Industrial wind tunnel analysis based on current modeling and future outlook

Transcription

1 Industrial wind tunnel analysis based on current modeling and future outlook Rafael José Mateus Vicente Instituto Superior Técnico Abstract The purpose of this work is to study the wind tunnel used by JDEUS, a Portuguese company part of the Denso Corporation. JDEUS produces and develops intercoolers and radiators in their factory unit located in Samora Correia, Portugal, where a closed circuit wind tunnel is used to support both certification and R&D activities. Given the amount of different geometries that JDEUS has in his portfolio the wind tunnel is required to accommodate a wide range of applications reason hence having a variable geometry and therefore being a challenge for the main goal of this work that is to uniformize the flow reaching the testing area where the intercooler is located during the test. To study the existing air flow a number of Computational Fluid Dynamics (CFD) simulations of the wind tunnel have been performed focused on three main cases, each one representing a different intercooler category: box, down face and full face. A special case of the box category was also included because of already existing data. Through misalignment correction, geometrical simplifications, changes in settling chamber volume, duct diameter reduction and pressure loss addition to the air stream resulted in an overall improvement of the air flow observed in the wind tunnel. In addition, two solutions (one based on spheres and another one based on fabric or cloth filters) are presented to uniformize the air flow reaching the test area. Introduction The main goal for this dissertation is to understand the air flow and provide a better understanding of the wind tunnel through the use of Computational Fluid Dynamics (CFD) simulations, obtained results were analyzed and then validated against experimental data so that relevant conclusions can be made and eventually propose changes to improve the wind tunnel air flow. This study is in Mechanical Engineering interest as it requires information from Thermodynamics, Aerodynamics, Fluid Mechanics and Computational Fluid Mechanics. The JDEUS wind tunnel is a closed loop geometry that is presented in the next figure, it s possible to identify two distinct areas: the first one containing the chamber on the left and the second contains the drive section and ducts (often called the machine room). Figure 1. - JDEUS closed circuit wind tunnel Intercooler s geometry depends on the application and the positioning within the vehicle, it is therefore possible to encounter three different types of intercoolers, box, full face and down face, each one of each is intended but not restricted to be mounted in a different position of the vehicle, reason why box intercoolers are typically mounted on top or nearby the wheels. To be able to accommodate all this geometry JDEUS developed the wind tunnel to be a versatile apparatus (made out of 8 metal plates) capable of changing between tests. Please refer to Figure 1.6 and Figure 1.7 The following variables can be recorded: atmospheric pressure and humidity; inlet and outlet air and water temperatures; rate of air and water flow; air flow velocity; pressure loss of both water and air side. 1

2 1. Mathematical and Numerical Model 1.1. Governing equations This work, much like all CFD, is based on three fundamental governing equations continuity (Mass conservation), momentum (Newton s second law) and energy (Energy conservation). The equation that express mass and momentum conservation in the flow, the Navier-Stokes equations, are expressed in 1 and 2 in their tensorial form. ρ t + (ρu x j ) = s m i (ρ u i ) + (ρ u t x j τ ij ) = p + s j x i i where t stands for time; x i the Cartesian coordinates (1) (2) (i=1,2,3,); u i is the velocity component in the x i direction; p represents the static pressure; ρ the density; τ ij the components of the stress tensor; s m the mass source and s i the momentum sources. In this case s m = 0 since there are no mass sources SIMPLE algorithm The model will make use of a SIMPLE solver algorithm [1] whose basics steps in the solution update are as follows: set the boundary conditions; compute the gradients of velocity and pressure; solve the discretized momentum equation to compute the intermediate velocity field; compute the uncorrected mass fluxes at faces; solve the pressure correction equation to produce cell values of the pressure correction; update the pressure field p k+1 = p k + urf p, where urf is the under-relaxation factor for pressure; update boundary pressure corrections p b ; Correct the face mass fluxes: m f k+1 m f + m f ; correct cell velocities: U k+1 u V p v ; where p is the a P pressure corrections gradient, a P v is the central coefficients vector for discretized linear system representing the velocity equations and V is the cell volume and update density due to pressure changes (only for compressible gases) Turbulence modelling Turbulence is present in the almost every flows encountered in nature and among several industrial applications. Turbulent flows are characterized by fluctuating velocity fields. These fluctuations mix transported quantities such as momentum, energy, and species who might be traveling with the flow, causing transported quantities to fluctuate as well. In this work, as will be seen, there was the need to investigate some turbulence models in order to obtain a good agreement between computational results and experiments. All models considered throughout this work are available in STAR-CCM Reynolds Average Navier Stokes (RANS) Due to the computational cost of both DNS and LES for most geophysical and engineering applications, modelers are restricted to RANS approaches commonly based on turbulent kinetic energy (TKE) closure schemes. The most widely used RANS models today are two equation models which solve two transport equations for the properties of the turbulence from which the eddy viscosity can be computed. The best known of these models is the k- ε model which requires the solutions of the turbulent kinetic energy equation and dissipation of turbulent kinetic energy equation. The other model addressed in this work is the k ω model Boundary Conditions In every CFD model, apart from the geometry definition and consequent mesh there is the need to prescribe boundary conditions. In wind tunnels, air flows by the action of a ventilator, represented in this work s simulation as a pressure increase on momentum equation. Walls are modeled as a no-split condition; which means that they are impermeable with a zero velocity condition of the wall. It is near the walls where the gradients are larger and where the momentum is more affected. It is therefore of high importance in numerical simulation to accurately predict the near-wall flow. 2

3 Pressure loss dp (Pa) 1.4. Additional physics models 2. Geometry presentation Physical mechanisms from the apparatus must be modeled in order to successfully represent the reality; a modulation must be performed so that Navier-Stokes equations can be accurately computed inside the computational domain Ventilator The ventilator was simulated using a constant momentum source in the momentum equation, since that was no data from the ventilator an iterative approach was conducted to ensure the intended prescribed velocity on test section Wind tunnel Geometry A The 3D wind tunnel geometry was created based on measurements and observations performed on site using both laser and standard measuring tools. The wind tunnel geometry is simplified to only include objects oft at least approximately 40mm, which means that small objects for example cables are not modeled Intercooler pressure loss function Experimental data from the outer flow was provided so that pressure loss functions could be computed and a custom negative momentum source term could then be applied to the simulation. External air flow 1600 y = 11.07x x 1200 Figure 1.2 Geometry side view perspective of the settling chamber and the testing area. 3D geometry to the left and corresponding pictures to the right. The ceiling of the wind tunnel consists of a complex set of heat exchangers (coolers) and fans to control the air temperature. These heat exchangers are made of the same material and the geometries of the intercoolers and uses a cold water circuit while fans are set in motion so that air is forced to pass through the heat exchangers while cooling Case Polinomial (Case) Velocity (m/s) Figure 1.1 Pressure loss and velocity data points obtained from the apparatus using a central geometry Resulting in the following equation for the pressure loss: P i,j,k x i,j,k (v) = 11,07 V i,j,k ,697 V i,j,k Figure 1.3 Geometry top view, heat exchangers As said, JDEUS uses a closed loop wind tunnel and the complete geometry is presented in the following figure where the air flow motion is also sketched. 3

4 Figure Sketch of the air flow, the Drive Section is colored in magenta 2.2. Intercooler Geometries There is a wide variety of products developed and produced by JDEUS; this section will introduce the reader to the reason behind such a wide range of geometries. Specific applications require different intercooler s output such as the required heat to be exchanged, not less important, the available space and positioning for the intercooler to be mounted, hence the vast range of products allowing JDEUS to provide tailored solutions for each set of requirements Testing section To be able to accommodate all the wide range of geometries presented in the test section JDEUS developed the wind tunnel with a versatile apparatus capable of changing between tests, this apparatus is made out of 8 metal plates that the test section with an inlet and an outlet. A picture of the apparatus and the equivalent 3D CAD is presented below. Figure 1.6 Front view of the test section apparatus (a picture on the left and the CAD model on the right) It is possible to encounter three different types of intercoolers: box, full face and down face, each one of each is intended, but not restricted to be mounted in a different position of the vehicle. Reason why box intercoolers are typically founded top mounted or nearby the wheels, full face intercoolers are typically mounted at the very front of trucks while down face intercoolers are positioned closer to the front bumper of sports cars. These requirements can be translated into geometric parameters such as H x W x T according to the terminology used across the automotive industry. Figure 1.7 Test section geometry configuration. (Clockwise starting from top left) Box, Central, Full-face and Down face. It is also possible to see the outlet duct through the testing area. Figure Intercooler terminology Intercooler geometry High [mm] Width [mm] Thickness [mm] Box Down Face Full Face Table Intercooler geometries under analysis Figure 1.8 Detailed side view of the test section with the intercooler colored in cyan the arrows representing the air flow. 4

5 2.4. Geometric issues Despite being the first geometry and without any CFD simulation so far it is possible to point out some geometrical problems that might affect the air flow. The lack of geometry symmetry due to physical restrictions inside the wind tunnel chamber. There is also a small misalignment between the inlet (of the settling chamber) and the outlet (of the test section) duct that might lead to a nonsymmetric flow reaching the settling chamber. The volume of the settling chamber might not be enough for the fluid to actually settle. All this small issue might influence the quality of the velocity profile reaching the test section and moreover the intercooler s upstream face, this is possible to confirm in the results section. 3. Error Analysis obtained in a cell of around 9.5 Million cells. The selected grid is presented in the next figure. Figure 1.9 Trimmed mesh Reader must be advised that the screen might not be visible on pictures but it s considered on the geometry, either if it s being used as a pressure drop or not. 5. Results and discussion From the analysis performed using the geometry and mesh already presented the following velocity field Numerical solutions of fluid flow and heat transfer problems are only approximate solutions, in addition to errors that might be introduced by boundary condition, numerical solutions always include three kinds of systematic errors: Modeling Discretization Iteration From the performed qualitative evaluation of the simulations it seems that CFD is delivering a good solution, which was confirmed, although a quantitative evaluation of the simulation an error of 6.5% was found for pressure loss quantification. 4. Mesh Figure 1.10 Geometry A with central geometry mean velocity profile 50 mm of the testing duct (On the left); Incident mean velocity magnitude on intercooler s upstream face (on the right) From these results, some geometry changes will be proposed. These are addressed in the next section. 6. Geometry Changes The following changes were made to the model: Discretization approximations introduce errors which decrease as the grid is refined, and therefore the order of the approximation is a measure of accuracy hence the importance of having a good mesh. The cubic cell and a trimmed grid will then be used as the standard cell/mesh type and the level of detail achieved using a characteristic cell length 0,025m with a reduction of 25% of the testing area resulted in a very small difference when comparing these mesh to results Outlet duct diameter reduction All deflectors in the settling chamber should be removed The volume of the settling chamber should increase Elimination or reduction of the geometric issue identified in sub section 2.4 5

6 As a conclusion one can affirm that deflectors are not a benefit for the flow since it increases instabilities inside the wind tunnel chamber. Figure 1.11 Wind tunnel cross section where it is possible to identify the Settling Chamber (A), the Test Section (B and Cyan colored), the Air Cooler (C), the Drive Section (D and Magenta colored)) and two additional pressure losses (E1 and E2) Figure 1.14 Test section geometry configuration (clockwise starting from top left: Central, Box, full-face and down face) 7. Results after geometry changes CFD result from the new wind tunnel geometry is presented below. IC Figure 1.12 Wind tunnel top view, ceiling heat exchangers and inlet duct. Figure 1.15 Geometry B1 with box geometry velocity profile 50 mm of the testing duct (On the left); Incident velocity magnitude on intercooler s upstream face (on the right) Figure 1.13 Wind tunnel geometry front view, ceiling heat exchangers (A) and cable tray (B). Despite all changes, testing section remained unchanged, hence all configurations already presented are still possible to be tested, it is still interesting to have an insight of the test section (after all these changes) this can be found on the next figure containing the already presented changes made to the surroundings. Comparing this with the previous result, it is possible to validate changes made to the original geometry, however flow approaching the intercooler still present non homogenized areas and therefore further steps are required to develop a suitable solution and achieve the objective of this work. 8. Flow homogenization To improve flow homogenization a pressure loss was proposed, it is required to compute the needed value in order to understand the required filter to be applied; as there is a wide range of possible filters to be implemented with the same purpose two solutions were selected: one based on cloth filters that are widely used across the industry as an anti-pollution measure, (as they are used to filter air flow particles in bag houses). It is required to understand both pressure loss value 6

7 and still have a suitable solution to accommodate all geometries. Regardless of the solution, it is important to compute the required pressure drop before selecting the feasible approach. That said, the following picture is the result from CFD a simulation where homogenization was achieved. Porous media IC Figure Geometry B1 with box configuration, velocity profile at 50 mm away from the testing duct (On the left); Incident velocity magnitude on intercooler s upstream face (on the right). Figure 1.17 Test section detail of the wind tunnel geometry B cross section, a packed bed of spheres is represented as a possible solution to increase flow homogenization. With a typical pressure drop ranging from 60 to 200 Pa [3], in addition, it is always possible to create patterns according to requirements. And for this range of pressure drop both woven or felt can be used however with a different weight, woven ranges from 170g/m 2 up to 340 g/m 2 and felt ranges from 340 g/m 2 up to 170g/m 2 [4]. From this CFD simulation a pressure loss of 6404 Pa is required to homogenize the flow, with value it is then possible to understand the required parameters for the for the required filter. For the packing bed spheres (presented in Figure 1.17), a common correlation to be used is the Ergum correlation with a correction for random packing made by Ribeiro, A.M., Neto, P. and Pinho, C. in [2]; to reduce the total solution cost spheres should have the same diameter and therefore a monosize random distribution of spheres is considered leading to the following equation: P (1 ε) (1 ε) = L d p ε 3 v (150μ ρv ) A c d p A c Using the required pressure drop as a constant for the iterative approach the parameter ε is computed for a given solution, in this case spheres with 42mm of diameter can be considered as a possible solution. (3) Figure Test section detail of the wind tunnel geometry B cross section, a piece of fabric/cloth is represented as a possible solution to increase flow homogenization. 9. Conclusions The purpose of this work was to study and provide a better understanding of a closed loop wind tunnel used by JDEUS to support both certification and R&D activities. After modeling the wind tunnel geometry using a 3D software, the model was imported to STAR-CCM where Computational Fluid Dynamics (CFD) simulations were performed, results were then analyzed and validated against experimental data supplied by JDEUS. After the first set of simulations it was possible to propose changes to the following changes proposed: Outlet duct diameter reduction; Settling chamber deflectors should be removed; 7

8 Settling chamber volume should increase; The distance between the intake for the settling chamber and the intake for the test section should be increase; Misalignment between test section and outlet duct should be possible to avoid; Misalignment between inlet and outlet duct should be reduced; Geometry symmetry should be increased to reduce secondary flows inside the settling chamber; filters were not the best technique to obtain such a pressure loss give the required pressure loss value. By applying the corrected Ergum equation presented in the last chapter of this work resulted on a feasible solution for the wind tunnel homogenization problem, a packed bed of table tennis balls (42mm) can be installed in the upstream area of the test section and provide the required pressure drop. This result, after validated will prove the usability of CFD simulations to improve other wind tunnel geometries. 10. Further work In addition to the already changes, an additional filter was recommended to increase flow homogenization on the outlet duct. Although far from what was expected as a result for this work, velocity profile proven to be more homogeneous in the second set of CFD simulations (where the used computational domain was already changed to accommodate all geometry changes proposed earlier in this work). JDEUS already applied the proposed changes to the geometry and new tests performed on the wind tunnel proven that it is already delivering better results than it was before, having that said it is clear that the objective for this work was clearly achieved, at the same time it was also provided a better understanding of the apparatus during the follow up meetings with the research and development team increases their confidence on new wind tunnel. To obtain the intended homogeneous velocity profile at the test section two different solutions were proposed - the first one based on a packed bed of spheres and another based on cloth filters, both representing an extra pressure loss that should homogenize the velocity profile. The wind tunnel geometry containing this additional pressure loss was simulated and presented in the last chapter of this work, where results proven to be as good as expected, knowing the required pressure drop it would be a question of understanding which filter should be applied, taking into account the results from CFD simulations for the pressure drop it became clear that cloth Current work focused on the air flow arriving to the test section, future simulations could account also the influence on the fluid, allowing a better understanding of the flow arriving the drive section and therefore understand even better how the flow behaves on the outlet duct. In terms of geometry this work focused on a selection of geometries which should represent the wide range required to be tested at JDEUS facilities, in order to have a better understanding on how pressure losses can help on improving wind tunnel results it might sound reasonable to simulate a wider range of geometries. The fullface geometry represented a challenge and therefore was not possible to include in this study, geometrical issues and convergence difficulties can be pointed out as major difficulties to obtain valuable results from the CFD simulations, being an intercooler type with a higher aspect ratio it would be interesting to study this geometry even further and to be able to obtain suitable results that after validation could be used to improve the wind tunnel even more. The velocities used for CFD simulations in this work were limited, a prescribed value for each geometry was used and the subsequent simulations would use this value as a reference, as it was possible to see during this work, the results are affected by the prescribed velocity and therefore it would be interesting to iterate for a wider range of velocities. The pressure drop equation was assumed to be the same for every geometry introducing another error to the CFD simulations, this could be reduced by using more 8

9 data and by computing a tailored equation for each geometry, as for the filter pressure drop, depending on the solution adopted by the company it might sound reasonable to review the pressure drop function for filters [Online]. Available: [Accessed 2015 January 10]. The test section is a complex apparatus that allows the wind tunnel to accommodate a wide range of geometries, it s composed out of 8 metal plates and it can be divided in two similar sets of 4 metal plates each, one set lays upstream and the other one downstream of the test section. It would be interesting for a future study to approach the need of the upstream component of the test section, allowing the air arriving the intercooler coming directly from the settling chamber. The cooler used to control the air flow temperature is a complex set of fans that forces the air passing through heat exchangers while it cools, given the complexity of this secondary flows the impact on the air moves was not accounted in this work however it would be interesting already to understand such impact. References [1] Adapco, STAR-CCM user's manual version [2] A.M.Ribeiro, P. Neto and C. Pinho, "Mean porosity and pressure drop measurements in packed beds of monosized spheres: Side wall effects," International Review of Chemical Engineering, vol. 2, pp [3] A. S. Mujumdar, Handbook of Industrial Drying, Fourth Edition, Boca Raton: CRC Press, [4] L. K. Wang, N. C. Pereira and Y.-T. Hung, Air Pollution Control Engineering, vol. I, Totowa: The Humana Press, Inc., [5] V. Brederode, Fundamentos de Aerodinâmica Incompressível, Lisboa: Vasco de Brederode, [6] S. K. Venayagamoorthy, J. R. Koseff, J. H. Ferziger and L. H. Shih, "Testing of RANS turbulence models for stratified," CENTER FOR TURBULENCE RESEARCH, 23 August 9