Cooling in poor air quality environments -- Impact of fan operation on particle deposition
SSUBMITTED FOR CONSIDERATION
TO IEEE SPECIAL ISSUE IN TRANS. CPMT, FEBRUARY 2021 1
Cooling in poor air quality environments -Impact of fan operation on particle deposition
Jason Stafford and Chen Xu
This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer beaccessible.
Abstract —Environmental pollutants are a source for reliabilityissues across data center and telecommunications equipment. Aprimary driver of this is the transport and deposition of particlematter (PM . , PM ) on printed circuit boards, electroniccomponents and heat exchange surfaces. This process is enhancedby turbulent air flows generated from cooling fans. Particlepollutants can persist after contemporary filtering, highlightingthe importance of elucidating particle transport mechanisms andutilising this information to design robust equipment. This studyinvestigates particle transport behaviour arising from axial fansoperating under varied aerodynamic conditions. Transient, multi-phase numerical simulations were performed to model the flowof millions of microscale particles in air and determine their fate.Across a comprehensive range of fan operation conditions, fromaerodynamic stall to free delivery, non-dimensional depositionvelocities spanned an order of magnitude. Deposition profilesvary from monotonic to non-monotonic behaviour, influenced bylocal flow impingement, blade tip vortices, and shear velocity.A simple flow control solution that mitigates the factors influ-encing deposition has been demonstrated for equipment alreadydeployed. The findings and numerical methods can be appliedfor the optimization of fan-cooled equipment intended for indoorand outdoor environments where air quality is poor, or pollutionlevels are high. Index Terms —Air cooling, fans, particles, fouling, reliability.
I. I
NTRODUCTION T He reliability challenge posed by atmospheric pollutantshas grown significantly since it was highlighted overthirty years ago by Comizolli et al. [1]. Corrosive gases(e.g. SO , H S, NO x ) and particle contaminants acceler-ate electronic equipment failures primarily through materialcorrosion and leakage current effects [2]. Micron and sub-micron scale particles have diverse constituents, deposit ontosmall feature electronic devices, and persist globally. Theyoriginate naturally from wind-driven dispersion of land dust,sea salt, or from human-related activities and industrial emis-sion sources [3]–[5]. These particles can be hygroscopic andcontain soluble salts, providing the ingredients to deliquescein environments with moderate relative humidity (typically 50-65% [2]).The sensitive interplay between temperature, relative humid-ity, particle matter (PM) and the resulting reliability issues hasled to recommendations for allowable operating environmentsof information technology equipment [6]. There are, of course,many instances where controlling these environments is either J. Stafford is with the School of Engineering, University of Birmingham,Birmingham B15 2TT, United Kingdom. (e-mail: [email protected]).C. Xu is with Bell Labs - CTO, Nokia, Murray Hill, NJ 07974 USA.Manuscript prepared February 17, 2021. equipment heat fan { filter { ω T ∞ C ∞ RH exit flow PCBs and components dust fouling on sensitive devices { Fig. 1. A typical arrangement for forced cooling electronic equipment andthe region susceptible to enhanced particle deposition and device failures. not possible (e.g. outdoor equipment deployments) or disad-vantageous on cooling efficiency. Free cooling of data centersusing airside economizers reduces air-conditioning energydemands by bringing in outside air to cool equipment [6].This introduces a reliability risk by exposing the electronicequipment to atmospheric pollutants and recent studies havebegun to investigate reliability issues [7].Identifying and excluding problematic contaminants wouldavoid these reliability concerns. Filtering is the widespreadapproach for removing large particles (e.g. d p > µ m),however, most fine particle matter remain. High-efficiencyclean room filters are generally impractical (e.g. above MERV13 rating), introducing energy penalties in forced-cooledequipment, as air movers must overcome the substantial flowresistance. The particles that do enter the equipment either passthrough or deposit onto surfaces via a number of mechanisms,ranging from gravitational settling to inertia-driven forcedflows [8], [9].Inertia effects can dominate in fan-cooled systems. Pref-erential deposition patterns in electronic equipment have re-cently been studied using in situ particle image velocimetrymeasurements and observations from long-term installationsin the field [9]. These patterns of enhanced particle depositionwere found to be a consequence of the fan exit flow angle andlocal turbulent events that enhance mass transport. Notably,the deposition patterns are non-unique, appearing in other fan-cooled systems and data center environments also (e.g. figure1 from ref. [9] and figure 3 from ref. [7]). These push flowconfigurations, illustrated in Fig. 1, are regularly used forcooling electronic systems. Utilising the swirling turbulent exit a r X i v : . [ phy s i c s . f l u - dyn ] F e b UBMITTED FOR CONSIDERATION
TO IEEE SPECIAL ISSUE IN TRANS. CPMT, FEBRUARY 2021 2 flow can provide gains in local heat dissipation [10].Mass transfer is also enhanced by turbulent air transport,accelerating the formation of particle deposition sites whichbecome a nucleus for corrosion and leakage current problems.Therefore, revealing the fundamental connection between de-position and the flow delivered by air movers is crucial forinforming the design of robust technologies exposed to atmo-spheric pollution. This would also support the advancementof next-generation multi-phase predictive tools that enablepackage and equipment designers to reduce the potential risksto reliability. Indeed, this strategy aligns with the exemplarywork and vision of Kraus and Bar-Cohen [11] and Bar-Cohen[12], who showed on many occasions how fundamental studiesin thermal sciences can form the basis of developing physicaldesign methodologies and solutions that address thermally-induced device failures.Previous experiments have shown that fan exit flows playa key role in mass transport of fine particles [9]. The aimof this study is to investigate the influence of cooling fanoperation point on the deposition of particle pollutants. Fanoperation point is influenced by the airflow resistance of theequipment being cooled, filter installations, filtration efficiencychanges over time, and other external effects such as back-pressure resulting from aisle containment configurations [13].The present investigation has been conducted using a multi-phase numerical method that can be readily incorporated intoexisting computational fluid dynamics codes currently used inthe thermal design of electronics. In addition, a simple flowcontrol method has been presented that reduces the impactof the exit flow driving enhanced particle deposition. Thiswas demonstrated by installing the solution in a fan-cooledtelecommunications equipment.II. M
ETHODS
The transport of air and atmospheric particles due to rotatingfan flows was numerically modelled. This section describesthe fan design, particle properties, and numerical techniquesused to perform these multi-phase simulations.
A. Fan characteristics and particle properties
A seven-blade axial fan with an outer diameter of 120 mmand hub-tip ratio of 0.417 was considered for the numericalinvestigations. The blade profile was chosen to be a NACA6409 aerofoil with 30 mm chord length. The angle of attackat the hub was set at 45 ◦ and the blade twist angle was -10 ◦ ,respectively. Fan characteristics of pressure rise and flow ratewere assessed by placing the axial fan concentrically within acircular duct, providing a gap of 1.2 mm between the wall andthe blade tips. The fan was located in the middle of this 600mm long duct. An illustration of this arrangement is shown inFig. 2. A constant volume flow rate was delivered at the inletto the duct, and the pressure rise across the fan was analysed.Different flow rates spanning maximum to minimum pressurerise were applied to construct the pressure-flow characteristicand determine the recommended operating region.Investigations on the transport of atmospheric pollutantsunder varied operating conditions were also conducted in the fan ω exit flow circular duct5DDinlet outlet M( d p ) C ∞ z Fig. 2. An illustration of the computational domain. d p ( m) M ( d p ) Coarse mode τ + < 1 Fig. 3. Size distribution of the fractional mass concentration of coarse modeatmospheric particles. same configuration discussed above. A particle distributionwas constructed to include the characteristic coarse modeobserved in outdoor environments [14]. Particles in the 2.5-10 µ m range are not completely removed from the environmentsof information technology equipment when using standardMERV 11 air filtering [9]. An indoor mass distribution ofthe coarse mode particles, M ( d p ) , was created considering acut-off at 10 µ m. This resulted in a range of particle diametersshown in Fig. 3. A constant particle density of 1500 kgm − was chosen based on an average of the seasonal anddiurnal measurements of the apparent density of atmosphericparticles [15]. Particles were introduced at the inlet to the ductwith an initial velocity equal to the air flow. The number ofparticles introduced at this boundary condition was prescribedto maintain a constant particle concentration ( C ∞ ) at the faninlet for all operating conditions. B. Numerical techniques
An Eulerian-Lagrangian computational fluid dynamics ap-proach was used to predict the transport of individual atmo-spheric particles in air. Transient, Reynolds-Averaged Navier-Stokes simulations (RANS) were performed using the k − (cid:15) UBMITTED FOR CONSIDERATION
TO IEEE SPECIAL ISSUE IN TRANS. CPMT, FEBRUARY 2021 3 turbulence model to describe the transport of turbulent ki-netic energy (TKE) and the rate of dissipation of TKE.This was implemented within the open source continuummechanics software OpenFOAM, using the PIMPLE algorithmfor pressure-velocity coupling. A no-slip condition was appliedat walls and the Spalding wall function was used for near-walltreatment. This Eulerian approach was coupled to a Lagrangiansolver that described the particle motion according to: m i d v i ( t ) dt = (cid:88) F i (1)where m i is the particle mass and v i the particle velocity. Thesum of forces acting on the particle depends on the particleand carrier fluid densities [16]. In the present work, ρ i (cid:29) ρ air and the dominant forces considered are particle drag ( F D )and gravity ( F g ), respectively. The drag force is computedaccording to the Ergun-Wen-Yu equation: F D = m i ( u ( x i ( t ) , t ) − v i ) τ i (cid:0) . Re . i (cid:1) (2)where u is the air velocity, τ i = ρ i d i / µ is the particlerelaxation time, and Re i = ρ air d i | u ( x i ( t ) , t ) − v i | /µ is theparticle Reynolds number.Individual particles were modelled according to (1) and eachsimulation introduced N p ∼ . The volume fraction of theseparticle pollutants in air was φ p ∼ − and permitted one-way coupling between the fluid and the particle. This meantthe air influenced particle motion, however, the particles hada negligible impact on the air flow properties. Turbulenceeffects on particles were considered using a stochastic dis-persion model that obtained turbulence information from theRANS solution and applied an isotropic assumption. Particlesdeposited on the surface of the duct wall when they reacheda particle-wall distance less than d p / .The motion of the axial fan was resolved directly by creatinga dynamic mesh to rotate the geometry at constant angularspeed, ω . This approach is computationally intensive comparedto compact fan models and MRF methods that are tradi-tionally used for the thermal design of air-cooled electronicand photonic systems [17]. However, it also provided themost accurate representation of the fluid mover and supporteddetailed analyses on the spatio-temporal characteristics of thismulti-phase flow. C. Data reduction and validation
The rate of particle deposition onto a surface can be describedby the deposition velocity: V d = JC ∞ (3)which relates the mass flux per unit time, J , to the concen-tration of particles in the air, C ∞ . The deposition velocitydepends on a number of physical processes. The typicalregimes include Brownian diffusion, turbulent diffusion andeddy impaction, and inertia moderated. These are categorizedthrough the non-dimensional deposition velocity and particlerelaxation time parameters: -2 -1 + -5 -4 -3 -2 -1 V d + Liu & Agarwal (1974)Sippola & Nazaroff (2005)Prediction τ Fig. 4. Predicted relationship between the dimensionless particle depositionvelocity, V + d , and the dimensionless particle relaxation time, τ + .TABLE IN UMERICAL GRID SENSITIVITY ANALYSIS N c × E (∆ P ) E ( u x , u y , u z ) E ( m p,d ) V + d = V d u ∗ (4) τ + = τ p u ∗ ν (5)where u ∗ is the friction velocity, τ p is the particle relaxationtime, and ν is the kinematic viscosity. In the present work,non-dimensional particle relaxation times span the turbulentdiffusion and eddy impaction regime ( . < τ + < , Fig. 3).The numerical method was verified against experimentalparticle deposition data for fully developed turbulent flow inducts [18], [19]. Fig. 4 shows the comparisons with the pre-dictions of the proposed model providing suitable agreementin the turbulent diffusion / eddy impaction regime and for thetransition to the inertia moderated regime ( τ + (cid:29) ).A sensitivity analysis was performed to determine thenumerical grid parameters that ensured sufficient accuracy.Numerical solutions for six different grids were tested with thefan running at an intermediary operating point near maximumefficiency. Comparisons between the predictions for globalquantities (pressure rise), local velocity profiles in three-dimensions, and total mass of particles deposited on theduct surface were used to assess the sensitivity of the gridsettings on the solution accuracy. For these assessments, a non-dimensional time of t ∗ = ωt/ π = 9 was selected across thetransient simulations. These results are presented in Table Iwhere the relative differences, E ( . ) , are calculated with respectto the predictions at the finest grid resolution. Local velocity UBMITTED FOR CONSIDERATION
TO IEEE SPECIAL ISSUE IN TRANS. CPMT, FEBRUARY 2021 4 differences (RMSD) are normalized by the mean flow velocityof 3 m s − . The numerical grid settings for N c = 1 . × cells converged to a maximum difference ≈ and wasselected to balance accuracy with computational time.III. R ESULTS AND D ISCUSSION
The interplay between fan operating characteristics, atmo-spheric particle transport, and particle deposition are discussedin this section. Following this investigation, a simple flow con-trol solution that addresses dominant deposition mechanismsis presented for a telecommunications equipment test case.
A. Airflow characteristics
The predicted characteristic pressure-flow rate and powercurves for the axial fan under investigation are shown in Fig.5. The stall region, defined by the pressure trough in the ∆ P − Q curve (points “i-iii”), is a regular feature in axial fanperformance curves. Here, the flow is unstable and separatesfrom the fan blades. It is a region normally avoided for coolingapplications, with surging flow effects, pressure variations,propensity for increased bearing wear, vibrations and noisegeneration. The region toward the maximum flow rate at zeropressure rise is also typically avoided. At this location nearpoint “viii”, the efficiency reduces considerably as reflected bythe reduction in aerodynamic power, q . The operating regionregularly designed for in fan-cooled systems is between points“iv” and “viii”.The flow resistance of these systems, such as heat ex-changers or printed circuit boards in Fig. 1, follow a systemcharacteristic behaviour also presented in Fig. 5. This curve isa function of the flow regime (e.g. laminar/turbulent) throughthe exponent, n , and various other losses through the systemaccounted for by the prefactor, K . While most thermal designsare arrived at for a fixed system characteristic curve, there arepractical scenarios where this curve can change throughout theequipment’s operational lifetime. For example, filter blockages Q (m s -1 ) P ( P a ) q ( W ) P ~ KQ n P System curve (n = 2) q Multi-phase analysis region i ii iii iv v vi viiviii
Fig. 5. Fan pressure rise and power curves for an operating speed of ω =2000 rpm. can shift the operating point towards the stall region. Similarly,the installation of additional circuit boards to card slots. Over-sizing or under-sizing fans for cooling applications can alsoresult in adverse operating conditions. To capture this rangeof possible operating scenarios, a wide exploration space hasbeen considered for multi-phase flow simulations. This ishighlighted in Fig. 5 and spanned from stall to free deliveryconditions. B. Particle transport and deposition mechanisms
Enhanced particle fouling in forced-cooled systems, illustratedin Fig. 1, predominantly occurs in close proximity to the fanexit where the highest shear velocity and turbulent stressesexist [9]. Therefore, an area ± . z/D ) upstream and down-stream of the fan exit flow was the focus for this study. Fig.6 shows a snapshot of the flow field in this region for aseverely stalled operating condition (point “i”, Fig. 5). Thistwo-dimensional plane intersects a fan blade at mid-chord forpositive r/D . In the lower half of the plot, at negative r/D , theplane is located at the intermediate angle between two adjacentfan blades. This plane was selected as it captures both the airflow during blade passage (+ve r/D ) and in the blade wake(-ve r/D ).Two dominant flow features are observed. The first arethe large vortices which emanate from the blade tip region,severely distorting the inlet flow to the fan. This flow recircu-lation at the inlet also influences the exit flow that continuesdownstream. Air and particles impinge on the duct wall atan off-axial exit angle with a significant radial and tangentialvelocity component. This flow behaviour is promoted by areduction in axial velocity in the stall region, the consequencesof a reduced flow rate at this operating point.The effect of these exit flow features on the depositioncharacteristics have been examined from ≤ z/D ≤ . .Fig. 7 presents the non-dimensional deposition velocities forall operating points investigated across the fan performance -0.5 -0.25 0 0.25 0.5z/D-0.5-0.2500.250.5 r / D -0.6-0.4-0.200.20.40.60.8 u t / ω r b Fig. 6. A snapshot of the three-dimensional air and particulate flows at ω = 2000 rpm and t ∗ = 10 . Particles have been colored green. Airflowis represented by velocity vectors in the radial-axial plane and contours oftangential velocity normalized by blade tip speed. UBMITTED FOR CONSIDERATION
TO IEEE SPECIAL ISSUE IN TRANS. CPMT, FEBRUARY 2021 5 characteristic. These deposition data for the circular ducthave been circumferentially averaged. Distinctive differencesin deposition profiles were found between recommended op-eration and when the fan enters stall conditions. Local non-dimensional deposition velocities across ≤ z/D ≤ . , areup to four times higher during stall operation compared to thelowest deposition case. This case occurred at the maximum fanaerodynamic power, q max . The shape of these profiles alsoshifts from a monotonically decreasing deposition velocitywith increasing z/D , to a non-monotonic behaviour as stall isestablished.These observations have important consequences for thepractical design of fan-cooled electronic systems. Firstly, themass flux of coarse mode particles with d p < µ m hassignificant local variations over a relatively short distancedownstream. In the operating region, the non-dimensionaldeposition velocities span V + d ∼ − − − . This sharpreduction in deposition velocity agrees with the locationsof patterns of preferential deposition observed on equipmentoperating in the field [7], [9]. Furthermore, if operating out-side of the recommended aerodynamic conditions, positioningsensitive components further downstream of the exit flow doesnot necessarily correlate with a reduced exposure to particledeposition.Secondly, a significant change in particle deposition canoccur with small changes in operating point. For example,deposition velocity varies by a factor of 3-4 when moving ± either side of the maximum fan power. In many coolingand ventilation systems, the target fan operation point is within of the maximum efficiency, thus placing it in a regionwhere high variability in particle mass flux occurs. Therefore,physical designs for forced air cooling in poor air qualityindoor and outdoor environments should carefully considerholistically the fan operation, thermal design and pollutionexposure levels to avoid the failure modes associated withparticle deposition (e.g. corrosion, mechanical, thermal).These trends in deposition with operating point becomemore apparent by overlaying particle deposition curves onthe fan performance characteristic. Fig. 8 shows two curvesfor average and maximum deposition velocity, V d , over theregion of interest, ≤ z/D ≤ . . As noted above, thedeposition minima occur at the point of maximum fan power(point “iv” in Fig. 5), with sharp increases either side of this.On the stall side, deposition velocity rapidly increases as theflow becomes unstable. Within the recommended operatingregion, deposition velocity increases at a lower rate, reachinga maximum mid-way between maximum fan power ( q max )and the free delivery condition ( ∆ P ≈ ).Notably, at this operating point, there is a twofold increasein average deposition velocity compared to V d at maximumfan power. The presence of this local maximum within the rec-ommended fan operation region also opens up the opportunityfor multi-objective fan and thermal design which incorporatesmass transport. The weighting of objectives in this designframework (e.g. maximize heat dissipation, maximize fanefficiency, and minimize particle deposition) might dependon the application, operating environment ( T ∞ , RH) and airquality conditions ( C ∞ , PM . , PM ). z/D V d + - iiiiiiiv max vviviiviii P 0 } operateregion } stallregion }} q Fig. 7. Axial profiles of the non-dimensional deposition velocity for eachoperating point and a rotational speed, ω = 2000 rpm. Q (m s -1 ) P ( P a ) V d ( m s - ) - P V d V d,max Fig. 8. Particle deposition behaviour operating along the fan characteristiccurve for a rotational speed, ω = 2000 rpm. The monotonic non-dimensional deposition velocity profilesacross the operating region (points “iv” to “viii”) are V + d ∼ − for z/D (cid:47) . . Further downstream, for z/D (cid:39) . ,this reduces to V + d ∼ − as the turbulent diffusion andeddy impaction deposition mechanisms weaken. In contrast,the unstable flows generated as the fan experiences stall remainat V + d ∼ − over the entire z/D examined. The transition tothese non-monotonic deposition profiles while operating in thestall region can be explained by the accompanying alterationto the air and particulate flow behaviour.Close-up views of the air velocity vectors and particleflow in the blade wake ( − . ≤ r/D ≤ − . ) and blademid-chord planes ( . ≤ r/D ≤ . ) for four sample fan UBMITTED FOR CONSIDERATION
TO IEEE SPECIAL ISSUE IN TRANS. CPMT, FEBRUARY 2021 6 -0.5 -0.25 0 0.25 0.5-0.5-0.25 r / D -0.5 -0.25 0 0.25 0.5-0.5-0.25 r / D -0.5 -0.25 0 0.25 0.5-0.5-0.25 r / D -0.5 -0.25 0 0.25 0.5 z/D -0.5-0.25 r / D r / D r / D r / D -0.5 -0.25 0 0.25 0.5 z/D r / D iiiivvi u p,t / ω r b u p,t / ω r b Fig. 9. The flow of microscale particles in the blade wake and mid-chord regions (left and right) for ω = 2000 rpm and t ∗ = 10 . Operating points correspondwith stall conditions (“i” and “ii”), maximum aerodynamic power (“iv”), and peak deposition velocity in the recommended region of operation (“vi”). Ductwall, fan inlet and fan exit flow planes are located at r/D = ± . , z/D = − . and z/D = 0 , respectively. operating points (“i”,“ii”,“iv” and “vi”) are shown in Fig. 9.Fan rotation introduces three dimensional flow behaviour withan increase in out-of-plane tangential particle speed, | u p,t | ,as the operating point is shifted towards higher pressure rise.Additionally, vortices form around the blade and grow in sizeas pressure rise increases. As discussed previously, in thesevere stall case (“i”) these vortices extend into the inlet sideand alter the inflow velocity distribution. Interestingly, the flowmaldistribution along the blade span produces protective lowconcentration bubbles, with many particles diverted around thevortex. Particles feed into the oblique jet that is formed at thefan exit in the stalled condition, resulting in high momentumparticle impingement at the duct wall for < z/D < . .This leads to a local enhancement of the wall normal massflux, J , reflected in the deposition velocity profile in Fig. 7.As the operating point is shifted to the trough of the stallregion (“ii”), the wake vortex reduces in size while the bladevortex flips to the pressure side of the fan during blade passage.The latter produces a low concentration bubble spanning < z/D < . . This flow feature explains the reduction indeposition velocity for points “ii” and “iii” in Fig. 7, followedby a recovery and peak in V + d downstream as particles impingeonto the duct wall.At maximum aerodynamic power (“iv”), the lowest local,maximum and average deposition velocities were observedin Figs. 7 and 8. The factors contributing to reduced masstransport include: 1) it is the lowest flow rate with stableoperation; 2) the tip vortex and low concentration bubble isretained in the near wall region; and 3) the radial velocity component in the exit flow decreases, reducing flow impinge-ment and lowering the wall normal mass flux. The latter is afeature dependent on pressure rise, with the impingement anglechanging from approximately − ◦ (“i”), − ◦ (“ii”), − ◦ (“iv”) to − ◦ (“vi”) along the fan characteristic. Collectively,these three factors reduced the local particle concentration andshear velocity immediately downstream.The peak deposition velocities in the operating region, V d and V d,max , occurred at point “vi”, mid-way between q max and ∆ P ≈ . Here, the blade tip vortex observed in other casesno longer exists on the blade pressure side, leading to enhancedmass transport to the wall at r/D ≈ . . The vortex persistsin the blade wake at z/D < , however, it is much reducedin scale compared to the previous operating points. The flowexits and impinges at r/D ≈ − . at an off-axial, obliqueangle which promotes mass transfer for < z/D < . aboveother operating points. This oblique exit flow angle reducesas pressure rise across the fan is lowered further, reducing themass transfer enhancements driven by this impinging flow.This results in progressively lower deposition velocities foroperating points “vii” and “viii” at free delivery, offsettingmass transfer enhancement from an increase in flow rate. C. A flow control solution for installed equipment
The multi-phase numerical methods and data analyses pre-sented in this work can be used as a predictive tool fordesigning reliable forced-cooled electronic equipment in poorair quality environments. For equipment already in service,however, the opportunity to modify board layouts, fan designs,
UBMITTED FOR CONSIDERATION
TO IEEE SPECIAL ISSUE IN TRANS. CPMT, FEBRUARY 2021 7 flow control devicefan tray installationequipmentfan tray (removed)tuft locationvanes a b flow controldevice standard
Fig. 10. A simple flow control solution for reducing particle impingement effects. a) A 3D printed flow control device mounted to the left fan housing withina fan tray. b) Fan tray installation in a telecommunications base station equipment. abc housing flow directionflat vanesvanes designedfor exit flow flow inside equipmentPCB surfacefan exit
Fig. 11. The flow control solution (left) alters exit flow impingement within the telecommunications equipment as represented by changes in fibre tuft angle(right). The images demonstrate exit flow for a) standard operation, b) control using flat vanes, and c) control using vanes designed for the exit flow. and other corrective hardware maintenance to solve reliabilityissues is generally not feasible and cost-prohibitive. An alter-native approach has been considered that performs remediationon already-deployed equipment using a basic flow controldevice [20].Qualitatively, the particle transport and deposition mecha-nisms uncovered in this numerical investigation are ubiquitousfor axial fans and recently confirmed in experiments by theauthors using particle image velocimetry [9]. The investi-gations on the impact of operating point in Section III-Bhave demonstrated that fan exit flow angle has a dominantcontribution to particle deposition velocity.Using telecommunications base station equipment as the testcase, shown in Figs. 10 and 11, stationary aerodynamic guide vanes were designed and installed on individual fans withinthe fan tray. These altered the exiting air flow angle so that theoblique flow impingement was reduced, while limiting adverseeffects to the bulk aerodynamic performance of the fan ( < reduction in volumetric flow rate). This concept utilized thespace available in the fan housing and protruded by only 3mm outside of this, allowing for installation of the existingfan tray without any equipment modifications.For the purposes of the study, and to investigate the alter-ation to exit flow angle, the equipment was modified with atransparent face plate to permit flow visualization using fibretufts. These tufts were placed in the exit flow plane at locationsindicated in Fig. 10a and images were captured using a DSLRcamera and halogen light source during fan operation. The UBMITTED FOR CONSIDERATION
TO IEEE SPECIAL ISSUE IN TRANS. CPMT, FEBRUARY 2021 8 vanes designed to have a flow entrance angle matching the exitflow in Fig. 11a, and a flow exit angle aligned parallel to theprinted circuit board, altered the downstream flow angle from − ◦ to ◦ . This concept, therefore, provides a simple, lowcost solution to one of the main drivers for particle deposition.IV. C ONCLUSION
Indoor and outdoor electronic equipment located in poor airquality environments presents reliability challenges. Forcedcooling, although integral in the prevention of thermally-induced failures, indirectly exposes components and devicesto particle pollutants. In this study, multi-phase numericalsimulations on axial fan flows were performed to determine theinfluence of operation point on the deposition of coarse modeatmospheric particles with diameters below 10 µ m. Depositionspanned the turbulent diffusion and eddy impaction regime andthe dominant deposition factors across the fan characteristic,from aerodynamic stall to free delivery, have been revealed.Deposition velocity was highest in the stall region, withparticle momentum and wall normal flux enhanced by theinstabilities and oblique flow impingement near the fan exit.Deposition velocity was lowest at maximum aerodynamicpower, as the presence of blade tip vortices and a maldistribu-tion of flow along the blade span provided a protective bubbleof low particle concentration, limiting the local mass transportof particles to the surface. Within the normal operation region,a local peak in deposition velocity was found, as a trade-offbetween increased flow rate and reduced impingement angleat low pressure rise occurred. Finally, a practical, low costflow control solution was developed and installed into realtelecommunications equipment to alter particle impingementangle and reduce deposition rate.A CKNOWLEDGMENT
J.S. acknowledges use of High Performance Computing fa-cilities at the University of Birmingham (BlueBEAR). J.S.would also like to acknowledge Mr. Jahan Hadidimoud whoprovided the fan CAD geometry for the numerical simulationsperformed in this study.R
EFERENCES[1] R. B. Comizzoli, R. P. Frankenthal, P. C. Milner, and J. D. Sinclair,“Corrosion of electronic materials and devices,”
Science , vol. 234, pp.340–345, 1986.[2] R. P. Frankenthal, D. J. Siconolfi, and J. D. Sinclair, “Accelerated lifetesting of electronic devices by atmospheric particles: Why and how,”
Journal of the Electrochemical Society , vol. 140, no. 11, pp. 3129–3134,1993.[3] K. Xiao, X. Gao, L. Yan, P. Yi, D. Zhang, C. Dong, J. Wu, andX. Li, “Atmospheric corrosion factors of printed circuit boards in a dry-heat desert environment: Salty dust and diurnal temperature difference,”
Chemical Engineering Journal , vol. 336, pp. 92–101, 2018.[4] B. Song, M. H. Azarian, and M. G. Pecht, “Effect of temperature andrelative humidity on the impedance degradation of dust-contaminatedelectronics,”
Journal of the Electrochemical Society , vol. 160, no. 3, pp.C97–C105, 2013.[5] W. Yue, X. Li, J. Liu, Y. Li, X. Yu, B. Deng, T. Wan, G. Zhang,Y. Huang, W. He, W. Hua, L. Shao, W. Li, and S. Yang, “Charac-terization of PM . in the ambient air of Shanghai city by analyzingindividual particles,” Science of the Total Environment , vol. 368, pp.916–925, 2006. [6] J. M. Shah, O. Awe, B. Gebrehiwot, D. Agonafer, P. Singh, N. Kannan,and M. Kaler, “Qualitative study of cumulative corrosion damageof information technology equipment in a data center utilizing air-side economizer operating in recommended and expanded ASHRAEenvelope,”
Journal of Electronic Packaging , vol. 139, no. 020903, pp.1–11, 2017.[7] J. M. Shah, A. Misrak, D. Agonafer, and M. Kaler, “Identification andcharacterization of particulate contaminants found at a data center usingairside economization,”
Journal of Electronic Packaging , vol. 141, no.031003, pp. 1–10, 2019.[8] M. Tencer, “Deposition of aerosol (“hygroscopic dust”) on electronics- mechanism and risk,”
Microelectronics Reliability , vol. 48, no. 4, pp.584–593, 2008.[9] J. Stafford and C. Xu, “Deposition of particle pollution in turbulentforced-air cooling,”
Aerosol Science and Technology , pp. 1–15, DOI:10.1080/02786826.2021.1873237 2021.[10] R. Grimes, M. Davies, J. Punch, T. Dalton, and R. Cole, “Modelingelectronic cooling axial fan flows,”
Journal of Electronic Packaging ,vol. 123, no. 2, pp. 112–119, 2001.[11] A. D. Kraus and A. Bar-Cohen,
Thermal analysis and control ofelectronic equipment . New York: Hemisphere Publishing Corporation,1983.[12] A. Bar-Cohen, “State-of-the-art and trends in the thermal packaging ofelectronic equipment,”
Journal of Electronic Packaging , vol. 114, no. 3,pp. 257–270, 1992.[13] S. Khalili, H. Alissa, K. Nemati, M. Seymour, R. Curtis, D. Moss,and B. Sammakia, “Impact of server thermal design on the coolingefficiency: Chassis design,”
Journal of Electronic Packaging , vol. 141,no. 031004, pp. 1–11, 2019.[14] W. E. Wilson and H. H. Suh, “Fine particles and coarse particles:Concentration relationships relevant to epidemiologic studies,”
Journalof the Air & Waste Management Association , vol. 47, no. 12, pp. 1238–1249, 1997.[15] Z. Liu, B. Hu, D. Ji, Y. Wang, M. Wang, and Y. Wang, “Diurnal andseasonal variation of the PM2.5 apparent particle density in Beijing,China,”
Atmospheric Environment , vol. 120, pp. 328 – 338, 2015.[16] J. G. M. Kuerten, “Point-particle DNS and LES of particle-ladenturbulent flow - a state-of-the-art review,”
Flow Turbulence Combustion ,vol. 97, pp. 689–713, 2016.[17] J. Stafford and F. Fortune, “Investigation of multiple miniature axialfan cooling solutions and thermal modeling approaches,”
Journal ofElectronic Packaging , vol. 136, no. 011008, pp. 1–11, 2014.[18] B. Y. Liu and J. K. Agarwal, “Experimental observation of aerosoldeposition in turbulent flow,”
Journal of Aerosol Science , vol. 5, no. 2,pp. 145 – 155, 1974.[19] M. R. Sippola and W. W. Nazaroff, “Particle deposition in ventilationducts: Connectors, bends and developing turbulent flow,”
Aerosol Sci-ence and Technology , vol. 39, no. 2, pp. 139 – 150, 2005.[20] C. Xu and J. Stafford, “Reducing dust deposition on electronic equip-ment by optimizing cooling air flow patterns,”
IPC APEX EXPO 2019,San Diego, CA, January 29-31 , 2019.PLACEPHOTOHERE
Jason Stafford is a lecturer in the School of Engi-neering at the University of Birmingham. He earnedhis Ph.D. in experimental thermofluids at StokesInstitute, University of Limerick, Ireland. He workedat Bell Labs as a Member of Technical Staff andjoined Imperial College London in 2017 as a MarieSkłodowska-Curie Fellow. His research interests areat the interface between fluid dynamics, heat andmass transport phenomena, and materials science,with application to electronics thermal control andsolution processing of nanomaterials.PLACEPHOTOHERE