approximately Re=255,000 with Reynolds Stress Model and K-Epsilon RNG compared to experimental data from Shettar and Murthy. Case 2 Wall y+ model of less than 60 computed for Re = 200,000 and 100000 with K-Epsilon RNG The two variations of Reynolds Number in Case 2 is to compare and analyze the effect of Reynolds Number on the reattachment length, turbulent kinetic energy, static pressure and velocity magnitude in the channels and branch. Parameters Fluid Material Reynolds Number Water 255,000 (0.850 m/s) 200,000 (0.670 m/s) 100,000 (0.340 m/s)

Using information from the model domain in which the channel width (h=0.3) and the kinematic viscosity of water ( =0.001003m2/s), the velocity magnitude in x-direction can be obtained. = (1)

in which is the width of junction. For all cases, excluding Velocity Magnitude and Viscous Model, all other parameters are kept constant. The turbulent kinetic energy for the channel inlet is changed to 0.001m2/s2. The discretization of pressure, momentum, turbulent kinetic energy, and turbulent dissipation rate are all changed to second order solution. Discretization remains similar except for turbulent dissipation rate which was changed to the original first order solution. This is done in order to reduce computation effort and produce a simulation that can converge in a shorter amount of time.

Figure 4: Wall Mesh for 22420 Elements

Figure 5: Wall Mesh for 23530 Elements

Figure 6: Wall Mesh for 25750 Elements

Figure 7: Wall Mesh for 26490 Elements

Figure 8: Wall y+ Value for 22420 Elements

Figure 9: Wall y+ Value for 23530 Elements

Figure 10: Wall y+ Value for 25750 Elements

Figure 11: Wall y+ Value for 26490 Elements Results

For comparisons between the two viscous models, the properties used are wall y+, X-Velocity, Turbulent Kinetic Energy (TKE), and Static Pressure. These values are plotted against the location on the passage normalized by the passage width (y/h). It can be seen from Figure 12 to 21 below, the recorded x direction velocity for the viscous model K-Epsilon RNG and Reynolds Stress Model has a close resemblance of the experimental fluid characteristics by Shettar and Murthy. The viscous model of KEpsilon RNG and Reynolds Stress Model should be considered as well for wall treated meshing in a wall bounded fluid flow. However, it should be noted that there are certain discrepancies especially in Figure 17, 19, and 21 where the peak value of X-Velocity is higher than both Shamloo and Shettar & Murthys result. These discrepancies happen especially after fluid flow past the branch region. This explained by Shamloo, was due to the three dimensional affects as the domain model computed in the case study is a simple two dimensional model. Also, the plot of X-Velocity for both Reynolds Stress Model and KEpsilon RNG compare well together, with both having similar trends. The most significant difference is in the magnitude of peak X-Velocity for cases x=3.15m and 3.45m. The predicted contour of static pressure of K-Epsilon RNG is similar to those predicted by Shamloo. It can also be observed that the static pressure for the K-Epsilon RNG model recorded a lower static pressure as compared to Shamloos prediction. This is because the computation for K-Epsilon RNG adopts a low Turbulent Kinetic Energy of 0.001 m2/s2. The contours of Velocity Magnitude computed matches well with Shamloos contour, abiding to the fluid characteristics when passes through the branch channel where there is a velocity shift away from the main channel passage into the wall region of the branch. It should be noted that there is one significant difference between Shamloos predictions of velocity magnitude contour to the K-Epsilon RNG. That is the separation zone developed after the fluid leaves the branch opening. From Ramamurthy, along the continuous far wall, separation due to flow expansion may occur. The prediction by Shamloo did not present a significant build up this said separation zone, which in contrast, K-Epsilon RNG managed to present in the contour. This may be due to the wall treatment of log law region. The contours of static pressure and velocity magnitude for Reynolds Number bear a very close similarity to both Shamloos case and K-Epsilon RNG. The static pressure magnitude also resembles those recorded by K-Epsilon RNG. The velocity magnitude contour on the other, mimics well of the velocity shift into the branch channel. Moreover, the Reynolds Stress Model also presents the characteristics of separation zone on the main channel flow after crossing the branch From these comparisons, it can be seen that both K-Epsilon RNG and Reynolds Stress Model are capable of solving t he computation of a fluid flow in wall bounded region with the almost the same amount of accuracy. However, the computation effort for Reynolds Stress Model is almost thrice as much as the K-Epsilon RNG due to the thriving number of equations (5 no.) that has to be solved directly. Therefore, it can be concluded that for a case of fluid flow through a wall bounded region, a K-Epsilon model would provide a sufficient computational accuracy coupled with wall treatment meshing for the log law region. Hence, the computation for Case 2 will be completed using K-Epsilon RNS model.

1 0.9 0.8 0.7 0.6 y/b 0.5 0.4 0.3 0.2 0.1 0 0 0.2 0.4 0.6 0.8 1 K-Epsilon RNS RSM

X-Velocity (m/s2)

Figure12 & 13: Comparison of Computed Results to Shamloo and Shettar&Murthys Results for x=1.35m

1 0.9 0.8 0.7 0.6 y/h 0.5 0.4 0.3 0.2 0.1 0 0 0.2 0.4 0.6 0.8 1 K-Epsilon RNS RSM

X- Velocity (m/s2)

Figure14 & 15: Comparison of Computed Results to Shamloo and Shettar&Murthys Results for x=2.85m

1 0.9 0.8 0.7 0.6 y/h 0.5 0.4 0.3 0.2 0.1 0 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 K-Epsilon RNS RSM

X-Velocity (m/s2)

Figure16 & 17: Comparison of Computed Results to Shamloo and Shettar&Murthys Results for x=3.15m

1 0.8 0.6 y/h K-Epsilon RNS 0.4 0.2 0 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 RSM

X-Velocity (m/s2)

Figure18 & 19: Comparison of Computed Results to Shamloo and Shettar&Murthys Results for x=3.45m

0.8

0.6 y/h

0.4

K-Epsilon RNS

RSM

0.2

0 0 0.1 0.2 0.3 X Velocity (m/s2) 0.4 0.5 0.6

Figure20 & 21: Comparison of Computed Results to Shamloo and Shettar&Murthys Results for x=5.1m

Figure 22 & 23: Comparison of Static Pressure Contours between between Computation Results of Shamloos Predicted, K-Epsilon RNG, and Reynolds Stress Model Contour (Clockwise)

Figure 24 & 25: Comparison of Velocity Magnitude Contours between Computation Results of Shamloos Predicted, K-Epsilon RNG, and Reynolds Stress Model Contour (Clockwise)

The comparison of Velocity Magnitude, Turbulent Kinetic Energy, and Static Pressure for different Reynolds Numbers for K-Epsilon RNG are shown in figures in the following page. It can be seen from Figure 26 that the Turbulent Kinetic Energy plot for x=1.35 recorded a decrease in its Turbulent Kinetic Energy on the wall bounded region when the Reynolds Number is decreased. The same can be observed from both Figure 27, and 28, where the trend of Turbulent Kinetic Energy across the surface at x=3.15 and 5.1 remains identical but each with a different magnitude. It can be observed that the magnitude will decrease with the Reynolds Number. The increase in Turbulent Kinetic Energy due to Reynolds Number can logically be explained with (1). Once the Reynolds Number is increased, the velocity will be increased as well, rendering a high inertial force. This will in turn create a higher turbulent kinetic energy. From Figure 29, the static pressure of the bottom wall surface after the branching presents a very similar trend for all Reynolds Number starting with a drop in pressure from x=3 to x=3.5, at which the pressure experience an increase again until it stabilizes to a constant pressure. With every decrease of Reynolds Number, it can be observed that the trend will experience a decrease in its magnitude. From Figure 30 to 32, the Velocity Magnitude plot recorded a decreasing magnitude with the decrease of Reynolds Number for all three locations in the channel(x=1.35, 3.15, 5.1). It can be observed that the general trend of these plots remained the same only varying in its magnitude. The reattachment length after the separation bubble in the branch channel observed for all three different Reynolds number presents with a significant change. Referring to [3], Shamloo suggest that the increase of discharge ratio will render the reattachment length of the separation bubble smaller. However, due to the nature of the model domain which is strictly a two dimensional model, discharge ratio cannot be applied in this case. Again due to the limitations of the unknowns in the benchmark exercise, the behavior of reattachment length due to discharge ratio is not observed in this exercise. However, it can be concluded that the decrease of Reynolds Number, on the other hand, will increase the reattachment length of the separation zone in the branch channel. Limitations and Difficulties There are a few different limitations and difficulties in this specific exercise. One problem is that the boundary condition is given in terms of Velocity, and not Reynolds Number. Therefore, it is impossible to find out the default kinematic fluid viscosity that is being used by the author of the case study. This simply leaves a gap in the study as to the unknown of whether the computation conducted is referring to the same Reynolds Number as the benchmark study. Another limitation of the exercise is the domain model used. Due to the fact that the model used is of two dimensional, the specification of discharge ratio is not applicable. Therefore the investigation of reattachment length due to discharge ratio is unachievable. Another repetitive difficulty in the assignment is converging the computation, as well as the indication of reversed flow. The reason behind this is the turbulent kinetic energy, which at default is 1, is too high for the input of velocity. Therefore, to acquire a converging solution, the turbulent kinetic energy has to be scaled down, in this case to 0.001m2/s2. The reason why the turbulent kinetic energy in this case study is less than in Backward Facing Step is due to the increase in fluid velocity.

1.2

0.8

y/h

0.6

Re=255000 Re=200000

0.4

Re=100000

0.2

0 0 0.0005 0.001 0.0015 0.002 0.0025 0.003 0.0035 0.004 0.0045 0.005

Turbulent Kinetic Energy (m2/s2)

Figure 26: Turbulent Kinetic Energy Plot for x=1.35 for Different Reynolds Number
1.2

0.8 y/h

0.6

Re=100000 Re=200000

0.4

Re=255000

0.2

0 0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009

Turbulent Kinetic Energy (m2/s2)

Figure 27: Turbulent Kinetic Energy Plot for x=3.15 for Different Reynolds Number

1.2

0.8

y/h

0.6

Re=100000 Re=255000

0.4

Re=200000

0.2

0 0 0.001 0.002 0.003 0.004 0.005 0.006 0.007

Turbulent Kinetic Energy (m2/s2)

Figure 28: Turbulent Kinetic Energy Plot for x=5.1 for Different Reynolds Number

Figure 29: Static Pressure Plot for Wall Surface 3m to 6m for Different Reynolds Number

1.2

0.8

y/h

0.6

Re=255000 Re=100000

0.4

Re200000

0.2

0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Velocity Magnitude (m/s)

Figure 30: Velocity Magnitude Plot for x=1.35 for Different Reynolds Number
1.2

0.8

y/h

0.6

Re=100000 Re=255000

0.4

Re=200000

0.2

0 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Velocity Magnitude (m/s)

Figure 31: Velocity Magnitude Plot for x=3.15 for Different Reynolds Number

1.2

0.8 y/h

0.6

Re=200000 Re=100000

0.4

Re=255000

0.2

0 0 0.1 0.2 0.3 0.4 0.5 0.6

Velocity Magnitude (m/s)

Figure 32: Velocity Magnitude Plot for x=5.1 for Different Reynolds Number

Figure 33: Contours of Stream Function for Reynolds Number 200000 with K-Epsilon RNG Model using Wall Treated Mesh

Figure 34: Contours of Stream Function for Reynolds Number 255000 with K-Epsilon RNG Model using Wall Treated Mesh

Figure 35: Contours of Stream Function for Reynolds Number 100000 with K-Epsilon RNG Model using Wall Treated Mesh

Conclusion 1. Wall Treated Mesh is an optimum strategy to compute fluid flow for High Reynolds Number in wall bounded flows 2. In a wall bounded channel passage flow, K-Epsilon RNG presents a more suitable computation as compared to Reynolds Stress Model, producing the same accuracy of computation with less effort and time. 3. With decreasing Reynolds Number, Turbulent Kinetic Energy, Static Pressure and Velocity Magnitude experience decrement. 4. Reattachment Length characteristics increases as Reynolds Number decreases, for a junction or branching flow.

References [1] Tota, P. V. (2009, April 21). Turbulent Flow Over a Backward-Facing Step Using the RNG k- Model. [2] Jovic, S., & Driver, D. M. (1994, February). Backward-Facing Step Measurements at Low Reynolds Number. Moffett Field, California, USA. [3] Shamloo, H., & Pirzadeh, B. (2008, May). Investigation of Characteristics of Separation Zones in. Tehran, Iran. [4] Shettar, A. S., & Murthy, K. K. (1996, June 4). A numerical study of division of flow in open channels. Bangalore, India. [5] Salim, S. M., & Cheah, S. (2008, December 3). Wall y+ Strategy for Dealing with Wall-bounded Turbulent Flows. Semenyih, Selangor, Malaysia. [6] Fluent 6.2 Documentation File. (2010). ANSYS Manual . Fluent. [7] Chilka, A., & Kulkarni, A. (n.d.). Modeling Turbulent Flows in Fluent. Retrieved November 1, 2010, from CFD Flow Modeling Software & Solutions from Fluent: http://www.fluent.com/software/university/blog/turbulent.pdf

You might also like