Sunteți pe pagina 1din 14

Turbulent Flows (MA6622) Mixing Layer Numerical Simulation Report Prepared by Low Chee Meng (G1002972A) Introduction Turbulent

flows are distinguished from laminar flows by its chaotic and stochastic nature. They have been described as being diffusive, dissipative, 3-dimensional, and unsteady. Most flows in nature are turbulent and there are many useful engineering applications for turbulent flows. The effects of turbulence can be advantageous or disadvantageous and it is thus important for engineers to understand the causes, effects and behaviour of turbulent flows so as to be able to predict and control it. In the study of turbulence, much progress have been made in both experimental and computational approaches. High-fidelity numerical simulations are able provide much insight into the behaviour of fluid flows and are usually cheaper and quicker to implement than physical experiments. However, the accuracy of the simulation depends on many factors and it crucial for the user to exercise judgement in implementating the simulation and intrepreting the results. The intent of this project was to gain some insight in the generation, evolution and behaviour of turbulence in a mixing layer. A commercial simulation software Fluent was used and the results are presented in this paper. Overview of Simulation and Analysis Parameters The simulation is of a plane mixing layer as shown in Figure 1. The domain is 15m in length with a height of 2m. A splitter plate of 2m length is located at the height of 1m on the left hand side of the domain. The fluid to be used in the computation is air. The simulation involves 2 air-streams entering the domain from the left boundary and exiting on the right boundary. They are initially separated by the splitter plate before joining up where the plate ends at 2m. Figure 2 summarizes the key parameters of the simulation. y

Ul=0.73 m/s Re = 50000 Uh=1.46 m/s Re = 100000

1m Splitter Plate x 1m 2m 15m Figure 1. Schematic of Simulation Domain

The grid was a structured grid consisting of 66000 cells. The turbulence models used in the 1

simulation to close the RANS equations were the k-, Spalart-Allmaras, and Reynolds-Stress models. Each simulation was run for 1000 iterations to minimize the residuals, which is a measure of numerical errors. Fluid Fluid Streams Air with =1.225kg/m3 =1.7894x10-5 kg/ms U1=0.730335 m/s U2= 1.46067 m/s (note: 2D flow with V=0 m/s) 66000 cells Domain Physical Dimensions Splitter Plate 2D, 15m x 2m 2m (length) x 1mm (thickness)

Computational Grid Size

Turbulence Models

k-, Spalart-Allmaras Reynolds-Stress

Figure 2. Key parameters of Simulation In the analysis of the plane mixing layer, it is also useful to define several parameters which will be used to describe the flow. For each turbulence model, data for the mean axial velocity <U>, turbulent kinetic energy k, Reynolds shear stress <uv>, variance of the fluctuating component of the axial velocity <u2> was either directly taken from the data pool generated by Fluent, or otherwise manually calculated. The results are shown in the following sections. Parameter <U> k <u2> <uv> Uh, Ul Characteristic Convection Velocity, Uc Characteristic Velocity Difference, Us y0.9, y0.1 Definition Mean axial velocity Turbulent kinetic energy (k = 1/2<uiui>) Axial-directional normal component of Reynolds-Stresses Reynolds shear stress component Uh and Ul are respectively the higher and lower velocities of the two fluid streams. Uc=1/2(Uh + Ul) Us=Uh - Ul Cross-stream locations at which U(x,y0.1)=Ul(x)+ 0.1Us(x) U(x,y0.9)=Ul(x)+ 0.9Us(x) (x) =1/2(y0.1+y0.9) (x) = 1/2(y0.1(x)+y0.9(x)) =(y-(x))/(x)

Characteristic width of flow, (x) Reference lateral position, Scaled cross-stream coordinate,

Scaled Velocity, f() (<U> - Uc)/(x) Figure 3. Mixing Plane Parameters 2

Reynolds-Stress Model Simulation In the Reynolds-Stress Model (RSM), closure of the Navier-Stokes equations is achieved by solving the transport equations for the Reynolds-Stresses. In contrast with k- and SparlartAllamaras model, the RSM model does not invoke the concept of turbulence viscosity. Figure 4 and 5 respectively tabulates the salient parameters for a plane mixing layer and shows the mean axial velocity <U> at five locations downstream of the splitter plate, where the trailing edge of the splitter plate is x=0. The cross-stream coordinate is given in actual physical dimensions in meters. Hence, the positions y=-1 and y=1 are thus respectively the upper and lower walls of the flow domain. Parameter Ul Uh Us Uc Y0.1 1m 1.46 0.7 0.76 1.08 0.1 Axial Position (x) 2m 4m 5m 1.46 0.69 0.76 1.07 0.17 1.44 0.67 0.77 1.06 0.3 1.44 0.66 0.78 1.05 0.36 6m 1.44 0.65 0.78 1.04 0.39

Y0.9 0.02 0.02 0.03 0.03 0.04 (x) 0.09 0.15 0.27 0.33 0.35 0.06 0.1 0.17 0.2 0.21 Figure 4. Salient Parameters at Various Axial Positions From Figure 5, it can be observed that the velocities are zero at the walls due to the no-slip condition. Away from the walls, the cross-stream variation of <U> can be broadly separated into 3 regions: <U> Uh, <U> Ul , Uh <U> Ul.. The first two regions are simply where the influence Uh and Ul have on each other is relatively low. In the middle region, in which the cross-stream rate of change of <U> becomes significant demarcates the region of mixing between the two initially separated fluid streams. The width of this mixing region is characterised by (x) and from Figure 4, can be observed by noting the portion of the curves in which the slope d<U>/dy is relatively significant.

Figure 5. Axial Variation of <U> Profile


1.600 1.400 1.200 1.000 1m 2m 4m 5m 6m

<U> - (m/s)

0.800 0.600 0.400 0.200 0.000 -1.500 -1.000 -0.500 0.000 0.500 1.000 1.500

Cross-Stream Position, y (m)

The axial variation of (x), from the data shown in Figure 3, is plotted and shown in Figure 6 below. It can be observed that (x) varies approximately linearly with x. In other words, from x1m to x=8m, and actually beyond, d/dx is approximately constant. This corroborates well theoretical analysis from which Equation 1, which introduces and defines the spreading rate S as a non-dimensional parameter, in terms d/dx, Us and Uc. S has been determined to be independent of the flow velocities Us and Uc, just as the spreading rate of a round jet is independent of Re. From the slope of the curve in Figure 6 and the data for U s and Uc, for this present simulation S was found to be 0.08. The value for S was down to be between 0.06 to 0.11 for a range of flows (Dimotakis 1991). S (Uc/Us)(d/dx) [1]

Figure 6. Growth of (x)


0.35 0.3 0.25 0.2 (x) Linear Regression for (x)

f(x) = 0.06x + 0.03

0.15 0.1 0.05 0 3m 4m 6m 7m 8m

Axial Position (m)

Another interesting effect worthy of note is that the mixing layer tends to spread towards the low speed stream. In Figure 7, the axial variations of y 0.9, y0.1 and are shown. They respectively demarcate the upper and lower boundaries and the mid-plane of the mixing layer. It can be observed all three planes vary linearly with x tend towards the upper, lower velocity flow region. From Figure 8, which plots the variation of (y0.9- ) and (y0.1- ) with x, it can be observed that the growth of the mixing layer is symmetric about .

Figure 7. Axial Variation of Y0.9, Y0.1 and Y-bar


0.45 0.4

Cross-Stream Position (m)

0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 3m 4m 6m 7m 8m

Y0.9 Linear Regression for Y0.9 Y-bar Linear Regression for Y-bar Y0.1 Linear Regression for Y0.1

Axial Position (m)

Figure 8. Axial Varation of Y0.9- and Y0.1-


0.2 0.15 0.1 y0.9 0.05 Linear Regression for y0.9 3m 4m 6m 7m 8m y0.1 Linear Regression for y0.1

f(x) = 0.04x + 0.01

Y-

0 -0.05 -0.1 -0.15 -0.2

f(x) = -0.04x - 0.01

Axial Position, x (m)

Figure 9 shows the scaled velocity profile of the mixing layer. In plotting the scaled velocity f() against the scaled cross-stream coordinate , it is observed that the velocity profiles are selfsimilar. It should be noted that the mixing layer will be self-similar only after it is developed, i.e. the mixing layer is not self-similar immediately downstream of the splitter plate where x is small.

Figure 9. Scaled Velocity Profile


0.6 0.4

(<U>-Uc)/Us

0.2 0 -1.5 -1 -0.5 -0.2 -0.4 -0.6 0 0.5 1 1.5

1m 2m 4m 5m 6m

Figure 10 shows the profiles of the turbulent kinetic energy, k, at various axial positions. A comparison with Figure 5 suggests that the peak of each curve is where the mean velocity gradients are highest in the mixing layer. The d<U>/dy curve has an inflection point which leads to instability and turbulence. The Kelvin-Helmholtz instability is common feature in shear layer flows in which velocity gradients are high such as the one in this simulation. Near the walls, the k also increases along with the velocity gradients. Figures 11 and 12 show the profiles for the Reynold's stresses <u 2> and <uv>, and upon comparion with Figure 5, it can be inferred that the Reynold's stresses are associated with high velocity gradients. Indeed, Equation 3 shows the assumed relationship between the ReynoldsStresses and velocity gradients. <uiuj> = 2/3kij T (d<Ui>/dxj + d<Uj>/dxi) [3]

Since the components of k include <u2>, the close similarity between Figure 10 and 11 should be expected. Upon closer inspection of the relative magnitudes of k and <u 2>, it can be also inferred that <u2> is a large component of k for this present simulation. Other observations are that 5

the axial variation of k and <u 2> is observed to be very small compared to the streamwise variations, and also that both are higher at higher flow velocities or Reynolds numbers.

Figure 10. Axial Variation of Turbulent Kinetic Energy


Turbulent Kinetic Energy (m2/s2)
0.050 0.045 0.040 0.035 0.030 0.025 0.020 0.015 0.010 0.005 0.000 -1.500 -1.000 -0.500 0.000 0.500 1.000 1.500 1m 2m 4m 5m 6m

Crosstream Position (m)

Figure 11. Axial Variation of <u2> Profiles


0.045 0.040 0.035

<u2> - (m2/s2)

0.030 0.025 0.020 0.015 0.010 0.005 0.000 -1.500 -1.000 -0.500 0.000 0.500 1.000 1.500

1m 2m 4m 5m 6m

y (m)

Figure 12. Axial Variation of <uv> Profiles


0.020000 0.015000 0.010000

1m 2m 4m 5m 6m 0 0.5 1 1.5

<uv>

0.005000 0.000000 -1.5 -1 -0.5 -0.005000 -0.010000

Crosstream Position (m)

Figure 12 shows the profiles of the Reynolds shear stress <uv>. It is observed that <uv> is high where the velocity gradients are high, similar with <u 2>. It can also be seen that at the upper wall (y=1) <uv> is positive whereas at the lower wall (y=-1) it is negative. This can be explained by noting the sign of Equation 3 below. Since turbulent viscosity T is positive, it can be seen that <uv> is negatively related to d<U>/dy. According to the coordinate system shown in Figure 1, it can be shown that at the lower wall (y=-1), d<U>/dy is positive, and thus <uv> is negative, whereas the opposite holds for upper wall where d<U>/dy is negative. <uv> = - T(d<U>/dy) [3]

Figure 13 and 14 respectively show <u2>/Us2 and <uv>/Us2 plotted against the scaled crossstream coordinate . It can be observed that the profile shapes of the reynolds stresses are generally rather similar. Thus is can be said that Reynolds-Stresses also exhibit self-similarity when scaled with the velocity difference Us and the local layer thickness (x). It is further observed that the scaled <u2> profiles at 1m and 2m seem to be marginally separated from the profiles at 4m,5m, and 6m. For the scaled <uv> profiles, the profile at 1m seems to be separated from the remaining profiles. This seems to indicate that the mixing layer is still not fully developed below 2m in terms of the of Reynolds-Stresses.

Figure 13. Scaled <u2> profiles


0.080 0.070 0.060

<u2>/(Us)^2

0.050 0.040 0.030 0.020 0.010 0.000 -1.500 -1.000 -0.500 0.000 0.500 1.000 1.500

1m 2m 4m 5m 6m

Figure 14. Scaled <uv> Profiles


0.030000 0.025000 0.020000

<uv>/(Us)^2

1m 0.015000 0.010000 0.005000 0.000000 -1.5 -1 -0.5 0 -0.005000 0.5 1 1.5 2m 4m 5m 6m

k- Model Simulation The k- model is a two equation model, in which model transport equations for k and , the dissipation rate are solved in order to determine T, which will be used to model the ReynoldsStresses. Equation 4 shows the relationship between T, k and . The k- model is stable and robust and less demanding on computational resources compared to RSM models, albeit with a compromise in accuracy. It is also not suitable for flows with boundary layer separation, or where there are sudden changes in the mean strain rate. However, for simple shear flows such the present simulation, the model is said to be capable and reliable. T = Ck2/ where C = 0.09 Figure 15 shows the axial variation of the <U> profile. It is largely similar to that generated by RSM. Figure 16 shows the scaled velocity profile of <U> and is shows the self-similarity of the mixing layer. Comparing Figures 15 and 16 to Figures 5 and 9, it is observed that the results obtained with the k- model do not show a kink in the vicinity of y=0 =0.6. Therefore the results for the k- model seem to more accurately reflect other experimental and computational results compared with the RSM model. [4]

Figure 15. Axial Variation of <U>


1.600 1.400 1.200 1m 2m 4m 5m 6m

<U> - (m/s)

1.000 0.800 0.600 0.400 0.200 0.000 -1.500 -1.000 -0.500 0.000 0.500 1.000 1.500

Cross-steam Position, y (m)

Figure 16. Scaled Velocity Profiles of <U>


0.6 0.4 0.2 1m 2m 4m 0 -0.2 -0.4 -0.6 0.5 1 1.5 5m 6m

f()

0 -1.5 -1 -0.5

Figures 17, 18, and 19 respectively show the axial variation of the mixing layer thickness (x), the direction of its movement, and the development of the boundaries of the mixing layer about its mid-plane. From the Figure 17, d/dx is observed to be 0.05, which give a spreading rate value Sk- = 0.07 compared with SRSM=0.08. However, the trends are similar, with Figure 18 showing the tendency of the mixing layer to move towards the lower speed stream, and its symmetric growth about the midplane .

Figure 17. Axial Variation of (x)


0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 3m 4m 5m 6m 7m 8m

f(x) = 0.05x + 0.06


(x) Linear Regression f or (x)

(x) - (m)

Axial Position (m)

Figure 18. Axial Variation Y0.1, Y0.9 and Y-bar


0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 3m 4m 5m 6m 7m 8m Y0.9 Linear Regression f or Y0.9 Ybar Linear Regression f or Ybar Y0.1 Linear Regression f or Y0.1

Y(m)

Axial Position (m)

Figure 19. Axial Variation of (y0.9-) and (y0.1-)


0.25 0.2 0.15 0.1 0.05 0 -0.05 3m -0.1 -0.15 -0.2 -0.25 4m 5m 6m 7m 8m Y0.9- Linear Regression for Y0.9- Y0.91- Linear Regression for Y0.91-

Figures 20, 21, and 22 show the variations of the profiles of k, <u 2>, and <uv> in the axial direction. The date for turbulent kinetic energy k was directly extracted from Fluent, while <u 2> 9

was calculated from k via the relationship <u2>=<v2>=<w2>=2/3k. <uv> was calcuated by extracting the data for T and du/dy using the relationship between them shown in Equation [3]. From Figure 20, it can be observed that the profile for k is generated by the k- model differs rather significantly from that generated by the RSM model. Firstly, the magnitudes of k calculated by the k- model is one order of magnitude larger than that calculated by RSM. The area under the curve is also fuller which suggests that a broader region of the flow has higher k. The data also shows that k increases with increasing downstream distance, which is contrary to the data shown in Figure 10. The

Figure 20. Axial Variation of k


0.200 0.180 0.160 0.140 0.120 0.100 0.080 0.060 0.040 0.020 0.000 -0.800 -0.600 -0.400 -0.200 0.000 0.200 0.400 0.600 0.800 1m 2m 4m 5m 6m

k (m2/s2)

Cross-stream Position (m)

Figure 20. Axial Variation <u2> Figure 21. Axial Variation ofof k
0.200 0.140 0.180 0.120 0.160 0.100 0.140 0.120 0.080 0.100 0.060 0.080 0.060 0.040 0.040 0.020 0.020 0.000 0.000 -0.800 -0.600 -0.400 -0.200 0.000 -0.800 -0.600 -0.400 -0.200 0.000 0.200 0.200 0.400 0.400 0.600 0.600 0.800 0.800

k - (m2/s2) <u2>(m2/s2)

1m 1m 2m 2m 4m 4m 5m 5m 6m 6m

Cross-stream Position (m) Cross-Stream Position (m)

Figure 22. <uv> at Various Axial Positions


0.120 0.100 0.080

-<uv> - (m2/s2)

0.060 0.040 0.020 0.000 -1.500 -1.000 -0.500 -0.020 0.000 -0.040 -0.060 0.500 1.000 1.500

3m 4m 6m 7m 8m

Cross-stream Position (m)

10

Figure 23 and 24 show the scaled profiles of <u 2> and <uv>. It can be observed that while the general shapes of the curves seem to resemble each other, the separation between them are significant. In Figure 23, it can be observed that the <u2> profiles at 1m and 2m seem to form a pair, and the profiles for 5m and 6m form another pair, while the the 3m profile is clearly separated from both pairs. In Figure 24, the profiles for 1m and 2m form a pair, while the profiles for 4m, 5m, and 6m form another group. Hence, for the Reynolds-Stresses, it is observed that while the shapes of the scaled curves seem to be similar, the placement of the curves are somewhat irregular. This seems to contradict the the concept of self-similarity. Unlike for the scaled velocity profile, the k- does not seem to generate data that will provide for self-similar <k>, <u2>, and <uv> profiles.

Figure 23. Scaled <u2>


0.200 0.180 0.160 0.140

<u2>/<Us>^2

0.120 0.100 0.080 0.060 0.040 0.020 0.000 -1.500 -1.000 -0.500 0.000 0.500 1.000 1.500

1m 2m 4m 5m 6m

Figure 24. Scaled <uv> Profiles


0.200

0.150

0.100

<uv>/<u2>^2

1m 2m 4m 5m 6m 0.500 1.000 1.500

0.050

0.000 -1.500 -1.000 -0.500 0.000 -0.050

-0.100

Spalart-Allamaras Model Simulation The Spalart-Allamaras model is a one equation model which solves the transport equation for the turbulent viscosity to provide closure for the RANS equations. The model is designed for 11

use with unstructured grids and is particularly suited for attached wall bounded flows but weak for free shear flows. The velocity profiles and scaled velocity profiles are shown in Figure 25 and 26 respectively. That the velocity profiles display self-similarity is apparent from the Figure 25.

Figure 25. Variation of <U> with Axial Position


1.600 1.400 1.200 1.000 1m 2m 4m 5m 6m

<U> (m/s)

0.800 0.600 0.400 0.200 0.000 -1.500 -1.000 -0.500 0.000 0.500 1.000 1.500

y (m)

Figure 26. Scaled <U>


0.6

0.4

(<U>-Uc)/Us

0.2

1m 2m 4m 0 0.5 1 1.5 5m 6m

0 -1.5 -1 -0.5 -0.2

-0.4

-0.6

Figure 27 and 28 respectively show the profile of <uv> in x-y coordinates and the scaled <uv>. Similar to the date provided by the RSM and k- models, the magnitude of <uv> is largest where the velocity gradients are the largest near the wall and in the mixing layer. The scaled <uv> curves show strong self-similarity between the curves for the 4m, 5m and 6m axial positions, with the curves at 1m and 2m somewhat separate from the others. The k profiles are shown in Figure 29.

12

Figure 27. Axial Variation of <uv> Profiles


0.020

0.015

<uv> - (m2/s2)

0.010

1m 2m 4m 5m 6m

0.005

0.000 -1.500 -1.000 -0.500 0.000 -0.005 0.500 1.000 1.500

y (m)

Figure 28. Scaled <uv> Profiles


0.03000 0.02500

<uv>/<Us>^2

0.02000 0.01500 0.01000 0.00500 0.00000 -1.5 -1 -0.5 0 0.5 1 1.5

1m 2m 4m 5m 6m

Figure 29 and 30 respectively show the rate of growth of the mixing layer and trajectory of the mixing layer downstream. Both sets of data agree with the data provided by RSM and k- models. It can observed that (x) grows linearly and that the mixing layer tends to move towards the lower velocity region downstream.

Figure 29. Growth of (x)


0.35 0.3 0.25

f(x) = 0.05x + 0.06

(x) - (m)

0.2 0.15 0.1 0.05 0 3m 4m 6m 7m 8m

(x) Linear Regression for (x)

Axial Position (m)

13

Figure 30. Axial Variation of y0.9, y0.1 and


0.35

Cross-Stream Position (m)

0.3 0.25 0.2 0.15 0.1 0.05 0 3m 4m 6m 7m 8m

Y0.9 Linear Regression for Y0.9 Ybar Linear Regression for Ybar Y0.1 Linear Regression for Y0.1

Axial Position (m)

Conclusion In the present simulations, a commercial software was used to analyze the development of a plane mixing layer. Three different turbulence models were used in the computations and the results were compared. It was clear that each turbulence model provided slightly different results. In plotting the curves for k, <u2> and <uv>, it was clear that there were significant difference in the data sets most directly related to turbulence. The difference was smallest when comparing the velocity profiles and scaled velocity profiles. The scaled <U> profiles for all three models indicated self-similarity. Also, all 3 models showed that the mixing layer tend to move towards the lower velocity region of the flow, which agrees well with other simulations and experimental evidence. All 3 models also showed that the Reynolds-Stresses and turbulent kinetic energy are high where the velocity gradients are high. The choice of turbulence model would then depend on their comparative computational expense, as well as their proven suitability for the simulation problem. Two of the models invoke the turbulent viscosity hypothesis. The k- model has proven to be robust, stable, economical and it provides relatively good predictions for many flows. It is however weak in simulating flows with high swirl and rotation, flows with high pressure gradients and separated flows. The SpalartAllamaras model is relatively cheap and simple to implement and is suited for flows which are attached to mild mild separated. It is weak for free shear flows and for predicting turbulence decay. The Reynolds Stress Model does not invoke the turbulent viscosity hypothesis and solves the transport equations for the Reynolds-Stresses directly. It is computationally expensive but also most suited for complex flows and flows with separation. Hence, the user has to exercise judgement in the selection of turbulence models.

14

S-ar putea să vă placă și