Design of ITA’s Research Wind Tunnel Contraction Using CFD Tools
Bento Mattos Embraer – São José dos Campos – Brazil
Nide G. C. R. Fico Jr. and Roberto M. Girardi
Instituto Tecnológico de Aeronáutica – São José dos Campos – Brazil -12228-900
Abstract speed along the centerline is larger than the
speed close to the wall near the contraction The present work describes the conceptual entrance and (iii) at its exit the flow velocity design of the contraction for the new near the wall, but outside the boundary subsonic research wind tunnel being layer, is larger than the velocity at the center constructed at the Technological Institute of line (Tulapurkara and Bhalla, 1988). Thus, Aeronautics (ITA), located in São José dos to provide the desired high quality flow at Campos, Brazil. The contraction was the test section, the contraction should have conceived to yield an excellent flow quality a geometry that minimizes the adverse at the test section. The method described by pressure gradients at its extremities, in order Morel was used to design a set of to avoid boundary layer separation. Besides, reasonable initial geometries, which were it is necessary to have in mind that at the labeled and organized in some groups. The contraction exit plane the boundary layer fully unstructured CFD code FLUENT was thickness should be small and the velocity then employed to analyze the flow inside of profile uniform. some selected wind-tunnel configurations. The CFD results clearly revealed the most There exist in literature countless suited configuration regarding the required methods for designing contractions. Klein et characteristics at the test section. The study al (1973), Chmielewski (1974) and Morel is part of a project of technological (1975) supply good bibliographical revision innovation being supported by FAPESP in on the subject. Most of the proposed which ITA and Embraer are involved. The methods solve the equations considering the final objective of this endeavor is to create a flow incompressible and potential. The flow methodology for extrapolation of solution supplies, among other information, experimental 2-D results to three- the streamlines. The segment of a dimensional configurations. streamline that results in a monotonic velocity distribution, with the desired Introduction. contraction ratio, is chosen to mold the contraction contour. Thwaites (1946) and Contraction is a most important element of Bossel (1969) make a different approach. a wind tunnel. As the fluid passes through They admit that the entrance and the exit the contractions, the flow accelerates and, planes of the contraction are equipotent consequently, an uniformization of velocity planes. Through a technique of separation profile and a decrease in the turbulence of variables, they obtained an expression, in level occurs. Real contractions have finite series form, for the velocity potential. From length and, as a consequence, it is observed that result, the streamlines are calculated. that: (i) an adverse pressure gradient exists The series coefficients are such that a near both ends of the contraction, (ii) the Copyright 2003 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved. monotonic velocity distribution along the contraction wall is guaranteed. The contraction of ITA’s Research Wind Tunnel (Fig. 1-3) was designed through the method proposed by Morel (1975). The project begins with the choice of the contraction’s geometry, that is, its length and contraction ratio. Then a numeric solution for the flow inside the contraction and ducts of constant cross section at each one of the contraction’s extremities are obtained. The contraction is considered of finite length, resulting in adverse pressure Fig. 2 – Inlet nozzle. gradients and non-uniformity of the velocity profile close to both ends. The Stratford Criterion (1959), to prevent boundary layer separation, is used to avoid the occurrence of this highly undesirable phenomenon, especially in the contraction of a wind tunnel of which an excellent flow quality at the test section is required. Tulapurkara and Bhalla (1988) made an experimental investigation of two nozzles; one had a contraction ratio of 3.464 while for the other this parameter was equal 12. Both were designed through Morel’s method. They Fig. 3 – test section. observed that: (i) in both cases there was no The design and construction of a flow separation (ii) the nonuniformity in the wind tunnel with low turbulence level at the velocity profile, at the contraction exit was test section (TI = 0,05%) is part of the 1% instead of the 2% expected (iii) the Project of Technological Innovation velocity profile is practically uniform at a financed by the Fundação de Amparo à distance equal to 0.1 D1 where D1 is the Pesquisa do Estado de São Paulo diameter at the contraction inlet section. (FAPESP), in which both ITA and Embraer Thus, it can be affirmed that the easy to use are involved. The objective is to develop Morel’s method is capable of supplying a methods for extrapolation of experimental good contraction. results obtained in 2-D bodies for 3-D configurations. ITA’s Research Tunnel, a fundamental tool in such an effort, will have a test section of 1.0x1.2 m. The maximum Reynolds number is 106, based on a model whose characteristic length is 0.30 cm. The last generation instrumentation includes equipment such as Particle Image Velocimetry (PIV) and Laser Doppler Velocimetry (LDV). Both, with capacity for 3-D measurements, will be of paramount Fig. 1 – Cutaway of the wind-tunnel importance in the present and future installation. experimental investigations undertaken at the Division of Aeronautical Engineering of ITA. Brief Explanation of Morel’s Method
The contraction’s contour is obtained
through cubic curves that join at a point denoted by xm. Results for the potential flow inside nozzles with contraction ratio varying from 2 up to 25 are supplied by Morel (1975). The parameters L/D1 and X= xm /L (see Fig 1) were varied from 0.75 up to 1.27 Fig. 4 – Sketch of a contraction made up of two cubic curves. and from 0.12 up to 0.8, respectively. Once the pressure distribution along the wall was Description of the Computational calculated, it was possible to get the Method pressure coefficients at the inlet and outlet of the contraction, Cpi and Cpe, The commercial fully unstructured code respectively. Thus, FLUENT was used in the simulations considered in the present work. FLUENT is where Vi and Ve are, respectively, the a well-known CFD package, developed in the context of unstructured grids that uses a 2 2 é V ù é U 2 ,¥ ù finite-volume formulation and is capable of C pi = 1 - ê1 - i ú C pe = 1 - ê1 - ú handling both internal and external flows êë U i ,¥ úû ë Ve û around complex geometries. The equations, minimum and the maximum velocity close after being discretized, can be to the entrance and to the exit of the simultaneously (coupled method) or contraction, while U1,¥ and U2,¥ are the sequentially (segregated method) solved. In undisturbed velocities upstream and the present work, the segregated method downstream of the nozzle. Graphics of Cpi was adopted. A point-implicit Gauss Seidel and Cpe as function of L/D1 and X= xm /L scheme in conjunction with a method of for several contractions were obtained by algebraic multigrid (AMG) is used to solve Morel (1975). With the aid of such results, the scalar system of equations for each L/D1 and X= xm /L are obtained for the dependent variable at a cell. wanted contraction ratio and for the chosen Turbulent flows are characterized by values of Cpi and Cpe. Once the values of fluctuations. Considering that among these L/D1 and X, the inflection point, are known fluctuations there are very small scales and the geometry of the wall contour is given very high frequencies, it becomes unfeasible by: to directly simulate such flows in practical D-D 3 engineering calculations. Thus, the 2 = 1 - 1 æç x ö÷ for x £ X instantaneous governing equations can be D -D X 2 èLø L 1 2 averaged in time or, in any other suitable D-D 1 æ xö 3 x way, to eliminate such small scales and high 2 = 1- - ³X 2 ç ÷ for 1 frequencies. The resulting system of D -D 1 2 (1 - X ) è L ø L equations, the so-called Reynolds Averaged Navier-Stokes Equations (RANS), requires where D is the diameter at a distance x from far less computational resorts to be solved the contraction inlet and D2 it is the (Table I). RANS equations require closure diameter at its exit. for Reynods stresses in order to develop turbulence models. Turbulence modeling is one of the pillars of CFD; the other two are mesh generation and development of algorithms. Unlike the other two, progress in turbulence modeling has been slow and the models are not, by any means, universal. variables close to the wall are smaller than This should not come as a surprise due to those of the k-e model. Consequently, the the great complexity of the physical Sparlat and Allmaras model is less sensitive phenomenon (Wilcox, 1998). In this to near-wall grid quality. context, FLUENT has several turbulence models, leaving the user the best choice for The Reynolds stress model is the the problem at hand. In this work, three most sophisticated one available in the turbulence models were used: Sparlat- FLUENT arsenal of turbulence models. Allmaras, standard k-e, nd a Reynolds Abandoning the hypothesis of isotropic Stress Model (RSM). For two-equation turbulence, RSM addresses the resolution of models, turbulence viscosity is correlated the Reynolds Averaged Navier-Stokes with turbulent kinetic energy (TKE) and the equations by solving transport equations for dissipation rate of TKE. The k-ε turbulence the Reynolds stresses, together with an model is the most widely used in the equation for the dissipation rate. This means industry. Therefore, its strength and that four additional equations are requested weaknesses are well documented. The k- for 2-D simulations flows while seven are equation is derived by subtracting the needed for 3-D floes. The complexity of the instantaneous mechanical energy from its RSM gives it a larger potentiality to cope time-averaged value, and the ε-equation is with complex flow fields than models based formed from physical reasoning. This model upon the Boussinesq hypothesis. However, is only valid in fully turbulent flows. In the accuracy of the RSM predictions is still order to overcome the intrinsic drawbacks limited by the assumptions related to several of the Standard k-ε model, some derivatives terms in the transport equations. The were developed such as the Realizable k-ε modeling of terms such as turbulence and the RNG k-ε. dissipation rate and pressure are particularly challenging and often considered responsible for degrading the accuracy of numerical simulations using RSM. Table II compares the weaknesses and strengths of some turbulence models10.
Table I– Solving the Navier-Stokes equations.
The one-equation model of Sparlat
and Allmaras solves a transport equation for the turbulent kinetic viscosity. This model Table II – Strong and weak points of some was especially developed for aeronautical turbulence models. applications involving internal flows and it has been supplying good results for Most k-ε and RSM turbulence boundary layers subjected to adverse models will not predict correct near-wall pressure gradients. In FLUENT, the original behavior if integrated down to the wall. model of Sparlat and Allmaras was Thus, special near-treatment is required. modified to use wall functions whenever the Concerning the near-wall treatment, Table mesh is not refined enough for low III presents the usual techniques and Reynolds computations. An important point provides a good comparison between is that the gradients of the transported them10. which in turn, softens the adverse pressure gradient. The pressure loss in the screens placed at the stilling chamber is proportional to the inverse of the square of the contraction ratio. High contraction ratios are, therefore, desirable. On the other hand, the larger the contraction, the more expensive is its construction with increased risk of boundary-layer separation. Thus, a Table III - Comparison of near wall compromise for small wind tunnels is a treatments. contraction ratio between 6 and 9 (Metha and Bradshaw, 1979). ITA’s wind tunnel Results was designed with a contraction ratio equal In terms of fluid dynamics the accurate to 10 in an effort to get low turbulence level modeling the internal wind-tunnel flow at the test section. requires the adequate computation of both flow field and boundary-layer development. Morel (1975), based on Stratford’s This section presents results in order to raise approach (1959), suggests maximum values important issues concerning the application. of Cpi and Cpe to keep the boundary layer The choice of FLUENT was the CFD tools attached. Contractions with small L/D1 and is based on previous successful experience X= xm /L (see Fig. 4) yield Cpi>0.5 and, in by de Mattos et all [11– 13] in the such a case, boundary-layer separation close application of FLUENT to usual practical to the nozzle entrance, is a real danger. problems, typical of the aerospace industry, Design criteria related with the nozzle exit in both aerodynamics and thermal analysis. flow conditions indicate, according to Morel (1975), values of Cpe£0.06. Table IV ITA’s Research Wind Tunnel has been shows the main parameters for the designed aiming outstanding flow geometries studied. characteristics to be achieved in the test section. One of the indicators of such flow quality is the turbulence level attained by the overall configuration. In the present case, the goal is to keep turbulence level Table IV: Parameter of the analyzed under 0.05% in the test section. An contractions. adequate design of the contraction section is a way to obtain low turbulence levels. The flow for all configurations Efficient contractions are characterized by a presented in Table IV was numerically uniform and steady velocity profile at the simulated. It is important to point out that exit cross section. An additional important the calculation domain covers the consideration is a small boundary-layer contraction and the whole length of the test thickness at the nozzle outlet with the section in order to estimate the boundary smallest length. A contraction that satisfies layer growth along it. these requirements will certainly present boundary-layer profiles that are close to the All simulations were performed separation point. Further, the non- using the Fluent’s segregated method. uniformity of the exit-velocity profile is Typical computational meshes were made equal to the tolerated maximum, typically up of 1.6 million hexahedral elements (Figs. ±1/2% outside the boundary layer. It is 5, 6). An important detail of the grid always possible to avoid boundary-layer construction is the rounding of the corners separation by increasing the nozzle length, of both the contraction and the test section. This rounding is convenient for the Fig. 7 compares the boundary-layer mounting of “corner fillets” used to make thickness at the end of the test section that the fine adjustment in the compensation for were obtained by calculations employing the boundary layer growth during tunnel three different turbulence models. All calibration phase. All computational meshes simulations were performed with 52.2 m/s were generated using the pre-processor code maximum speed at the test section. It can be Gambit, which is also developed and easily observed that there is a significant marketed by FLUENT Inc., New difference between the results obtained with Hampshire. The baseline meshes possessed the standard k-e model and the other two about 1,600,000 hexahedra but after grid closure schemes. The good agreement of the adaptation to ensure enough points inside Sparlat-Allmaras model with the RSM the viscous sub layer, this number raised to approach gave the authors confidence in the about two million. results obtained by the former.
Fig. 7 – Boundary-layer thickness at the end
of the test section obtained with several turbulence models. Configuration #1.
Fig. 8 exhibits the static pressure
contours along the tunnel walls for configuration #1. A very smooth pressure Figure 5 - Computational mesh for configuration #1. Notice the round corners. distribution is observed along the contraction and the test section. Configuration #1 is, in fact among all others, the one presenting the smoother static pressure variation along the tunnel’s centerline. Thus, although it is the longest one, configuration #1 is a good candidate to equip ITA’s Research Wind Tunnel.
Figure 6 – Another view of the superficial
mesh for configuration #1.
Boundary conditions were such that
the mass flow rat, at the inlet of the computational domain, and the static Fig. 8 - Static pressure contours along the pressure, at the exit plane, were fixed. The wall of configuration #1. mass flow was prescribed in order to obtain the desired speed at the test section. The The static pressure contours along outlet pressure corresponds to the local the walls of configuration # 7 are shown in atmospheric pressure. Fig 9. This contraction is considerably shorter than configuration #1 and its 0.87. By comparing the curves referring to inflection point, X, is located at x=L/2 (see configurations 1 and 2 it is possible to Fig 4). It can be noticed, that is this case, verify the influence of the inflection point the pressure distribution no longer presents position upon the pressure distribution a variation as smooth as the one in Fig 7. along the contraction’s centerline. The The flow seems to accelerate more abruptly results seem to indicate that the closer the and low-pressure areas near the inflection point is to the contraction’s inlet contraction’s end (beginning of the test the smoother is the pressure distribution. section) become visible, especially at the Finally, it is observed that a linear pressure corners. gradient exists along the test section and, therefore, the tunnel walls need to be diverged accordingly.
Fig. 9 - Static pressure contours along the
wall of configuration #7. Fig. 10 - Static pressure contours along the Fig. 10 shows the same type of wall of configuration #5. result presented in Figs. 8 and 9, now for configuration #5. This configuration is even shorter than number 7. The inflection point is located around 65% of its total length, which is closer to its end than others presented so far. Comparing Fig. 10 with Figs. 8 and 9, all drawn to the same scale, it a significant smoothness loss on the pressure distribution can be noticed. The flow accelerates even more abruptly than in configuration #7 and the areas of low Fig. 11 – Static pressure distribution along the pressure close to the exit plane, became centerline of configurations #1, #2, and #5. much more pronounced. Systematic analysis of the In Fig. 11 is shown the pressure configurations shown in Table IV indicates distribution along the centerline for that configuration #1 is the more suited one configurations 1,2, and 5. The to equip ITA’s Research Low Speed Wind computational domain includes both the Tunnel. In order to further illustrate the flow contraction and the test section. characteristics related to contraction #1 Configurations 1 and 2 have the same some velocity plots for a cross-section plane length, 4.97m, and L/D1= 1.27. The at the test section are shown. Fig. 12 difference between them lies in the exhibits the velocity profile at a given X- inflection point, which for configuration #1 station in the vertical plane of symmetry for is closer to the inlet plane. Configuration #5 the calculation performed with the Standard is shorter with 3.40 m in length and L/D1= k-ε turbulence model. The calculated velocity profile is seen to be uniform, as it should be. In order to verify flow angularity Figs. 13, 14, and 15 show the velocity- component profiles. One can observe that the velocity components in the Y and Z directions are two orders of magnitude smaller than the component in the X- direction, along the tunnel axis. In other words, the flow is parallel to the tunnel centerline.
Fig. 15 – Profile of the Z- velocity along the
Y-axis at the test section for configuration #1.
Fig. 16 shows the total pressure
variation at the symmetry plane xz. The head loss is only 87.2 Pa, which corresponds to about 0.10% of the total pressure at the contraction inlet plane. The boundary layer is thin at the test section Fig. 12 – Velocity magnitude profile entrance and remains that way along it. along the Y-axis for configuration #1. Results indicated a half-degree divergence of the test section walls.
Fig. 13 - Profile of the X-velocity along
the y-axis for configuration #1.
Fig. 16 – Total pressure contours for
configuration # 1 using Reynolds Stress Turbulence model.
Concluding remarks
A successful design of a contraction to
equip ITA’s Research Low Speed Wind Tunnel was achieved with the aid of CFD tools. Several geometries, obtained through Fig. 14 – Profile of Y-velocity along the Y-axis the method introduced by Morel (1975), at the test section for configuration #1. were studied. The numerical simulations of the turbulent, three-dimensional flow, inside each configuration, were accomplished Wind Tunnel Contractions,” ASME Journal of Fluid using Fluent. Among the studied Engineering, Vol 110, pp. 45-47. contractions, configuration 1 was the one 7. Stratford, B. S., 1959, “The Prediction of that presented smoother pressure Separation of the Turbulent Boundary Layer,” distribution along the axial direction as well Journal of Fluid Mechanics, Vol. 5, Part 1, pp. 1-16. as greater flow uniformity at the test 8. Thwaites, B., 1946, “On the Design of section. In spite of its largest length, it was Contractions for Wind Tunnels,” Aeronautical verified that boundary-layer growth along it Research Council, Londres, R&M 2278. is not prohibitive. Further, the probability of boundary-layer separation is practically 9. Wilcox, D. C., 1998, Turbulence Model in CFD, DCW Industries, La Cañada, CA, EUA. zero. Therefore, configuration #1 was chosen to equip the tunnel. Another very 10. FLUENT 5.3 User’s Guide, Vols. 1-4. important result of this work was FLUENT Incorporated, Lebanon, New Hampshire, quantifying the boundary-layer growth USA, 1996. along the test section. Because of this result, 11. de Mattos, B. S., Ferarri, M. A. S. and the top and bottom walls were designed to Leahy-Dios, F, “Transonic Euler Flow Calculation diverge by half degree. The tunnel around a Transport Configuration with Powered- construction ended in December 2002. Engine Effects,” 37th AIAA Aerospace Sciences and Calibration is schedule to take place during Exhibit, AIAA Paper # 99-0529, January 1999, Reno, NV, USA. February 2003.
Acknowledgements 12. de Mattos, B. S. and Oliveira, G. L., “
Three-dimensional Thermal Coupled Analysis of a Wing Slice Slat with a Piccolo Tube,” 18th AIAA The present work is inserted in the Project Applied Aerodynamics Conference, AIAA Paper # of Technological Innovation PITE 2000-3921, August 2000, Denver, CO, USA. supported by FAPESP (www.fapesp.br) under grant 00/13769-0. 13. Fernandes, F. C., and de Mattos, B. S., “Aerothermodynamic Flow Simulation Inside an Aircraft Cabin with Consoles and Screens,” Bibliography European Congress on Computational in Applied Sciences and Engineering, ECCOMAS 2001 1. Bossel, H. H., 1969, Computations of Conference, September 2001, Swansea, UK. Axisymmetric Contractions,” AIAA journal, Vol. 7, No. 10, pp. 2017-2020.
2. Chmielewski, G. E., 1974, “Boundary Layer
Considerations in the Design of Aerodynamic Contractions,” Journal of Aircraft, Vol. 11, No. 8, pp. 435-438.
3. Klin, A., Ramjee, V., e Venkataramani, K.
S., 1973, “An Experimental Study of the Subsonic Flow in Axisymmetric Contractions, ” ZFW, Vol. 21, No.9, pp 312-320.
4. Mehta, R. D. e Bradshaw, P, 1979, “Design
Rules For Small Low Speed Wind Tunnels,” Aeronautical Journal, Paper No. 718, pp. 443-449.
5. Morel, T, 1975, “Comprehensive Design of
Axisymmetric Wind Tunnel Contractions,” ASME Journal of Fluids Engineering, Vol. 97, No. 2 pp. 225-233