Experimental and numerical investigations of cooling drag Downloaded from: https://research.chalmers.se, 2021-04-09 20:18 UTC Citation for the original published paper (version of record): Hobeika, T., Sebben, S., Löfdahl, L. (2017) Experimental and numerical investigations of cooling drag Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering, 231(9 N.B. When citing this work, cite the original published paper. research.chalmers.se offers the possibility of retrieving research publications produced at Chalmers University of Technology. It covers all kind of research output: articles, dissertations, conference papers, reports etc. since 2004. research.chalmers.se is administrated and maintained by Chalmers Library (article starts on next page)
11
Embed
Experimental and numerical investigations of cooling drag · Experimental and numerical investigations of cooling drag Downloaded from: , 2020-10-15 18:22 UTC Citation for the original
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Experimental and numerical investigations of cooling drag
Downloaded from: https://research.chalmers.se, 2021-04-09 20:18 UTC
Citation for the original published paper (version of record):Hobeika, T., Sebben, S., Löfdahl, L. (2017)Experimental and numerical investigations of cooling dragProceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering, 231(9
N.B. When citing this work, cite the original published paper.
research.chalmers.se offers the possibility of retrieving research publications produced at Chalmers University of Technology.It covers all kind of research output: articles, dissertations, conference papers, reports etc. since 2004.research.chalmers.se is administrated and maintained by Chalmers Library
(article starts on next page)
Experimental and NumericalInvestigations of Cooling Drag
As the target figures for CO2 emissions are reduced everyyear, vehicle manufacturers seek to exploit all possiblegains in the different vehicle attributes. Aerodynamicdrag is an important factor that affects the vehicles fuelconsumption, and its importance rises with the shiftfrom the New European Driving Cycle (NEDC) to theWorldwide harmonized Light vehicles Test Cycle (WLTC)which has a higher average speed. In order to reduce vehicledrag, car manufacturers employ the use of grill/spoilershutters which reduces the amount of air going through
the vehicles cooling system, also known as cooling flow,thus reducing both its cooling capability and the resultantcooling drag. Moreover, the introduction of grill shuttersopens up many possibilities for optimizing the cooling flow,especially on hybrid electric vehicles and fully electricalvehicles. Even more control over cooling efficiency can be
1Chalmers University of Technology, Gothenburg, Sweden
Corresponding author:Lennart Lofdahl, Applied Mechanics, Chalmers university of Technol-ogy, Horsalsvagen 7A, Gothenburg, 41296, Sweden
Prepared using sagej.cls [Version: 2015/06/09 v1.01]
2 Journal Title XX(X)
obtained by partly opening grill shutters or creating newpackaging scenarios, yet the increase of uncertainties inoptimization parameters is a challenge, both experimentallyand numerically.It is important to be able to predict the drag and liftchanges due to different grill configurations in a constantoptimization cycle between reducing drag and havingsufficient cooling flow. With different alterations to grillopenings it is difficult to isolate the increase in drag dueto cooling flow from the interference effects this flowhas with other vehicle geometry. Thus the term coolingdrag emerges which is in practice1 the difference in dragbetween completely blocked cooling inlets and open inletfor the different grill configuration. Depending on theanalysis being performed, there are many formulations usedfor cooling drag in both simple and complex forms2–5.Garrone6 has performed an experimental investigationwhere the drag on the cooling package is measuredseparately using load cells then compared to cooling drag.In his comparison the cooling package drag contribution of17 count (0.017 Cd) was significantly larger that the 3 countcooling drag measured on the vehicle. As the coolingpackage inlets and outlets are both ducted then the coolingpackage contribution is mostly attributed to the pressureloss across the radiator; by placing the cooling outlets in thelow pressure wheelhouses a low cooling drag was achieved.Also for even more complex vehicles, Williams7 has shownthat the losses in the engine bay are minor compared tolosses from the cooling package.Although challenging to separate the force acting on theradiator and the cooling drag in wind tunnel tests, it is quiteeasy in Computational Fluid Dynamics (CFD) simulations.In previous works, simulations have been able to predictgeneral trends in overall vehicle forces yet the predictionof magnitudes remains challenging8. The cooling dragmeasured in wind tunnels is still influenced by blockageeffects, and when specific quantities like radiator forcesare being measured a correction could be needed as theblockage could affect the mass flow in the radiator9. Also,discrete pressure based mass flow measurements couldintroduce errors to cooling flow quantification due to thecomplexity of the flow and the measurement grid density10.For the mentioned reasons, this work quantifies the pressure
loss through the radiator core by means of radiator forceand mass flow measurements under the effect of largegrill blockages, with the primary goal of comparing toCFD simulations. Effects of wind tunnel blockage andmeasurement grid discretization are discussed from bothexperimental and numerical perspective. Also a coolingflow indifference to wheel rotation, moving ground, andmesh refinement is presented. Finally, the changes in overallvehicle forces due to cooling flow are discussed and theradiator force is put in perspective with cooling drag.
Methodology
Full scale wind tunnel experiments and numericalsimulations have been performed on a Volvo S60 Y283model. Flow through the radiator, force acting on theradiator core, as well as overall vehicle forces are measuredand analyzed. As the radiator flow and force are the mainaim of this investigation, the geometry around it has beenreplicated with high accuracy in simulations to match thetest object.
Experimental Setup
The wind tunnel tests are performed at the Volvo Carswind tunnel, which is a closed loop wind tunnel witha slotted wall test section and equipped with a five beltmoving ground system11. The tests reported in this paperare performed at 100 km/h with an active moving groundsystem and rotating wheels, unless otherwise specified. Thecar engine bay has been modified for this particular test;most components of the cooling system have been removedwith only the radiator and the fan shroud remaining. Bydoing so, the forces acting on the radiator core can bemeasured using a simple setup with two load cells oneach side of the radiator. In order to eliminate interferencefrom oncoming flow and to make it more representableof the normal vehicle conditions, special aluminum ductsguide the air from the bumper inlets straight to the radiatorcore thus shielding off the water tanks. With this setup sixdifferent grill configurations have been tested to evaluatethe effect of grill blockage ranging from completely opento completely closed as presented in Figure 1. Thesegrill blockages only cover the grill opening while the
Prepared using sagej.cls
Hobeika et al. 3
(a) Grill 1: Open grill (b) Grill 2: Production grill
(c) Grill 3 (d) Grill 4
(e) Grill 5 (f) Grill 6: Closed grill
Figure 1. Grill designs investigated.
lower bumper opening is left completely open for allconfigurations. A representation of the bumper openingsand aluminum ducts is shown in Figure 2a, while the fanshroud and load cells are shown in Figure 2b. Note thatthe fan itself and all three flaps have been removed thusresulting in one large circular opening and three smallrectangular ones.The flow through the radiator is measured using 48 Ruijsinkprobes12 placed over a rectangular grid in the radiatorcore, shown in Figure 3. The probes measure a local totaland a local static pressure from which a local velocitythrough the probe can be determined. They have beenmounted in the radiator and calibrated in a test rig, beforeplacing the radiator in the car, which resulted in individualcorrection curves that convert the local probe velocity tolocal radiator core velocities13. The correction curves arevelocity dependent thus a different correction is given forlow and high velocities, however at speeds below 2 m/s thecorrections vary largely with very small changes in velocity.From these 48 measurement points the velocity distributionacross the radiator can be plotted, using linear interpolationor extrapolation, and the average radiator core velocitycan be determined. However, as all pressure measurementtechniques have limitations at low velocities where theviscous effects are strongly pronounced and they affectthe measurement accuracy14, all velocities below 2 m/s areunreliable10.
(a) Front blockage: bumper and aluminum ducts
(b) Rear blockage: fan shroud and load cells (F1 and F2)
Figure 2. Figure showing radiator blockage from front andrear.
(a) Ruijsink Probe (b) Probe arrangement in the radiator
Figure 3. Geometrical setup of the 48 probes.
Numerical Setup
The process for CFD simulations is carried out usingcommercial software: Ansa, for CAD cleaning, Harpoon,for meshing, and Fluent for solving. Much care wasdedicated to ensure that the numerical model replicatesthe geometry in the vicinity of the radiator. The front ofthe vehicle, aluminum ducts, and known flow outlets are
Prepared using sagej.cls
4 Journal Title XX(X)
(a) Open road
(b) Virtual tunnel
Figure 4. The test vehicle in: (a) open road, (b) virtual windtunnel conditions.
resolved down to 2.5 mm while the radiator, load cells, fanshroud, and potential leakage areas are resolved down to1.25 mm. The remaining of the vehicle refinements varyfrom 1.25 to 5 mm following general external aerodynamicsguidelines. The simulations have all been performed inopen road conditions as well as in a virtual model of theVolvo wind tunnel15. The flow inlet and outlet boundaryconditions in the wind tunnel are tuned in order toreplicate the reference pressure drops measured by twoprobes located in the nozzle. This setup led to two setsof simulations, in open road and virtual tunnel, with 190and 300 million cells respectively, shown in Figure 4.The simulations are solved with the k-epsilon turbulencemodel, enhanced-wall treatment, Green-Gauss node-basedgradient scheme, and second order convective discretizationschemes for pressure, momentum, turbulent dissipationrate, and turbulent kinetic energy. They are performed for avehicle velocity of 100 km/h with Moving Reference Frame(MRF) for modelling rim rotation.
Figure 5. Percentage blockage correction needed to getequivalent of virtual tunnel values to open road.
Results and Discussion
Wind-tunnel blockage effect
Wind tunnels are a tool used by aerodynamicists, in order toquantify and improve the different aerodynamic propertiesof a vehicle. As it strives to get as close as possible toon road conditions, the result is a complex environmentwhich introduces different errors, in the forms of blockageeffects and measurement uncertainties, which need to beaccounted for. In general the aerodynamic forces measuredin the wind tunnel are an over prediction to the equivalenton road conditions. For example the drag force measured inthe wind tunnel throughout these tests requires a correctionof about 6% in order to represent the on road conditions.This raises concerns about the type of corrections thatneed to be applied to force measurements on the radiatorcore, and if the force on the core is over predicted dueto the blockage effects then so is the mass flow throughthe radiator and therefore a correction is needed there aswell. This has been investigated using CFD simulationswhere the car has been simulated in a virtual model of theVolvo wind tunnel and compared to open road simulationsfor the six different grill configurations. The numericalresults of mass flow change, radiator core force change,and drag coefficient change can be seen in percentagesin Figure 5. Distribution wise, no major changes couldbe seen apart from a general increase in the velocitymagnitudes around the high velocity areas, an example isshown on Grill 3 in Figure 6. Figure 5 shows that on
Prepared using sagej.cls
Hobeika et al. 5
(a) Grill 3: Open road (b) Grill 3: Virtual tunnel
Figure 6. Comparison of velocity distribution for Grill 3 whensimulated on open road or in virtual wind tunnel.
road conditions have around 6% lower overall Cd whichmatches well with the correction applied in the wind tunneltests, however it could also be seen that the correctionneeded for the force acting on the radiator core is not of thesame magnitude, more importantly it varies with a constantdrop as the blockage increases. A similar consistent trendis also observed in the mass flow of the radiator whichfurther confirms the need for a correction. However, asfinding such a correction is difficult, an alternative approachhas been adopted where all comparisons of mass flow andforces acting on the radiator are performed between testsand simulations in a virtual tunnel, instead of open road.Although this potentially introduces additional errors tothe simulations as the modelling of the tunnel in CFDis not necessarily accurate, the changes on the overallvehicle forces due to the presence of the tunnel are quitesimilar to those resulting from the test corrections andthus the blockage effect introduced in CFD is consideredrepresentative of the physical tunnels blockage effect.
Effect of the 48 probe measurement grid
In the wind tunnel tests, a measurement grid of 48 probesis used thus resulting in 48 local velocities. These pointsneed to be interpolated/extrapolated over the radiator facein order to plot the distribution and to calculate the massflow through the radiator. In the CFD simulations however,the velocity distributions and mass flows presented arethe result of a significantly finer grid with a few hundredthousand points. The 48 probe grid distorts the distributionand the definitions while also missing out on most of the
(a) Grill 1: CFD plot (b) Grill 1: 48 Probe plot
(c) Grill 3: CFD plot (d) Grill 3: 48 Probe plot
Figure 7. Comparison of numerical velocity distribution fortwo different grills when plotted directly from CFD vs plottedusing only 48 measurement points similarly to experiments.
velocities outside the grid area. Thus for a more accuratecomparison, the local velocities at the location of the 48probes are extracted from the CFD simulations and then thedistribution is plotted in an identical manner to that used inthe wind tunnel tests. A further motivation can be seen inFigure 7 which compares the velocity distribution over twogrills between continuous normal CFD plots and discrete48 probe plots. The qualitative distribution varies and adeviation of quantitative mass flow results is introducedwith an under prediction of 1.6% for Grill 1 and an overprediction of 2.0% for Grill 3. Also note that with theinterpolation, larger errors in quantifying the effects ofconfiguration changes are introduced. For example, a CFDcomparison between Grill 1 and 3 shows a 6.3% decrease inmass flow, yet using the 48 probe approach the equivalentreduction is 9.6%.
Mass flow and force predictions
Given the effects discussed, six different grill configurationshave been tested in CFD on open road conditions andin a virtual tunnel. The results of the radiator core forcedifference to experiments on the radiator core are presentedin Figure 8. The uncertainty of the load cell measurements
Prepared using sagej.cls
6 Journal Title XX(X)
Figure 8. Delta in force prediction of the radiator core forcebetween CFD and experiments. The dotted line outlines theuncertainty of the force measurement in experiments.
is ± 2 N, thus the forces computed in CFD are mostlywithin the uncertainty of the measurements, with a slightover prediction for Grill 2. It could also be seen thatthe force acting on the radiator increases between openroad and virtual tunnel. Figure 9 presents the mass flowcomparison to experiments, while extracting mass flowsfrom CFD using both full CFD integration as well asthe 48 probe approach as it is believed to be morerepresentative of the experimental setup. Although theinitial error estimate of the pressure probes is believed tobe within ± 5% when using 48 probes, the results reflectfar larger deviations as the blockage over the grill increasesup to the extreme condition where the grill is completelyclosed. It is interesting to note that looking only at thedifference from open road simulations the delta betweenCFD and experiments is within the 5% margin, except forGrill 6 which is clearly off. Figure 9 shows how using the48 probe approach influences the mass flow differently fordifferent configurations, while a clear trend of mass flowincrease can be seen between open road to virtual tunnelwhich agrees well with the increase in forces. With a goodmatch in forces between CFD and experiments, a significantmismatch in mass flow was not to be expected and thusa closer look into the distribution is performed based onFigure 10 comparing the virtual tunnel simulations andexperiments. There is quite a clear qualitative difference indistribution between the CFD simulations and experimentalresults however improvements are detected when plottedusing the 48 probe approach as this is what the experimentalvalues are based on as well. However, low velocities below
Figure 9. Delta in mass flow prediction through the radiatorbetween CFD and experiments. The results are shown usingdirect CFD prediction and prediction using 48 virtual pointmeasurements at identical positions to the probes in theexperiments.
4.5 m/s seem to cause the largest error as they seem tobe lower in the experiments than the CFD predictions.These can be seen behind the crash beam and the differentgrill blockages as low velocity areas in the experimentsand although they are reasonably low velocities in CFDas well they are off by 1 to 2 m/s thus resulting insignificant mass flow errors when integrated over the areathey cover. Furthermore, using the velocity distributionfrom pressure measurements, the theoretical force acting onthe radiator can be calculated13. A clear mismatch betweenthe theoretical and measured force values is detected withthe pressure probes under predicting the force by 0.6, 4.8,1.6, 7.9, 5.4, and 14.6%, from Grill 1 through 6 respectively.Even though these figures cannot be directly compared tothe mass flow percentage differences, presented in Figure9, as the force and mass flow are not linearly proportional,it does give an indication of a large discrepancy between themeasured velocity distribution and the measured forces. Apossible explanation to this could be that the probes loseresolution at high angles of oncoming flow, for exampledownstream of large blockage areas like a crash beam ora fully closed grill where pitch angles of above 65 degreescould be seen in the CFD simulations. This would likelyaffect the pressure readings as a probes calibration curveis obtained in ideal stable conditions in a test rig whichnaturally does not take into account high turbulence andseparations at the inlets of the probes.
Figure 10. Velocity distribution over the radiator for the sixdifferent grill configurations from experiments and virtualtunnel CFD simulations, plotted using 48 virtual probe grid.
Effect of wheels, moving ground, and meshresolution
Grill configurations have shown clear effects on coolingflow and cooling drag, however this is to be expected giventhe large blockage they impose at such close proximityupstream of the radiator. The outlets for cooling flow onthe other hand are not clearly defined. After the flow passesthrough the fan shroud it exits in three major locations: thetunnel at the center of the car where exhaust pipe is located,or the two front wheelhouses. Rim designs and movingground systems are known to have significant effects onthe vehicles’ overall drag and cooling drag by alteringthe underbody flow. This has been investigated by runningthe following configuration: fully covered rims, stationaryfront wheels, stationary rear wheels, and fully stationaryfive belt system. All of these configurations had significantchanges on the overall vehicle drag yet only small changes,within uncertainty limits, could be detected on coolingflow, radiator velocity distribution, and the force acting onthe radiator core. Thus it is believed that, for the testedvehicle, these configurations affect the interference drag atthe cooling flow outlets yet have little to no effect on theflow inside the engine bay.In agreement with the experimental results, simulationswith fully closed rims showed very small effects on thecooling flow and radiator forces. However the position ofthe fan shroud had a large effect where moving the fanshroud 10 mm downstream changes the flow through theradiator by 10%, roughly corresponding to 1% per mm.Also a mesh dependency study has been performed wherea coarser and finer overall vehicle mesh with a max surfacecell size of 2.5 mm have been analyzed with the changesin cooling flow not exceeding 2% between the coarsest andfinest resolution, thus showing mesh independent results forthe cooling flow and radiator force quantification. A minorshift of around 2 drag count and 5 to 10 lift count in theoverall vehicle forces was observed which is attributed toa minor mesh dependency of the results.
Effects on vehicle forces
The cooling drag prediction for CFD simulations andexperiments are presented in Figure 11. Note that the open
Prepared using sagej.cls
8 Journal Title XX(X)
road CFD simulations are to be compared with the correctedwindtunnel results while the virtual tunnel simulations areto be compared with uncorrected windtunnel results thuseliminating the effects of the corrections used in the tunnel.The open road simulation results under predict cooling dragin most cases by about 5 count, however the results improvesignificantly when a virtual tunnel is used where the coolingdrag predictions lie within 1.5 count which is within theuncertainty margins of both experiments and simulations.Even though the front lift changes are quite extreme as
Figure 11. Cooling drag comparison between experimentsand CFD. For a representative comparison, the experimentalcooling drag figures are presented with and without tunnelblockage correction as the virtual tunnel CFD simulationsshould be compared to the uncorrected tunnel results.
shown in Figure 12, the simulation results seem to predictthe results significantly better than with the virtual tunnelwhere the largest over prediction for Grill 1 is reduced from16 to 8 count. However for a change of 114 count, an overprediction of 8 count is still acceptable.The changes in rear lift are of smaller magnitude, as shown
Figure 12. Cooling front lift comparison between experimentsand CFD.
Figure 13. Cooling rear lift comparison between experimentsand CFD.
in Figure 13, However the virtual tunnel did not seem toimprove the predictions as the results seem to be a constantunder prediction. This under prediction is of the order of5 count which is considered within the uncertainty of thetunnel measurements and CFD simulations.
Cooling drag vs radiator drag
Given that the numerical results for force predictionacting on the radiator and overall cooling drag resultsmatched well, the following section is solely based onexperimental results. After the force acting on the radiatorcore is measured, a drag contribution can be estimated,this is reported as radiator drag in Figure 14. The coolingdrag for each of the configurations is also presented andin comparison the cooling drag is roughly three timeslower than the drag force acting on the radiator. HoweverGarrone6 reports a cooling drag more than five timeslower than the cooling package drop, yet his configurationincluded ducted outlet into the wheel houses which havelow pressure, thus significantly reducing the interferencedrag and possibly even having negative interference.
Conclusion
In summary, an investigation of cooling flow quantificationhas been performed both numerically and experimentally.A particular effect investigated in CFD simulations hasbeen that of wind tunnel blockage which showed significanteffects on the results. Although this cannot be generalizedto all wind tunnels, it is an effect worth investigating whenlooking at accurate comparison between CFD and tests.
Prepared using sagej.cls
Hobeika et al. 9
Figure 14. Contribution of radiator force to drag incomparison with cooling drag.
Also for validation purposes, a more detailed analysis ofthe accuracy of pressure probes is required as a significantdiscrepancy in experimental results could be seen betweenprobe and force measurements when the blockage imposesa highly non-uniform distribution through the radiator andhigh pitch angles. Although the forces seem to match quitewell with CFD predictions, it is believed that the probes areunder predicting the flow through the radiator when locatedin the wake of upstream blockages. In the case of a morecomplete cooling package with a condenser upstream ofthe radiator, this could act as a flow straightener and enablethe pressure probes to give more accurate measurements,yet this remains to be investigated in future work.It has also been shown that using virtual measurementpoints in CFD results in a more comparable qualitativedistribution when validating to experiments even thoughit could offset the results significantly. Finally, includingthe geometry of the physical wind tunnel in CFD hasshown to improve the prediction of cooling drag and liftdistribution changes to a level where the error lies withinthe uncertainty of the tests and simulations.Although the cooling package in this work is relativelysimplified, the baseline changes in cooling drag and lift arevery similar to tests performed with a complete coolingpackage thus similar trends are expected.
Acknowledgements
The authors would like to thank Volvo Cars and Volvo Trucks
for providing access to their testing facilities. Special thanks to
Dr Peter Gullberg at Volvo Trucks for his help with calibration
and Dr Christoffer Landstrom for support during the wind tunnel
tests and discussion of results.
Funding
Energimyndigheten (Swedish Energy Agency) project number
37195-1.
References
1. Hucho WH. Aerodynamics of Road Vehicles. Fourth ed.
Warrendale, USA: Society of Automotive Engineers, Inc,
1998. ISBN 0-7680-0029-7.
2. Tesch G, Demuth R and Adams N. A new approach
to analyzing cooling and interference drag. SAE Int J