International Journal of Heat and Fluid Flow 32 (2011) 11601172

Contents lists available at SciVerse ScienceDirect

International Journal of Heat and Fluid Flow

journal homepage: www.elsevier.com/locate/ijhff

A study on the ow eld and local heat transfer performance due to geometric scaling of centrifugal fans
Jason Stafford , Ed Walsh, Vanessa Egan
Stokes Institute, Mechanical, Aeronautical and Biomedical Engineering Department, University of Limerick, Limerick, Ireland

a r t i c l e

i n f o

a b s t r a c t
Scaled versions of fan designs are often chosen to address thermal management issues in space constrained applications. Using velocity eld and local heat transfer measurement techniques, the thermal performance characteristics of a range of geometrically scaled centrifugal fan designs have been investigated. Complex uid ow structures and surface heat transfer trends due to centrifugal fans were found to be common over a wide range of fan aspect ratios (blade height to fan diameter). The limiting aspect ratio for heat transfer enhancement was 0.3, as larger aspect ratios were shown to result in a reduction in overall thermal performance. Over the range of fans examined, the low prole centrifugal designs produced signicant enhancement in thermal performance when compared to that predicted using classical laminar ow theory. The limiting non-dimensional distance from the fan, where this enhancement is no longer apparent, has also been determined. Using the fundamental information inferred from local velocity eld and heat transfer measurements, selection criteria can be determined for both low and high power practical applications where space restrictions exist. 2011 Elsevier Inc. All rights reserved.

Article history: Received 25 February 2011 Received in revised form 1 September 2011 Accepted 4 September 2011 Available online 28 September 2011 Keywords: Radial ow Centrifugal fan Electronics cooling Miniature scale

1. Introduction The widespread use of centrifugal fans in engineering has resulted in many geometric variations of designs in order to meet application requirements. Such applications range from large scale industrial dryers and air conditioning units, to smaller scale blowers for the purpose of augmenting heat transfer in portable electronics. The requirement of fans in the electronics industry has substantially driven the demand for high performance, low noise, and low cost units that can contribute to maintaining adequate component temperatures within space restricted environments. In addition, the continual increase in density of electronics within devices suggests that future cooling solution designs will be further limited by available space. Therefore, there is a necessity to address the topic of miniaturization within the area of thermal management, to prevent thermal issues from stalling the development of future technologies. This is reected in recent literature examining such areas as phase change materials (Tan and Tso, 2004; Fok et al., 2010), thermo electric coolers (Wilson and Simons, 2005; Garimella et al., 2008), and microheat pipes (Langari and Hashemi, 2000). However, despite the widespread use of fanheat sink combinations in electronics cooling, there is limited information available that fundamentally examines the inuence of geo-

Corresponding author.
E-mail address: jason.stafford@ul.ie (J. Stafford). 0142-727X/$ - see front matter 2011 Elsevier Inc. All rights reserved. doi:10.1016/j.ijheatuidow.2011.09.002

metric scaling on the ow eld and local heat transfer distributions produced by miniature centrifugal fans. At larger scales, the extended use of centrifugal fans for uid movement has resulted in detailed research into the performance attributes of many designs. Wu et al. (2008) investigated the velocity eld at inlet, outlet, and tip leakage planes for a centrifugal design with seven unequally spaced blades that were also staggered at different angles along the blade span from hub to shroud. The authors present this design as an effective way to improve aerodynamic performance and reduce noise. High levels of positive and negative vorticity exist on fan outlet measurement planes indicating counter rotational vortices which were generated by the backward curved airfoil blades in rotation. The majority of the mass ow tended towards the impeller hub, with increased velocity uctuations at the shroud side, aided by small vortices created by leakage ow near the impeller shroud. The inuence of a scroll housing on the non-dimensional fan performance was noted as being insignicant at a certain ow coefcient, however below this point the scroll housing offered an increase in total pressure and efciency, with a decrease in the same observed at the higher ow coefcients. Yen and Liu (2007) used a phase-locked PIV technique to determine the outlet ow eld of a shrouded centrifugal fan design which has dimensions suitable to laptop sized electronic applications. Two planes were considered in detail, and the exit ow from the shroud was shown to exit at an off-angle to the fan housing. This was similarly noted by Egan et al. (2009) in a study of the ow

J. Stafford et al. / International Journal of Heat and Fluid Flow 32 (2011) 11601172

1161

Nomenclature A ar c C D Dh Din fmax FC H Hf h I k NETD Nu DP q00 _ Q r r Re T Tu u, v u0 , v0 U Uex V surface area, m2 fan aspect ratio chord length, m specic heat capacity, J/kg K fan diameter, m hydraulic diameter  4pDH/2(pD + H), m fan inlet diameter, m max. frequency detectible, Hz forward curved distance between plates, m fan prole height, m heat transfer coefcient, W/m2 K current, A thermal conductivity, W/mK noise-equivalent temperature difference Nusselt number static pressure difference, Pa heat ux, W/m2 volumetric ow rate, m3/s radial direction from fan center, m non-dimensional distance from fan blade  ((r (D/2))/ Dh)/(ReDh Pr) Reynolds number surface temperature, K p  02 v  02 =U ex turbulence intensity  1=2 u radial, axial velocity components, m/s uctuating component of velocity, m/s p velocity magnitude  u2 v 2 , m/s mean fan exit velocity, m/s voltage, V Vt x, y, z   blade tip velocity, m/s Cartesian coordinates, m time average

Greek symbols e emissivity l dynamic viscosity, kg/m s x fan rotation q density, kg/m3 r StefanBoltzmann constant, W/m2 K4 p  02 =h rh normalized uctuations in heat trans. coeff :  h fc s time, s / ow coefcient w pressure coefcient Subscripts aw adiabatic wall temperature, K c blade chord conditions as reference c conduction f foil (SS304) fc forced convection gen input max maximum nc natural convection p paint r radiation 1 ambient

entering miniature heat sinks which were positioned adjacent to a shrouded centrifugal fan outlet. It was found that increases of up to 20% in the overall thermal performance of the miniature cooling solutions could be achieved by aligning the fan exit ow with the heat sink channels. This highlights the benet of designing fan and heat sink collectively rather than separately as is commonly considered. The advantage of using a nless heat sink design at this scale over a conventional nned design was also another outcome of this work. Stafford et al. (2009a) showed that the thermal performance of the nless design is under predicted using laminar duct ow theory. It was hypothesized that unsteady ow structures generated by the centrifugal fan were conserved in the nless geometry, thereby promoting heat transfer. The nned design however, suppressed these ow features to the longitudinal direction, forming a closer representation with theory. The authors also presented a prediction tool to determine the cross over in design choice for nned and nless geometries. Previous studies examining the bulk performance of rotating fan designs indicates a degrading effect on aerodynamic performance when fans are geometrically scaled below a critical point. Grimes et al. (2005) initially noted the adverse geometric scaling effect on the performance of an axial fan design. A datum fan design with a 120 mm diameter was geometrically scaled down to 1/3, which indicated a reduction in fan efciency. Quin and Grimes (2008) examined the same designs including a 1/20 scale of the same axial fan design for a range of blade Reynolds numbers from 283 to 39,700 based on chord length and blade velocity at the midspan. Below a Reynolds number of 1980, a viscous scaling effect was observed, where fan performance was adversely affected and could no longer be determined by the non-dimensional ow and pressure coefcients of the datum fan. Neustein (1964) also determined a Reynolds number effect on axial fan performance to occur

below 2000. The resultant inuence of this scaling effect on local heat transfer distributions using a miniature axial fan has recently been documented by Stafford et al. (2010a). In a complimentary study by Stafford et al. (2010b), a larger axial fan with different blade geometry and hub-tip ratio was found to produce similar surface heat transfer distributions. This was attributed to the similarity in motor support layout on the exit ow plane. The miniaturization of centrifugal fan designs also results in a similar scaling effect on fan performance as shown by Walsh et al. (2009a, 2010). In the rst study by Walsh et al. (2009a), the inuence of fan prole scaling for fan diameters of 1530 mm was examined to address the issues associated with implementing miniature fan designs in low prole applications. The fan characteristics of ow rate, pressure rise, and power consumption were experimentally measured while varying the blade prole alone. A low Reynolds number effect was noted at 650 based on chord length and blade tip velocity which resulted in a reduction of ow rate, and a simultaneous increase in power consumption over that predicted using conventional scaling laws (Bleier, 1997). At the miniature scale, these fan scaling laws were found to be valid only for fan aspect ratios between 0.12 and 0.17. In a separate study, Walsh et al. (2010) examined the same fan characteristics and range of fan diameters but in this case varying blade chord length. Similar trends in reduced fan performance were noted at low Reynolds numbers, and the authors applied simple boundary layer theory to determine the main contribution to this scaling effect for miniature fans. In doing so, the authors proposed an alternative empirical based correlation for determining the performance of centrifugal fan designs operating at low Reynolds numbers. Aside from studies relating to fan ow and pressure characteristics, investigations into the acoustic emissions of centrifugal fan designs has also received attention (Wolfram and Carolus, 2010;

1162

J. Stafford et al. / International Journal of Heat and Fluid Flow 32 (2011) 11601172

Walsh et al., 2009b). The acoustic emission of miniature centrifugal fans ranging in diameters of 1532 mm has been investigated by Walsh et al. (2009b) for numerous fan proles down to just 0.5 mm. The application of the work was focused on handheld electronic devices, and as a result measurement procedures were designed to reect this. It was determined that the acoustic measurements at the fan outlet were dominant over that measured above the inlet. A new scaling law was introduced which accounts for rotational speed, diameter, height and aspect ratios between 0.06 and 0.16. Guidelines were also outlined for the design of miniature centrifugal fans to minimize acoustic levels. In summary, a wide range of fan designs have been investigated through velocity eld, fan performance, and acoustic emission analyses. Studies on miniature designs in terms of fan performance characteristics and acoustic emissions are now evident due to the anticipated move towards miniaturization in electronics cooling. Although the inuence of miniaturization on fan performance has now been documented, only preliminary studies exist on the exit ow eld and thermal performance of combined miniature fanheat sink cooling solutions. The nless heat sink concept has been shown to provide thermal performance at a similar level to nned designs of equal exterior dimensions at miniature scales, however further investigation is required to fundamentally understand the reason this unconventional design produces the level of enhancement shown by Egan et al. (2009) and Stafford et al. (2009a). Even at the larger scales, there is an absence of experimental studies examining the inuence of unsteady and non-uniform velocity elds from fan assemblies on local thermal performance. This is despite the primary intended use of fans in electronic systems being the promotion of heat dissipation. Therefore, the present experimental study examines the velocity eld and local heat transfer performance of centrifugal fans that discharge air between two parallel plates, representing a nless heat sink design. A primary aim is to examine the uidic mechanisms that result in local heat transfer enhancement and spatial variation of surface heat transfer coefcient. Six scaled versions of a centrifugal fan design have been used to investigate the inuence of geometric scaling on the velocity eld and thermal performance within a nless heat sink. A range of fan diameters (15 59 mm) and proles (26.5 mm) have been chosen to highlight the local heat transfer performance that can be achieved through geometric scaling, while also maintaining a relatively low prole for implementation into space constrained environments.

observed from the local heat transfer measurements. All fans were manufactured using polycarbonate, and designed to operate without a volute, allowing air to discharge in the radial direction. Fig. 1e provides the geometrical details of the chosen fan design. The non-dimensional ow and pressure coefcients are dened in Eqs. (1), (2). These coefcients are plotted in Fig. 2 for all fan designs in Table 1 operating at blade Reynolds numbers Rec > 1000, where Rec = qVtc/l. The data presented in this section was experimentally measured using a test facility designed in accordance with BS848 (1980). This test facility was developed to accurately measure the fan performance characteristics of ow rate and pressure for miniature designs similar to that examined in the current study. A detailed description of this test facility is provided elsewhere (Grimes et al., 2005).

_ Q

xD2 Hf
DP

qx2 D2

2. Centrifugal fan design and performance For the velocity eld and heat transfer analyses, a number of geometrically scaled centrifugal fans were investigated which have aspect ratios 0.068 6 ar 6 0.433. The geometric specications of these fans are included in Table 1. In Fig. 1, the four different centrifugal fan diameters are presented. All fans consist of a forward curved blade design, however one half of the fans were designed to rotate in a clockwise direction and the remaining designs rotated in an anti-clockwise direction. This was chosen to investigate if the direction of the tangential velocity component of the fan outlet ow had an inuence on the surface heat transfer phenomena
Table 1 Fan specications. Diameter (mm) Rotor speed (rpm) Blade prole height (mm) No. of blades Blade design Aspect ratio, ar 15 50010,000 2 18 FC 0.133 15 50010,000 4 18 FC 0.267

Fig. 2 indicates the issue associated with accurately predicting fan characteristics for a range of aspect ratios using conventional scaling laws. This observation was noted by Walsh et al. (2009a) who presented results for a constant diameter forward curved radial fan with various aspect ratios from 0.01 to 0.63. As aspect ratio was decreased, the maximum pressure coefcient was shown to decrease which is also shown in Fig. 2. Walsh et al. (2009a) determined that the fan scaling laws could only predict the performance of the fan design investigated for the limited range 0.12 6 ar 6 0.17. The 15 mm fan with ar = 0.133, and 24 mm fan with ar = 0.167 are within this range and appear to resemble similar non-dimensional performance attributes with maximum ow and pressure coefcients within 14%. Fig. 2 also highlights the importance of selecting a blade prole within the maximum limit where the inlet chokes the ow and no further benets in ow rate are experienced. The maximum blade prole recommended by Bleier (1997) for forward curved blade designs is Hf,max = 0.6Din (ar = 0.431) which equates to 6.46 mm for the 15 mm fan design. In contrast, Walsh et al. (2009a) found that above an aspect ratio of 0.35 no benet in ow rate was achieved. This is conrmed in the maximum ow rate measurements, as shown in Fig. 3, of the 15 mm fan with ar = 0.431 which provides minimal increase in ow rate over the 15 mm fan with ar = 0.267. Consequently, velocity eld measurements on the 15 mm fan were only considered for the maximum ow rate fan design with Hf = 4 mm (ar = 0.267). Velocity eld measurements were also considered for the other fan diameters with Hf = 4 mm to examine the inuence of decreasing aspect ratio on the ow eld. Heat transfer measurements were compiled for all fan sizes outlined in Table 1. Although an aspect ratio ar = 0.133 reduces the ow rate, the static pressure level is maintained as the ow accelerates during the 90 turn in the uid from an axial to radial direction. An acceleration in ow occurs when the inlet area  to .the  blade passages (pDinHf) is less than the inlet orice area pD2 in 4 . This is reected in Fig. 3 for a range of fan speeds. The linear regime for ow rate

15 50010,000 6.5 18 FC 0.433

24 50010,000 4 18 FC 0.167

32 5008000 4 18 FC 0.125

59 2005000 4 18 FC 0.068

J. Stafford et al. / International Journal of Heat and Fluid Flow 32 (2011) 11601172

1163

Fig. 1. Centrifugal fan designs of (a) 15 mm, (b) 24 mm, (c) 32 mm, and (d) 59 mm diameters with fan prole height of 4 mm. (e) Geometrical details of the selected fan design.

0.1

D 15mm 15mm

ar 0.133

10000 rpm

6 5
a = 0.133
r r

30 25 20 15 10

0.267 10000 rpm 0.433 0.167 0.125 0.068 10000 rpm

0.08

15mm 24mm

a = 0.267 ar = 0.433

(m /s)

4000 rpm 4000 rpm 3000 rpm

0.06

59mm

0.04

Q x 10

3 2 1 0

0.02

P
0 2000 4000 6000 8000

5 0 10000

0.2

0.4

0.6

0.8

Fan speed (RPM)

Fig. 3. Variation in maximum ow rate and static pressure (gray data points) with fan aspect ratio using a 15 mm fan.

Fig. 2. Non-dimensional ow and pressure coefcients.

and power law trend for static pressure also conrms the scaling relationship with fan rotational speed (Bleier, 1997). 3. Experimental details The geometric specications and bulk performance of the centrifugal fans considered for this study have been presented in Section 2. In the following sections, the velocity eld and heat transfer measurement procedures are discussed, including a description of the experimental facilities created for the purpose of assessing the inuence of geometric scaling. 3.1. Velocity eld analysis A velocity eld analysis was undertaken to characterize the radialaxial exit ow eld of the scaled centrifugal fans. Particle Image Velocimetry (PIV) was chosen to complete this analysis as it is a full eld, non-intrusive measurement technique. The experimental arrangement for the velocity eld analysis is provided in Fig. 4. Each centrifugal fan was positioned between two parallel plates, with the top plate containing an orice that acted as the fan inlet, having an equal diameter to the fan. In all cases a clearance gap of 0.5 mm was set between the base of the fan and base plate, and also the top of the fan blade and top plate. An Edmund Industrial Optics translation stage was used to achieve accurate fan positioning relative to the top and base surfaces. This stage allowed incremental movements of 0.01 mm in the vertical direction. The fan was rotated above the inlet to accommodate an infrared camera and provide full optical access for experiments which visualized the base plate heat transfer. Any effects of fan blockage due to the presence of the motor (Maxon 110124 22 mm diameter

12VDC) and the positioning stage were alleviated by extending the 3 mm diameter input shaft such that the motor to fan inlet distance was 45 mm. A TTi dual DC power supply was used to control fan rotational speed that was monitored using an Omega HHT13 optical tachometer. The experimental arrangement was contained within a glass walled enclosure of dimensions 600 mm (L) 300 mm (W) 300 mm (W) 300 mm (H). In this enclosure, tracer particles were introduced using a glycol solution and Rosco 1700 fog machine, and illuminated using a Nd:YAG laser in a single plane of interest. An 11 mega-pixel CCD camera was positioned perpendicular to the laser sheet. Images were recorded at 1 Hz and processed using TSi Insight 3G software providing randomly sampled, uncorrelated velocity eld data. As a result, the convergence of velocity eld statistics was monitored against sample size to ensure sufcient data samples were considered. An ensemble of between 750 and 1000 vector maps was chosen to represent the velocity eld statistics. In the proceeding results section, this ensemble averaged data shall be referred to as time-averaged. All velocity eld data was recorded with both top and base plates at ambient temperature, as the main focus was to examine the effect of geometric scaling on fan outlet ow. 3.2. Heat transfer analysis The local heat transfer performance was quantied using infrared thermography and a heated-thin-foil technique. A schematic of the experimental apparatus for the measurement of local heat transfer coefcients is also presented in Fig. 4, where the heatedthin-foil represents the top and base plates being cooled by the centrifugal fan. A stainless steel 304 grade foil with a measured

P (Pa)

32mm

1164

J. Stafford et al. / International Journal of Heat and Fluid Flow 32 (2011) 11601172

Fig. 4. Experimental schematic for velocity eld and heat transfer measurements.

thickness of 14.3 lm is clamped and tensioned using copper busbars and a tensioning mechanism that prevents deection of the plates. An electric current is passed through the electrically resistive thin-foil resulting in heating of the plate by Joule effect to produce a constant heat ux condition. The air leaving the fan outlet is conned to exit in the radial direction, preventing recirculation of heated air back into the fan inlet. Base and top plates could be heated separately, and only the plate of interest was heated when recording the temperature measurements. Due to the orice in the top plate, it was not possible to achieve a uniform heat ux over the entire top surface when joule heated. Ideally, a fully heated top plate with an inlet orice centrally located on the heatedthin-foil would determine the full eld heat transfer distribution on the top plate. However, initial experiments of this design were unsuitable due to the local variation of the input heat ux q00 gen , over the surface of the heated-thin-foil. Due to the discontinuity of the fan inlet orice, electrical current is diverted around the orice resulting in non-uniform q00 gen . To overcome this, only a portion of the top plate highlighted in Fig. 4 and tangent to the inlet orice was heated. This was sufcient as the fan outlet ow was found to be axisymmetric about the axis of fan rotation. The thermal images of each plate were acquired using a ThermaCam Merlin camera with an InSb detector operating in the 35 lm MWIR spectral range. A 25 mm lens was used giving a eld of view of 22 16 and providing a temperature resolution of 312.5 lm for all cases examined. A calibration of the infrared camera was conducted to ensure accurate temperature measurements (Stafford et al., 2009b). To ensure accuracy in the IR camera calibration during image recording, a single K-type thermocouple remained mounted to the foil in a location void of large gradients in temperature. A K-type thermocouple was used to obtain the ambient air temperature and was positioned 200 mm upstream of the fan inlet. The effect of ambient air circulations and nearby radiation sources was removed by positioning the experimental facility in a large enclosure with a single provision for the infrared camera lens. 60 thermal images were recorded at 1 Hz once the foil reached a quasi-steady state. In this state, time-varying uctuations in temperature were noted due to the unsteady uid ow

interacting with the thin-foil surface. In the time-averaged analysis, the images were averaged to reduce noise and time-varying uctuations in the temperature prole that were a magnitude of 103 of the averaged temperature map. On the camera observation side, the thin-foil is coated with an opaque matt black spray paint to provide an emissivity of 0.96 on the surface. Both foil and paint thicknesses were measured to account for the contribution of tangential conduction in the energy balance of Eq. (3) which has been shown by Stafford et al. (2009b) to produce signicant errors in the forced convection heat transfer coefcient if ignored at this scale. Eq. (3) denes the forced convection heat transfer coefcient

hfc

00 00 00 q00 gen qnc qr qc T T aw

00 00 where q00 gen is the input heat ux, qr is the radiation heat ux, and qc is the contribution of the conductive heat ow in the foil and paint layers, all of which are dened in Eqs. (4)(6). The surface of the thin-foil on the camera observation side also dissipates heat by natural convection, q00 nc , which was measured experimentally.

q00 gen

VI A   4 4 q00 r er T T 1 @2T @2T kf tf kp tp @ x2 @ y 2 ! qf C f t f qp C p t p   @T @s

4 5

q00 c

The time-averaged heat transfer coefcient was solved using a time-averaged temperature prole and neglecting the energy storage term in Eq. (6) which contains the additional effect of heat ow over time interval, os. The root-mean-square of the uctuations in heat transfer coefcient was calculated to determine the effect of turbulence and uid unsteadiness on the surface heat transfer distribution. This was achieved by solving for the instantaneous heat transfer coefcient using the energy balance of Eq. (3) and using the full expression in Eq. (6) over the recording interval.

J. Stafford et al. / International Journal of Heat and Fluid Flow 32 (2011) 11601172

1165

The acquisition of the uctuating heat transfer coefcient is limited due to the time constant of both the foil and paint layers. Consequently, the amplitude of the recorded temperature uctuations is dependent on both the amplitude and frequency of the heat transfer uctuations due to the thermal inertia of the combined foil and paint layers. Eq. (7) has been adapted from Nakamura (2009) to reect this and determine the maximum frequency of the uctuations which can be detected.

4.1. Velocity eld Fig. 5 presents the time-averaged velocity and streamlines in the radialaxial plane between base and top plates when using a 15 mm fan with aspect ratio ar = 0.267. This example has been chosen to present the similar ow features observed for all cases examined. All velocity eld data has been normalized with the mean fan exit velocity Uex. The air ow exits the fan blades producing a high shear ow with a large velocity gradient on the base plate. The high shear ow enters the nless channel at approximately one half of the channel height, and extends in the radial direction to a radial distance which is dependent on Reynolds number. This type of fan exit ow prole results from much of the airow tending towards the back plate of the fan due to the inertia forces that are generated as the fan redirects the axial inlet ow 90 to produce a radial outlet ow. For the Reynolds number range examined using this fan, a primary vortex is evident which is promoted by the high velocity exit ow interacting with the lower velocity uid in the remaining half of the channel. This velocity prole at the fan exit also inuences the downstream ow features within the channel. As air exits the lower section of the fan blade, it expands in the radial and axial directions. This expansion results in an adverse pressure gradient which forces the high velocity ow to separate from the base plate. This separation drives the impingement which occurs on the top plate. The impinging air ow is then deected back towards the base plate resulting in a secondary impingement also highlighted in Fig. 5. A product of the separation and both impingements is a secondary vortex which rotates in the opposite direction to the primary vortex. It is apparent when comparing Fig. 5ac that increasing Reynolds number shifts the point of impingement in the radial direction away from the fan. This shift is due to an increase in pressure in the fan exit ow over the adverse pressure produced by the expanding air ow. Consequently, at ReDh = 1615, the secondary vortex is elongated in the radial direction compared to that observed for the lower Reynolds number examples. A comparison between the instantaneous ow elds shown in Fig. 6a and b indicates the high level of unsteadiness which exists at the outlet of centrifugal fans and within the nless channel. Vortices which are evident in the upper half of the channel intermittently disrupt the high velocity shearing ow along the base plate. This is caused by axial movement of vortices which encompass a level of vorticity that results in acceleration and deceleration of the shear ow upon interaction. This is observed in the instantaneous velocity elds in Fig. 6 as regions of increased and decreased velocity magnitude are apparent in the region r/D < 1. The instantaneous distributions also highlight the detachment due to the adverse pressure gradient produced by the radial expansion. In Figs. 5 and 6, the location of the ow impingements and detachments are indicated by arrows which point to the radial u location where the radial wall shear stress l @ 0. @z The time-averaged turbulence statistics for this example are presented in Fig. 6c. The radialaxial turbulence intensity increases to 50% of the mean fan exit velocity in the region near the fan blades. Two shear layers emerge as a result of the interaction of the high velocity fan exit ow with the surrounding low velocity uid within the channel. The upper shear layer exists from the fan outlet until impingement occurs on the top plate. The lower shear layer is only evident once the high velocity ow is forced to detach from the base plate. Over the range of Reynolds numbers examined for the 15 mm fan, both time-averaged vortices and impingements were evident. Common uidic mechanisms within the channel were also found to exist with geometric scaling. However, one main difference with the exit ow eld of the 15 mm fan is the absence of a secondary vortex within the radialaxial ow eld above a critical Reynolds

fmax

eDhfc T T 1 2pqf C f tf qp C p tp DT NETD

where DTNETD is the noise-equivalent temperature difference of the infrared camera. For the range of experiments considered, the maximum detectible frequency of heat transfer coefcient uctuations was approximately 5 Hz. The Nusselt number is dened in Eq. (8) based on the characteristic dimension Dh. This represents the hydraulic diameter where the ow exiting the fan enters the channel.

NuDh

hfc Dh kair

where kair is the uid thermal conductivity. The relationship between heat transfer and uid dynamics can be examined through the scaling of non-dimensional Nusselt number with Reynolds number. The previously dened Rec is inappropriate for this as it only characterizes fan aerodynamics and is independent of the at plate heat transfer due to the scaling effect that exists at miniature scales. Therefore, Reynolds number was dened as:

ReDh

qU ex Dh l

where Uex is the mean fan exit velocity entering the channel, calculated from ow rate measurements. This provided a range 45 < ReDh < 5700. 3.3. Uncertainty The inuence of measurement uncertainties on the calculated data presented has been accounted for using an uncertainty analysis (Moffat, 1997). Uncertainties in the measurement of pressure and volumetric ow rate were 5% and 2.8% respectively. Uncertainty in the measurement of velocity was determined to be 5.44%. The maximum uncertainty in the heat transfer coefcient, Nusselt and Reynolds numbers were estimated at 10.9%, 11.2% and 3.1% respectively. The optical tachometer used for measuring fan rotor speed has an accuracy related to the resolution limit of 1 rpm. For the fan speeds considered however, uncertainty was noted as approximately 10 rpm, due to variations in speed monitored over the test duration. Experimental uncertainty bands have been neglected when presenting the data for clarity. 4. Results and discussion The experimental results of the local velocity eld at the fan blade exit are presented rst to determine the ow features that are generated by various scaled centrifugal fans with radial discharge. This information is then used to discuss the local and average heat transfer distributions which occur on the surface of the parallel plates that conne the exit ow, and represent the nless design. Finally, non-dimensional radial distributions in heat transfer are presented to examine the relationship between the fan exit ow, and the resultant heat transfer proles.

1166

J. Stafford et al. / International Journal of Heat and Fluid Flow 32 (2011) 11601172

(a)

(b)

(c)

Fig. 5. Time-averaged velocity eld for a 15 mm fan (ar = 0.267) and (a) ReDh = 85, (b) ReDh = 705, and (c) ReDh = 1615.

(a)

(b)

(c)

Fig. 6. Instantaneous velocity eld at time (a) s and (b) s + 1 s. (c) Turbulence intensity for a 15 mm fan (ar = 0.267) and ReDh = 1615.

number. These critical Reynolds numbers were found to be 1350, 1010, and 880 for the 24 mm, 32 mm and 59 mm diameter fans of constant fan prole height. This has been attributed to a change in expansion with increasing fan diameter, as it is the expanding fan outlet ow which governs the adverse pressure gradient. As fan diameter is increased, the ratio of the fan exit ow area (inlet area to channel) to the channel exit ow area approaches unity, and the radial expansion is no longer dominant. Consequently, the adverse pressure gradient produced by the expanding ow is insufcient to result in ow detachment along the base surface. When the detached ow along the base plate is absent, the local region of impingement on the base plate is removed and may result in reduced thermal performance at this local position. However, by

overcoming the adverse pressure gradient, the detachment of the high velocity ow exiting the fan no longer occurs, and velocity gradients at the surface are increased which would lead to heat transfer enhancement. In the following section, the inuence of this secondary vortex on local heat transfer of both plates is discussed. Sample velocity elds for the fan diameters 2459 mm and r/ D < 1.1 are presented in Fig. 7. At ReDh = 2770, the ow eld produced by the 24 mm fan contains only a primary vortex. In contrast, the ow eld due to the 32 mm fan and ReDh = 415 contains both counter rotating vortices as it is below the critical ReDh previously discussed. In Fig. 7c, the location of a single vortex accommodates a much smaller area of the ow eld than the primary vortex for the larger aspect ratio fans of Fig. 7a and b. The velocity prole appears to be directed towards the top plate to some extent for 0.585 6 r/D 6 0.754. It is anticipated that the pressure variation across the curved streamlines of the single vortex results in a net force acting perpendicular to these streamlines and towards the center of curvature (Massey, 2006). This net force causes the high velocity ow to bend towards the top plate, following the curvature of the primary vortex rather than continuing in a solely radial direction along the base plate. In contrast to the larger aspect ratio fans (including that of the 15 mm fan in Fig. 5), the outlet ow of the 59 mm fan is distributed over a larger portion of the blade prole, and therefore the channel. Consequently, the low velocity region within the channel is much smaller than for ar > 0.125. The normalized velocity proles of Fig. 8 suggest that as fan aspect ratio is decreased for a constant plate spacing, the ow within the channel approaches a near parabolic prole much sooner than for aspect ratios ar > 0.125. This is particularly evident when using the 59 mm fan with ar = 0.068. A theoretical velocity prole for r/ D = 1.09 is included in Fig. 8 based on the measured ow rate and the Hagen-Poiseuille prole for fully developed ow between parallel plates. For r/D > 0.754, the experimental velocity prole almost resembles a parabolic shape with a maximum velocity near the center of the channel. All of the velocity proles presented in Fig. 8 have been normalized with the maximum velocity magnitude at r/D = 0.585. For all fan diameters investigated, the velocity eld analyses have illustrated an unsteady ow exiting the fan which contains vortex structures that disrupt the high velocity shearing ow in the entrance region to the channel. This disruption of the boundary

J. Stafford et al. / International Journal of Heat and Fluid Flow 32 (2011) 11601172

1167

(a)

(b)

(c)

Fig. 7. Time-averaged velocity eld for (a) 24 mm fan (ar = 0.167) and ReDh = 2770, (b) 32 mm fan (ar = 0.125) and ReDh = 415, and (c) 59 mm fan (ar = 0.068) and ReDh = 1075.

layer along the plate surfaces can enhance heat transfer with turbulent diffusion. In Sections 4.2 and 4.3, local and radial heat transfer distributions are presented to determine the level of enhancement the unsteady ow within the channel produces, and also the contribution of the previously discussed exit ow features on the spatial and temporal heat transfer performance. 4.2. Local heat transfer The apparatus for the measurement of local heat transfer coefcients using a centrifugal fan has been presented in Section 3.2. This was used to determine if a wide range of fan aspect ratios, with the same geometric design, provided similar heat transfer trends that could be related to the velocity elds illustrated in the previous section. The base and top plate Nusselt number distribution are presented in Fig. 9 for a 15 mm fan (ar = 0.267) and ReDh = 1615. The outline of the fan has been superimposed on all local distributions for the purpose of discussion, with the location of the fan center at (x/D, y/D) = (0, 0). In Fig. 9a, the inuence of the fan outlet ow along the base plate on local heat transfer performance is apparent. In this region, local Nusselt numbers of over 50 are produced from the large velocity gradients at the exit of the fan blades. The area occupied by the secondary vortex is also evident, as well

0.085

0.068

z/D

0.051

r/D 0.585 0.754 0.924 1.09 (1.09)

0.034

0.017

0.2

0.4

0.6

0.8

Normalized U
Fig. 8. Normalized velocity proles for a 59 mm fan (ar = 0.068) and ReDh = 1075.

as an annular region of increased heat transfer produced by the secondary impingement observed in the velocity eld measurements of Fig. 5. The primary impingement and vortex zones provide an increase in local heat transfer on the top plate, as shown in Fig. 9b. The top plate has a lower magnitude of Nusselt number due to the majority of air exiting the lower portion of the fan blades. As previously discussed in Section 3.2, the heated portion of the top plate is that shown in the color contour region of Fig. 9b which is tangent to the fan inlet. The slight dissimilarity in the local Nusselt number between x/D < 0 and x/D > 0 is due to an unheated entrance effect, as thermal boundary layers only begin to develop at the heated foil leading edge. However, this asymmetry is relatively minor, and the local heat transfer distributions are adequately captured. Fig. 10 presents the local Nusselt number on the base plate for the larger fan diameters investigated and corresponding to the velocity elds of Fig. 7. By operating above the critical Reynolds number where ow detachment is avoided, the heat transfer rates on the base surface due to the 24 mm fan decrease gradually from the fan blades in the radial direction. This is also observed for the 59 mm fan in Fig. 10c. In Fig. 10b however, detachment produces a local heat transfer distribution similar to that shown in Fig. 9a for the 15 mm fan. A twofold increase in the local Nusselt number over the secondary vortex region is observed in an annular area where the uid deected from the top plate (Fig. 7b) impinges the base surface. The local heat transfer distributions highlight regions where signicant improvement in heat dissipation can be achieved, but also regions to be avoided if the intended use of the centrifugal fan is to cool discrete heat sources. Therefore, the heat transfer proles on base and top plates can differ substantially when vortices are apparent, from the typical monotonic behavior that is commonly assumed when predicting local thermal performance for ow between parallel plates. The spatial variation of Nusselt number using centrifugal fans has been discussed using Figs. 9 and 10, however the uctuating nature of unsteady fan ows is often overlooked when analyzing heat transfer performance using time-averaged information. Hence, it is also important to determine the inuence of unsteady fan outlet ows on heat transfer performance over time. This may be particularly useful if, like in electronics, reliability of components can be adversely affected by cyclic thermal loading. In the current study, the 15 mm fan with ReDh = 1615 has been chosen

1168

J. Stafford et al. / International Journal of Heat and Fluid Flow 32 (2011) 11601172

(a)

(b)

Fig. 9. Local Nusselt number on the (a) base and (b) top plates for a 15 mm fan (ar = 0.267) and ReDh = 1615. Contour level: 1.5.

0 0.5

Nu
90 80 70 60

0 0.5

Nu
25

0 0.2

Nu
40 35 30

20

y/D

y/D

1 1.5 2 2 1.5 1 0.5 0 0.5 1 1.5 2

y/D

50 40 30 20 10

15

0.4 0.6 0.8

25 20 15 10

1 1.5 1.5

10

0.5

0.5

1.5

1 0.5

x/D

x/D

x/D

0.5

(a)

(b)

(c)

Fig. 10. Local Nusselt number on the base plate for (a) 24 mm fan (ar = 0.167) and ReDh = 2770, (b) 32 mm fan (ar = 0.125) and ReDh = 415, and (c) 59 mm fan (ar = 0.068) and ReDh = 1075. Contour levels: (a) 3, (b), (c) 1.

to convey this information. In Fig. 11, the normalized uctuations in heat transfer coefcient (rh) on the base plate are up to approximately 10% and cover an annular region on the surface where the separated ow and secondary vortex occurs, shown previously in the time-averaged heat transfer measurements of Fig. 9a and velocity eld of Fig. 5c. The results presented in Fig. 11 imply a larger amplitude of normalized uctuations in heat transfer coefcient exist when this secondary vortex is apparent in the ow eld. However, due to the limitations of the experimental technique in resolving the true amplitude of high frequency uctuations (Section 3.2), it is not possible to condently arrive at this conclusion. It is however, possible to conclude that normalized uctuations in heat transfer coefcient with an approximate frequency less than 5 Hz are a maximum when this secondary vortex is in the ow eld. Using this information, this annular region of increased heat transfer uctuations could potentially be avoided if positioning discrete heat sources near centrifugal fan ows. 4.3. Radial heat transfer This section discusses the inuence of fan prole and diameter scaling on thermal performance of centrifugal fans through measurements of radial distributions in heat transfer. Although fan aspect ratio has been shown in Section 2 to be an important parameter in the selection of fan designs from a bulk ow rate and pressure perspective, the inuence of this parameter on heat transfer augmentation has not yet been conrmed. Fig. 12 presents the axisymetric radial distribution of heat transfer coefcient, circumferentially averaged and normalized by the maximum heat transfer coefcient hfc(max) on base and top plates. In Fig. 12a, hfc(max) is the maximum heat transfer coefcient over the entire range of 15 mm fan aspect ratios examined. SimFig. 11. Normalized uctuations in hfc on the base plate for a 15 mm fan (ar = 0.267) and ReDh = 1615. Contour level: 0.005.

ilarly, Fig. 12b is normalized with the equivalent maximum observed on the top plate. A range of fan aspect ratios are shown using a 15 mm radial fan and operating at a constant 10,000 rpm. The base plate data show that similar heat dissipation levels are achieved in the shearing ow region for all aspect ratios. For ar = 0.267, a greater surface area is covered by the shearing ow and secondary impingement zones due to the increased ow rate for this fan prole. At this fan rotational speed, the ar = 0.433 fan has a similar heat transfer distribution however with the absence of an increase in heat transfer due to a secondary impingement. It is anticipated that the secondary vortex does exist in the ow eld, as observed for the other aspect ratios. However, due to the

J. Stafford et al. / International Journal of Heat and Fluid Flow 32 (2011) 11601172

1169

(a)

(b)

Fig. 12. Radial distribution of the normalized heat transfer coefcient on (a) base and (b) top plates for 15 mm fans with aspect ratio 0.133 6 ar 6 0.433.

larger channel spacing combined with no increase in ow rate over the ar = 0.267 fan (Fig. 3), the primary impingement on the top plate is at a much lower velocity. This is conrmed when examining the top plate radial heat transfer distribution in Fig. 12b. The heat transfer performance in this zone using an ar = 0.267 fan is approximately twice that of the ar = 0.433 fan. Consequently, with the deection of air from the top plate to the base plate, the approach velocity for the second impingement is also considerably lower when using a fan with ar = 0.433. On the base plate, the normalized heat transfer coefcient beneath the fan back plate is similar for all proles as expected. In this region, it is anticipated that a Couette ow drives heat transfer, and is independent of fan prole for a constant fan diameter and blade Reynolds number, Rec. A velocity gradient exists between the base plate and the underside of the fan back plate. Although a piezometric pressure difference is absent in this region, this gradient occurs as a result of the moving boundary that is the fan back plate. The ar = 0.133 fan produces the greatest peak in heat transfer at r/D = 0.6, which may be attributed to the increase in acceleration  .  of the uid upon exiting the blade passage pDin Hf < pD2 in 4 , as previously discussed in Section 2. As aspect ratio increases, the reduced acceleration results in a peak of lower magnitude at r/

D = 0.6. Bleier (1997) indicated that the uid can accelerate over the blade tip speed, when exiting at the fan blade pressure side. This is conrmed in the heat transfer measurements, as the peak in maximum heat transfer is outside the blade tip location of r/ D = 0.5. For r/D > 0.6, a decrease in heat transfer is experienced as the high momentum uid exiting the fan detaches from the base plate, due to an adverse pressure gradient, and provides impingement cooling for the top plate at r/D = 0.7 using the ar = 0.133 fan. The fan with ar = 0.267 has this peak in heat transfer at r/ D = 0.9, owing to the increase in ow rate extending the shearing area on the base plate surface. It is postulated that the largest aspect ratio fan ar = 0.433 produces this peak at a lower r/D as the outlet ow is subjected to a greater expansion while maintaining a ow rate with similar magnitude to the lower aspect ratio fan of ar = 0.267 (Fig. 3). Upon impingement, the high momentum uid is then deected back towards the base plate resulting in the creation of a secondary peak in heat transfer for the ar = 0.133 and ar = 0.267 fans over the entire range of Reynolds number investigated. For ar = 0.433 presented in Fig. 12a however, this secondary peak is absent and there is a gradual reduction in heat transfer from the local maximum. The level of heat transfer from the top plate surface in Fig. 12b is also greatly reduced over the ar = 0.133 and ar = 0.267 fans, as the uid predominately exits the fan blade along the base plate. _ for the ar = 0.267 Although the ar = 0.133 fan supplies 57% of Q fan at this rotational speed, the ar = 0.133 fan provides a similar magnitude of radial heat transfer on the top plate as the ar = 0.267 fan. It is estimated that the mean velocity within the channel is similar for both, as %40% reduction in ow rate is coupled with a 40% reduction in exit ow area due to the plate spacing reducing from 5 mm (Hf = 4 mm) to 3 mm (Hf = 2 mm). Therefore fan prole selection can greatly inuence the heat transfer performance of both base and top plates of this design. The radial distribution of Nusselt number is presented in Fig. 13 for fan diameters of 24 mm, 32 mm, and 59 mm and a range of Reynolds numbers, 190 < ReDh < 5700. Non-dimensional heat transfer data is presented using the laminar ow relationship with p Reynolds number, NuDh = ReDh . The theoretical solution for laminar duct ow is also included to provide a comparison with the measured data for centrifugal fans. This is an idealized solution, derived from the theoretical relationship for at plate heat transfer and based on a zero pressure gradient assumption; however it is useful for assessing the thermal performance of the centrifugal fans. The theoretical model assumes laminar ow in the channel entrance region with a constant heat ux boundary condition on p the surface, NuDh 0:453= r Pr1=6 where r is the non-dimensional distance from the fan blade. This model has been selected as previous studies by Sparrow (1955) and more recently by Stafford et al. (2009a) indicate that this approach can adequately represent the entrance region Nusselt numbers for simultaneously developing hydrodynamic and thermal boundary layers within parallel plate and rectangular channels. The Reynolds numbers where the secondary peak in heat transfer occurs are highlighted (blue1) in Fig. 13a and c. This peak shifts outwards in the radial direction as Reynolds number is increased, until a point where the expanding ow no longer causes a ow separation on the base plate. The shift in the peak in Nusselt number on the top plate is also discernible in Fig. 13b and d. This peak becomes less pronounced as Reynolds number increases, as it is the ow detachment from the base plate which provides the primary impingement on the top surface. For all cases examined, the theoretical solution under predicts the thermal performance of the centrif-

1 For interpretation of color in Figs. 1, 411, 13, 14, the reader is referred to the web version of this article.

1170

J. Stafford et al. / International Journal of Heat and Fluid Flow 32 (2011) 11601172

(a)

(b)

(c)

(d)

(e)

(f)

Fig. 13. Nusselt and Reynolds number scaling relationship on base (left) and top (right) plates using (a) and (b) 24 mm (ar = 0.167); (c) and (d) 32 mm (ar = 0.125); and (e) and (f) 59 mm (ar = 0.068) geometrically scaled centrifugal fans.

ugal fans up to r/D % 2 for the base plate, and r/D % 1.5 for the top plate. Above these non-dimensional distances from the fan blade, the heat transfer performance begins to correlate with theory, as the unsteady vortical structures have dissipated into the mean ow. This suggests that centrifugal fans, used for heat dissipation purposes, should be selected based on these limiting radial distances. In the regions near the fan exit, signicant improvements in thermal performance are observed along the base surface aside from the regions where separation occurs. This can be related to the velocity eld measurements presented in Figs. 5 and 7. The outlet velocity above the base surface in this region ranges be-

tween 1.5Uex and 2.5Uex depending on fan aspect ratio. Consequently, the velocity gradients that result from the centrifugal fan outlet ow are greater than that assumed in the prediction, as the heat transfer prediction utilizes the mean outlet velocity Uex. The increased velocity gradients combined with the ow unsteadiness observed in Fig. 6, result in the substantial increase in radial distribution of heat transfer in this region. In Fig. 13a and c, the heat transfer performance in the secondary vortex region (r/D % 1) reduces considerably, and is of similar magnitude to that theoretically predicted. Once separation no longer occurs at the higher Reynolds numbers, this reduction in heat

J. Stafford et al. / International Journal of Heat and Fluid Flow 32 (2011) 11601172

1171

Fig. 14. Collapse of Nusselt and Reynolds number scaling relationship on the base plate using a 24 mm (ar = 0.167) centrifugal fan and a Reynolds number exponent of 0.65. Symbols correspond to that presented in Fig. 13a.

transfer performance is avoided. However, the separation and resultant secondary impingement, also increases the thermal performance locally for 1 < r/D < 1.6. The inuence of the primary vortex and impingement on heat transfer performance on the top surface is shown in Fig. 13b, d and f. The low velocity region of the primary vortex near the inlet orice, as shown in the velocity eld measurements of Fig. 7, results in a lower heat transfer performance compared to the theoretical solution. The thermal performance then increases to a peak which is produced by the primary impingement. This peak eventually attens out with increasing Reynolds number for the reasons discussed previously. Fig. 13 also indicates that the local thermal performance does not scale towards a laminar ow regime for the majority of radial locations. If this was so, the measured data for each Reynolds number would collapse to a single prole, as is the case for the theoretical prole. This is due to the unsteady ow produced by the fan, where vortices interact with the base and top surfaces, as shown in Fig. 6a and b for the 15 mm fan. This unsteady ow increases mixing and results in the thermal performance to scale towards that of a turbulent ow regime. This is shown for the example of the 24 mm fan and base plate heat transfer in Fig. 14. Fig. 14 consists of the data of Fig. 13a, however expressed using a Reynolds number exponent of 0.65. This exponent was chosen as it produces the lowest local standard deviations (<8%) between the six individual Reynolds number proles (250 6 ReDh 6 3485) over the r/D range examined. For a Reynolds number above the critical ReDh where separation no longer occurs, the heat transfer data generally collapses for r/D > 0.5. The main deviation from this relationship in the lower Reynolds numbers is due to the change in the velocity prole when separation occurs, resulting in a secondary vortex and impingement along the base. Through examination of bulk thermal performance of miniature and low prole nless cooling solutions, Stafford et al. (2009a) hypothesized that such designs typically correlate towards that of a turbulent ow regime, despite operating at low Reynolds numbers in many cases. The current study conrms this nding, providing an insight into the uidic mechanisms which promote heat transfer when using low prole centrifugal fans as part of an integrated cooling solution. 5. Conclusions The inuence of geometric scaling on the velocity eld and local heat transfer of centrifugal fans have been highlighted over a wide

range of fan dimensions, with the primary focus on low prole designs. Local velocity eld and heat transfer measurements have been developed to assess the uid mechanisms produced by centrifugal fan exit ows which are fundamental to cooling solution performance and optimization. It was determined that miniature centrifugal fans should be designed with an aspect ratio less than 0.3 to avoid reductions in thermal performance. This has been attributed to a reduction in ow coefcient, as no further increase in ow rate is achieved above this aspect ratio. It is postulated that the common exit ow prole for centrifugal fans occurs due to inertia forces, as the fan directs the axial inlet ow to a radial outlet ow. As aspect ratio decreases and ow rates increase, this effect is reduced and the outlet ow is distributed over the majority of the blade height. The expanding ow from centrifugal fans within a parallel plate channel produced an adverse pressure gradient that resulted in ow separation along the base plate, creating multiple time-averaged vortices and impingements. As fan diameter increased and aspect ratio decreased, this separation was no longer evident above a critical Reynolds number. When this occurs, the spatial and temporal heat transfer distributions are substantially altered, and begin to approach a monotonic decay in local heat transfer with distance from the fan blade tip. Unsteady ow within the parallel plate channel results from the interaction of vortices with the high velocity exit ow. The interruption of the fan exit ow by these unsteady uidic mechanisms disrupts boundary layer growth within the channel, promoting heat transfer above that predicted using laminar ow theory.

References
Bleier, F.P., 1997. Fan Handbook: Selection, Application and Design. McGraw-Hill. BS848: Fans for General Purposes, Part 1: Methods for Testing Performance, 1980. Egan, V., Stafford, J., Walsh, P., Walsh, E., 2009. An experimental study on the performance of miniature heat sinks for forced convection air cooling. J. Heat Transfer. 130 (7), 071402 (19). Fok, S.C., Shen, W., Tan, F.L., 2010. Cooling of portable hand-held electronic devices using phase change materials in nned heat sinks. Int. J. Therm. Sci. 49 (1), 109 117. Garimella, S.V., Fleischer, A.S., Murthy, J.Y., Keshavarzi, A., Prasher, R., Patel, C., Bhavnani, S.H., Venkatasubramanian, R., Mahajan, R., Joshi, Y., Sammakia, B., Myers, B.A., Chorosinski, L., Baelmans, M., Sathyamurthy, P., Raad, P.E., 2008. Thermal challenges in next-generation electronic systems. IEEE Trans. Comp. Packag. Technol. 31 (4), 801815. Grimes, R., Walsh, E.J., Quin, D., Davies, M., 2005. The effect of geometric scaling on aerodynamic performance. AIAA J. 43 (11), 22932298. Langari, A., Hashemi, H., 2000. A system level cooling solution for cellular phone applications. Electron. Cool. 6 (2), 17. Massey, B., 2006. Mechanics of Fluids, eight ed. Taylor & Francis, Inc. Moffat, R.J., 1997. Uncertainty analysis. In: Azar, K. (Ed.), Thermal Measurements in Electronics Cooling. CRC Press, pp. 4580. Nakamura, H., 2009. Frequency response and spatial resolution of a thin foil for heat transfer measurements using infrared thermography. Int. J. Heat Mass Transfer. 52 (2122), 50405045. Neustein, J., 1964. Low Reynolds number experiments in an axial ow turbomachine. J. Eng. Power. 86 (3), 257295. Quin, D., Grimes, R., 2008. The effect of Reynolds number on microaxial ow fan performance. J. Fluids Eng. 130, 101101 (110). Sparrow, E.M., 1955. Analysis of laminar forced convection heat transfer in entrance region of at rectangular ducts. NACA Tech. Note 3331, 142. Stafford, J., Walsh, E., Egan, V., Walsh, P., Muzychka, Y.S., 2009a. A novel approach to low prole heat sink design. J. Heat Transfer. 132 (9), 091401 (18). Stafford, J., Walsh, E., Egan, V., 2009b. Characterizing convective heat transfer using infrared thermography and the heated thin-foil technique. Meas. Sci. Technol. 20, 105401 (111). Stafford, J., Walsh, E., Egan, V., 2010a. Local heat transfer performance and exit ow characteristics of a miniature axial fan. Int. J. Heat Fluid Flow 31 (5), 552560. Stafford, J., Walsh, E., Egan, V., Grimes, R., 2010b. Flat plate heat transfer with impinging axial fan ows. Int. J. Heat Mass Transfer. 53 (2526), 56295638. Tan, F.L., Tso, C.P., 2004. Cooling of mobile electronic devices using phase change materials. Appl. Therm. Eng. 24 (23), 159169. Walsh, P., Egan, V., Grimes, R., Walsh, E., 2009a. Prole scaling of miniature centrifugal fans. Heat Transfer Eng. 30 (1), 130137. Walsh, E., Walsh, P., Punch, J., Grimes, R., 2009b. Acoustic emissions from active cooling solutions for portable devices. IEEE Trans. Comp. Packag. Technol. 32 (4), 776783.

1172

J. Stafford et al. / International Journal of Heat and Fluid Flow 32 (2011) 11601172 Wu, Y.D., Ouyang, H., Tian, J., Du, Z.H., 2008. Particle image velocimetry investigation of the ow eld in a centrifugal impeller with unequally spaced blades. Proc. Inst. Mech. Eng. Part A J. Power Energy 222 (A4), 389401. Yen, S.C., Liu, J.H., 2007. PIV measurements of exit ow eld of centrifugal fans with conditional sampling. J. Mar. Sci. Technol. Taiwan 15 (3), 232240.

Walsh, P.A., Walsh, E.J., Grimes, R., 2010. Viscous scaling phenomena in miniature centrifugal ow cooling fans: theory, experiments and correlation. J. Electron. Packag. 132 (2), 021001 (18). Wilson, J., Simons, R.E., 2005. Advances in high-performance cooling for electronics. Electron. Cool. 11 (4), 122. Wolfram, D., Carolus, T.H., 2010. Experimental and numerical investigation of the unsteady ow eld and tone generation in an isolated centrifugal fan impeller. J. Sound Vib. 329 (21), 43804397.