You are on page 1of 9

Application of CFD in Stack Interference

Prapanj Ganeshan

, Suresh Kumar , Jon Galsworthy

October 1, 2009

Tall stacks (or chimneys) are typically located in open terrain conditions. Considering that these are simple geometric shaped structures likely in open environment (sometimes called textbook type structures), nding loads on these structures is not cumbersome and is possible using analytical calculations. However, when these structures are in groups, they are sometimes subjected to severe wind loads depending on the distance between them in relation to their size and angle of attack. When stacks are in a group, the downstream stacks are subjected to multiple eects caused by upstream stacks. Some of these phenomenon are vortex shedding and wake bueting. Computational Fluid Dynamics (CFD) has been critically challenged so far in applications to such problems involving blu bodies and high Reynolds numbers. With steady improvements in methods of handling turbulent ow simulations, the application of CFD for simulation of interference eect on stacks is in the foreseeable future. An attempt has been made in the present paper to validate CFD simulation method applied to stack interference problem using Large Eddy Simulation (LES) for resolving/modeling turbulence. The results are compared with available wind tunnel results.

Abstract

Introduction

Computational Fluid Dynamics (CFD) is the study of uid ow by numerical simulation and is a eld advancing in leaps and bounds. These numerical simulations are used either for fundamental research or for prediction of ow parameters in practical problems. The former requires highly accurate data and the latter requires production of prototype or optimization of a design with minimal cost, in which case the desired predictions are mean values or extremes. The constraints on the quality of representation of ow, diers for both these cases. However, it goes without saying that the quality of prediction improves with richness of the algorithm or physical model used in the simulation. With this in mind, availability of better methods and models have to be explored and applied progressively for practical problems. It becomes imperative then, to check the viability of any new method for practical application. In the eld of CFD, computing power and time of simulation are two major constraints. In spite of the exponential growth in computing

1 RWDI 2 RWDI

email: ganeshan.prapanj@rwdi.com, phone: +91 471 4060010


India Pvt. Ltd., T-5, Thejaswini, Technopark, Trivandrum-695581 Inc., Guelph, Ontario, Canada.

power, most of us are skeptical about sustenance of this growth curve and hence we look for tweaks and methods to make the best out of available technology. Hence, successful application of CFD boils down to attaining a good compromise between richness of physical model, maximum utilization of computational resource, computing cost and accuracy of solution. LES claims a generalized approach-resolving the important turbulent scales and modeling the universal scales. Application of a generalized method for a practical problem is important and hence, its viability with available resources has to be studied. Amplication of across-wind forces on cylinders due to the presence of an upstream cylinder is a classical problem. One real world scenario where this poses an engineering challenge is when stacks are to be constructed in groups in an open terrain. The observed behaviour in these cases is an increase in across-wind forces due to the presence of the upstream chimneys. It is to be noted that these are problems in the very high Reynolds number regime of 107 . Subsequent to a eld study, an experimental program was conducted at Monash University, Australia [1]. The Reynolds number achieved using large subsonic wind tunnel was 106 compared to full-scale Re of 107 . Solution of stack interference issues demand a generalized ow solution methodology as arriving at an analytical solution that can handle dierent stack congurations is impractical. Modeling of turbulence cannot be considered a generalized approach. Hence possibility of simulating interference eect in stacks using a method that will resolve the eects is important. LES is the most viable way of doing this. The present paper is a report on one such experiment that keeps practical applicability in scope, in terms of computing power and time taken. Throughout the experiment, a maximum computing power of a 8 core mini-cluster was used. The grid resolution was limited to keep the time of computation within practical limits. Initially a single cylinder case was studied and compared with the wind tunnel experiment mentioned earlier. The same grid resolution was then used to two cylinders in tandem. The results are compared with the same experimental results.

Summary of the Wind Tunnel Experiment

Initially, as a baseline comparison for the two cylinder congurations, a set of wind tunnel runs for a single cylinder was conducted [1], for which the Reynolds number range was 2.2 105 to 106 . The lift and drag coecients obtained after integrating forces were corrected for blockage eects using empirical results of Modi and El-Sherbiny [2]. Two circular cylinders were used spanning the width of the working section of the wind tunnel, with the downstream cylinder xed and the upstream cylinder capable of moving with a forklift truck [1]. The longitudinal spacing ranged from 1500 mm to 3000 mm. Of this, only the longitudinal spacing of 1500 mm is being compared in the CFD experiment. Three dierent upstream cylinder diameters were studied, of which one case where both the cylinders are 500 mm in diameter is being used for comparison. The upstream cylinder was instrumented with pressure taps at the stagnation point and at angles 90 , 180 and 270 from the stagnation point. For the downstream cylinder, a ring of 32 taps at equal angles were installed on the centerline of the wind tunnel. The sampling frequency used was 200 Hz for 120

seconds. It is noted in the thesis that the upstream cylinder comes as close as two diameters to the oor of the wind tunnel. This would have some impact on the ow pattern around the upstream cylinder and concomitantly on the orientation of the wake behind it.

CFD and LES

CFD has been in application since the early 1980s in industry. CFD has evolved into a much more reliable method to predict uid ows with advancements in various modeling and solving techniques, not to mention the advancements in High Performance Computing facilites. Turbulence in uid ows has been a topic of great interest for research in lieu of the fact that successful simulation of uid ow is directly connected to successful simulation of turbulence. The wide range of turbulence scales is an enigmatic feature of any turbulent ow problem. A number of methods and schemes have been proposed in the past, some of which have found profound application in a variety of ow problems. Solving Reynolds Averaged Navier-Stokes (RANS) equations has been the industry standard for a couple of decades now, which involves obtaining an ensemble average of the various uctuating variables in the domain. Such an averaging results in introduction of new unknowns in the form of the Reynolds stress tensor, , which represents the eects of all turbulent uctuations. The Reynolds stress tensor has to be modeled to close the system. These models usually use an empirical result, based on experiments. Some of these also use intuitive approaches. In essence, RANS closure equations decide how the turbulent eddies of dierent length scales transfer energy among each other. A number of such models, generally called turbulence models came into practice, most of which are being used widely in industrial ow simulations.

Figure 1: Decomposition of energy spectrum in solution associated with RANS Unfortunately, the RANS approach suers from a principal shortcoming in that, it is virtually impossible to model the eects of large scales of turbulence (while the small scales are dependent on viscosity and are somewhat universal) in the same way in ows that are very dierent. LES is a result of an attempt to generalize the turbulent ow regimes. Resolving the complete turbulence scale is impossible with current computing power for a problem of considerable complexity. But it is possible to resolve the major universal scales of turbulence and model the smaller scales, by using a lter to demarcate between the scales. This dierence between RANS and LES can be noticed from gures 1 and 2. Figure 3 shows the use of a simple scale

Figure 2: Decomposition of energy spectrum in solution associated with LES separation operator. The presence of large and small eddies in scope can be seen in the physical space representation. Though LES does not resolve the full scales of turbulence, it captures a much larger range of scales than the RANS approach.

Figure 3: A simple scale separation operator: The sharp cut o wave number kc in the Fourier space is a direct function of the grid size in physical space. Since LES attempts at resolving the non-universal scales, the problem size tends to increase tremendously for high Reynolds numbers. This is owing to the fact that it is required by the computational grid to be as rened as to resolve the tubulent boundary layer. Hence, computational cost is a major factor in deciding whether LES is practical for a given ow problem.

Solution Procedure and Analysis


1. Single cylinder with Re 2.2 105

The following cases were studied and results presented.

Table 1: Y + values for the CFD experiments Re 2.2 10 Re 4.5 105 Re 6.5 105

Case

y+

Downstream Cylinder

min 0.286 0.472 0.656 5.05

y+

max 29.67 55.55 76.62 478.6

y+

6.103 11.603 16.10 131.2

mean

2. Single cylinder with Re 4.5 105 3. Single cylinder with Re 6.5 105 4. Two cylinders in tandem with Re 1 106

Figure 4: The Geometry of the CFD domain An open source C++ library called OpenFOAM (version 1.5.x) was used for the simulations. The geometry of the domain can be seen in Figure 4. Since LES resolves the major turbulence scales, a suciently realistic correlated turbulent uctuation has to be provided at the inlet. This was achieved by using a turbulent-inlet boundary condition available with OpenFOAM. An inlet turbulent intensity of 5% was used, as the wind tunnel experiment also uses 5% turbulence. LES requires that the turbulent boundary layer be resolved for good accuracy. Rening the computational grid down to the viscous sublayer is dicult for such high Reynolds numbers. Hence, a wall function was applied to subgrid scale turbulent viscosity, that comes with OpenFOAM. This was used as a compromise for not meeting renement standards of a pure LES simulation. The non-dimensionlized grid spacing values (y + ) used are tabulated in Table 1.The grid resolution can be seen in Figure 5.

Figure 5: Resolution of Grids used (a) single cylinder domain (b) close up of wall Table 2: Coecients of Drag and Lift for Single Cylinder

Re

CD
5

2.2 10 4.5 105 6.5 105

- Experimental 0.26 0.35 0.38

CD -

LES 0.7 0.398 0.355

CL -

Experimental 0.3 0.1 0.12

CL

0.150 0.082 0.082

- LES

For the single cylinder cases, the ow simulations ranged over a time of 3 seconds. For the cylinders in tandem, the simulation was for uid ow of 1 second. Simulation of 1 second of ow for the 2 cylinder case took 72 hours on a 8 core processor with 16 GB of RAM.

5
5.1

Results
Single Cylinder

A number of simulations had to be performed to make sure the grid is suciently rened to conduct the experiment. Hence, the baseline single cylinder results of the experimental work were used for comparison. Grid with the same renement level was used for all the cases. The coecient of drag values are tabulated in Table 2 and can be seen in Figure 6. It may be seen that, the wind tunnel experimental results have a leftward shift compared to the results of Farell (the classical drag bucket curve). This was claimed in [1] as due to the roughness of cylinder and the inlet turbulence. The present results match with the subcritical and supercritical parts of the curve, though the drag predicted near the critical region is deviant. It is to be noted that a time constraint was placed on these simulations and with increase in simulation time, the averaged values could change.

Figure 6: Coecient of drag comparison [1, 7]

8 Experimental LES 7

Reduced Spectral Density fS(f)/ ~CL 2

0 0 0.1 0.2 0.3 Reduced Frequency fD/V 0.4 0.5 0.6

Figure 7: Spectrum of downstream cylinder

Table 3: CD and St for downstream cylinder

Experimental LES

St 0.41 0.23 0.47 0.20


CD

5.2

Two Cylinders

After matching reasonably with experiments, the same grid resolution was used to construct a domain with two cylinders in tandem as shown in Figure 4. The drag value and the Strouhal number of the downstream cylinder can be seen in Table 3.

Figure 8: Pressure contours around the cylinder for all cases that were studied The reduced spectral density of coecient of lift uctuations of the downstream cylinder can be seen in Figure 7. Pressure contours showing the shedded vortices for all the cases can be seen in Figure 8.
5.3 Summary and Remarks

Though performing simulations of this scale are quite demanding in terms of time and computational resources, it is clear from the results that use of LES for complex ow problems like stack interference is in the foreseeable future. This paper is just an interim result of ongoing experiments with applicability of

LES for wind engineering problems at Rowan Williams Davies & Irwin (RWDI) India Ltd. The experiment will be conducted for longer time ranges, and with dierent intereference congurations.

References
[1] Galsworthy, Jon Kenneth, Ph. D Thesis: Aspects of Across-Wind Loads And Eects On Large Reinforced Concrete Chimneys, University of Western Ontario. 1972[2] Modi, VJ and El-Sherbiny, S. Eect of Wall Connement on Aerodynamics of Stationary Circular Cylinders, Proceedings of the Third International Conference on Wind Eects on Buildings and Structures, Tokyo, September 1971, pp. 365-375. [3] Meng Wang, Pietro Catalano, Gianluca Iaccarino Prediction of high

Reynolds number ow over a circular cylinder using LES with wall modeling, Center for Turbulence Research Annual Research Briefs, 2001 pp. 45-50.

[4] Forum discussions at http://www.cfd-online.com/Forums/openfoam/ [5] Pierre Sagaut, Large Eddy Simulation for Incompressible Flows, Springer 1998. [6] Eugene de Villiers, Ph. D Thesis: The Potential of Large Eddy Simulation for the Modeling of Wall Bounded Flows, Imperial College of Science, Technology and Medicine. 2006 [7] Farell, C. Flow Around Fixed Circular Cylinders: Fluctuating Loads, Proceedings of ASCE, EM3 pp. 16330, 565-588 [8] Shoei-Sheng Chen Flow-Induced Vibration of Circular Cylindrical Structures, Springer-Verlag 1987

You might also like