NASA Technical Memorandum 107515 Army Research LaboratoryAIAA–97–2878 Technical Report ARL–TR–1461

Aerodynamic Synthesis of a CentrifugalImpeller Using CFD and Measurements

L.M. Larosiliere and G.J. SkochU.S. Army Research LaboratoryLewis Research CenterCleveland, Ohio

P.S. PrahstNYMA, Inc.Brook Park, Ohio

Prepared for the33rd Joint Propulsion Conference and Exhibitcosponsored by AIAA, ASME, SAE, and ASEESeattle, Washington, July 6–9, 1997

U.S. ARMY

RESEARCH LABORATORYNational Aeronautics andSpace Administration

https://ntrs.nasa.gov/search.jsp?R=19970025160 2020-04-06T22:49:30+00:00Z

1

AERODYNAMIC SYNTHESIS OF A CENTRIFUGAL IMPELLER USING CFD ANDMEASUREMENTS

L. M. Larosiliere , G. J. Skoch1 1

U.S. Army Research LaboratoryNASA Lewis Research Center

Cleveland, OH 44135and

P. S. Prahst1

NYMA Inc.Brook Park, OH 44142

ABSTRACTThe performance and flow structure in an

unshrouded impeller of approximately 4:1 pressureratio is synthesized on the basis of a detailed analysisof 3D viscous CFD results and aerodynamicmeasurements. A good data match was obtainedbetween CFD and measurements using laseranemometry and pneumatic probes. This solidifiedthe role of the CFD model as a reliable representationof the impeller internal flow structure and integratedperformance. Results are presented showing the lossproduction and secondary flow structure in theimpeller. The results indicate that while the overallimpeller efficiency is high, the impeller shroud staticpressure recovery potential is underdeveloped leadingto a performance degradation in the downstreamdiffusing element. Thus, a case is made for a follow-on impeller parametric design study to improve theflow quality. A strategy for aerodynamic performanceenhancement is outlined and an estimate of the gainin overall impeller efficiency that might be realizedthrough improvements to the relative diffusion processis provided.

INTRODUCTIONINTRODUCTIONSignificant progress has been made in

understanding impeller aerodynamic performance andalso in predicting certain local flow details. A struggleis now ensuing to dislodge the last remaining deficitsin performance for machines of low to moderatepressure ratios. Developers who place a premium onoptimum performance are pursuing a synergisticapproach based on a rational deployment of advancedaerodynamic, structural, and manufacturing methods.However, the question of what is the most effectivestrategy for improving both range and efficiency isstill very much unresolved.

The most popular guide to impeller design isa diffusion parameter of some sort. Dean [1] discussedthe influence of internal diffusion on impellerefficiency. His results, from calculations based upontwo actual stages of medium and high pressure ratio,showed a trend of increasing efficiency with anincreased overall diffusion ratio. Overall diffusionratio is defined as the ratio of impeller inlet relativevelocity, usually taken at the shroud, to impellerdischarge relative velocity (W / W ). It was1 2

postulated by Dean that if an average overall diffusionratio of 2.0 could be realized in the impeller, asignificant increase in efficiency over conventionaldesigns would follow. Kano et al. [2] presented resultsshowing that in addition to the overall diffusion ratio,the rate of diffusion and maximum loading (i.e., 2Dloading diagram) can significantly impact impellerpeak efficiency and range. Kano’s conclusions werebased on boundary layer arguments supported byperformance measurements on three machines ofdifferent design-intent loading distributions.

Moore et al. [3] used a three-dimensionalviscous CFD method to examine the flow in amedium pressure ratio impeller. The CFD results,although not directly compared with measurements,showed several aspects of loss production in theimpeller. Loss production was high over most of theshroud particularly within the clearance flow region.As expected from the impeller geometry, the internaldiffusion process is likely to be very inefficient. Inmost measurements of impeller efficiency, theinefficiency of the internal diffusion process is hiddenby the large centrifugal pressure rise. This nearlyisentropic pressure rise is bought at the unavoidableexpense of a high absolute exit kinetic energy; as aresult the efficiency of the downstream process isgreatly compromised. Vavra [4] offered an interestingcommentary on the impeller internal diffusion process

NASA TM-107515

2

and he subsequently introduced a so called "wheelefficiency" to assess the quality and effectiveness ofthis process. In the calculations of Moore et al., awheel efficiency of 60 percent was computed althoughthe impeller polytropic efficiency was calculated to be91 percent.

Currently, gaps in knowledge concerningimpeller loss sources and magnitudes remain. Forexample, there is no definitive resolution of whetherthe existence and location of large regions ofthroughflow velocity deficit adversely impact lossgeneration within the impeller. Detailed measurementsof the internal flow made by Krain [5], Hathaway etal. [6], and more recently Skoch et al. [7] are helpingto fill some gaps. Moreover, application of CFDmoored to these benchmark data sets can greatlyincrease the information content and also enhance ourability to make design choices. Hirsch et al. [8]calibrated their CFD method using Krain’s data andperformed numerical simulations guided by theoreticalnotions concerning secondary flow to assess thedifferent contributions to secondary flows and theireffect on the overall flow structure.

The intent of this paper is to synthesize theperformance and flow structure in a moderate pressureratio unshrouded centrifugal impeller through anapplication of Computational Fluid Dynamics (CFD)anchored to the measurements of Skoch et al. (Ref. 7).This synthesis is executed with an awareness of theprevailing impeller theoretical process models forinternal diffusion, jet-wake flow, and secondary flowtransport. Thus, a rational framework for a follow-onimpeller parametric design study is established.

This paper is organized as follows. First, adescription of the impeller design-intent and theexperimental setup for the measurements is provided.Next, results of a data match between measurementsand CFD are presented. Finally a discussion on thepossibility of performance improvement is offered.

IMPELLER DEFINITION ANDEXPERIMENTAL SETUP

The impeller was designed to produce a stagepressure ratio of 4:1 at a corrected mass flow of 4.54kg/s (10 lbm/s) when coupled with a vane-islanddiffuser. A quasi-3D flow analysis developed in theearly seventies was used to derive the flowpath anddesign-intent axisymmetric flow. The dimensionlessspecific speed is 0.60 with an impeller corrected tipspeed of 492 m/s (1615 ft/s). At the aerodynamicdesign point, the intent was to keep the impellerloading roughly constant along the flowpath whiledoing most of the internal diffusion over the first 30-50% of the impeller meridional chord. The overalldiffusion ratio along the shroud surface was set at

about 1.4 with the goal of achieving an 83.3% total-to-static efficiency for the stage (i.e., impeller withvane-island diffuser and 90 degree bend) at a pointwith 8% minimum surge margin. Note that only theconfiguration consisting of the impeller discharginginto a vaneless diffuser is of concern in this paper.Details of the aerodynamic and mechanical designincluding blade coordinates are given by McKain andHolbrook [9].

The impeller consists of 15 full blades and15 splitter blades with 50 degrees of backsweep fromradial. Splitter blade leading edges are located at 30percent of full-blade chord and offset slightly towardthe full-blade suction surface in order to produce aneven flow split. The impeller surfaces are composedof straight-line elements from hub to shroud. Ameridional cross-section of the flowpath, a view of theimpeller, and some relevant geometric parameters areshown in Fig. 1. The exit diameter is 431 mm (16.986in), and the impeller exit shroud clearance is 0.203mm (0.008 inch).

The impeller was configured with a vanelessdiffuser in a test-rig for overall performanceevaluation and local flow diagnostics. Overallperformance was derived from total pressure andtemperature rakes located at a radius ratio of 1.18(Fig. 1). Total pressure was measured using six, four-element, total pressure rakes which were evenlyspaced about the circumference of the vanelessdiffuser. Four, three-element, total temperature rakeswere located at the same radius ratio and were alsospaced evenly about the circumference of the vanelessdiffuser. Rake data were area averaged to determineoverall pressure ratio and efficiency. The mass flowrate was determined using an orifice plate.

A single-component laser Doppleranemometer operating in the backscatter mode withoutfrequency shifting was used to measure the velocityfield within the impeller and vaneless diffuser. A fulldescription of the anemometer, seeding system, anddata reduction technique is given by Skoch et al. (Ref.7). The uncertainty in the measured velocities rangedfrom less than 2 percent away from solid surfaces to30 percent or more near the shroud and impellersurfaces. Additional local diagnostics were acquiredusing pneumatic probes. Static pressures weremeasured at several circumferential positions alongthe shroud from impeller leading edge to exit. Theimpeller discharge total pressure profile was measuredwith a constant blockage probe located at a radiusratio of 1.1.

NASA TM-107515

3

DATA MATCH BETWEEN CFD ANDMEASUREMENTS

Computational MethodThe computational modeling of the impeller

thermofluid-dynamic process was executed using theADPAC computer program. Briefly, the ADPACnumerical methodology utilizes a finite volume,multigrid-based Runge-Kutta (four stages) time-marching algorithm to solve a time-dependent form ofthe 3-D Reynolds-averaged Navier-Stokes equations.Residual smoothing is applied after each stage toextend the stability domain of the algorithm.Turbulence closure is obtained by an adaptation of theBaldwin-Lomax mixing length model. Convectivefluxes are handled using a second-order centeredscheme stabilized with scalar artificial dissipation. Thecode employs a multiple-blocked structured meshdiscretization which provides extreme flexibility foranalyzing complex geometries. Further details aboutADPAC are described by Hall et al. [10].

A five-block mesh was created using asimple algebraic grid generation technique. The firstblock represents part of the impeller passageextending from the full blade suction surface to thesplitter pressure surface including the impellerentrance duct; the second block covers the remainingpart of the impeller passage and entrance duct. Blockthree is the vaneless diffuser and extends from theimpeller trailing edge to a radius ratio of 1.5. Thesethree blocks have a circumferentially periodic H-Hmesh structure. The fourth and fifth blocks have a C-H mesh structure and occupy the space in the tip gapover the full and splitter blades respectively. Themesh consists of 161x49x33, 161x49x33, and77x49x113 points, in the throughflow, spanwise, andcircumferential directions, for blocks 1, 2, and 3respectively. Block 4 has 241x9x13 points, with 9 H-lines over the full blade gap height and 13 C-linesacross the blade profile. Similarly, over the splitterblade gap height, block 5 has 161x9x13 points. Thus,the total number of mesh points is 994,057. Parts ofthe mesh are shown in Figure 2, including views ofthe tip clearance grid close to the splitter leading edgeand the blunt trailing edges. The mesh spacings werecontrolled near solid surfaces to provide as muchresolution as possible without overly disrupting thegrid quality. ADPAC automatically switches to a wall-function approximation for the wall shear stress wheninadequate resolution exists.

The computational clearance gap paralleledthat measured, which varied from the impeller inlet toexit. The measured running clearance distribution was0.1524 mm (0.006 inch) near the leading edge, 0.61mm (0.024 inch) near mid chord, and 0.203 mm(0.008 inch) near the trailing edge. In order to avoid

backflow at the outlet boundary of the computationaldomain, the outlet portion of the vaneless diffuserwas contracted. At the inlet, the measured totalpressure and temperature profiles along with zeroswirl angle were specified. A constant static pressureboundary condition was prescribed at the exit of thecomputational domain.

Overall PerformanceThe overall performance from inlet to a

radius ratio of 1.18 at the design speed (21789 rpm)is shown in Figure 3. Both pressure ratio andadiabatic efficiency are adequately predicted at thenear-design point flow rate of 4.57 kg/s (10.06 lbm/s)and also for higher flow rates. However, thecomparison is not as good at the flow rate less thandesign. No attempt was made to predict the completecharacteristic including the stalling flow since the goalwas to closely match the performance near the designflow rate. Near the design point, the CFD predictedflow rate is 4.70 kg/s (10.35 lbm/s) with a pressureratio of 4.16 and an adiabatic efficiency of 87.7percent. The predicted efficiency is higher near thedesign flow by about 1% and tends to be higher at thelower flow because of the higher predicted pressureratio. A comparison of the measured and CFDpredicted total temperature rise showed very closeagreement. Thus, the higher pressure ratio is due tolower predicted losses rather than higher work input.

Local DiagnosticsThe computed and measured

circumferentially-averaged static pressure distributionsalong the shroud are presented in Figure 4 for thenear design point operating condition. Also shown, isthe isentropic static pressure ratio due to centrifugalstatic enthalpy rise along the shroud. This iscalculated by defining an intermediate state (U) suchthat

Where U is the wheel speed and h is the staticenthalpy. For this intermediate state (U), an isentropicstatic pressure ratio is obtained from,

The measurements represent the time-mean or steadypressure distribution along the shroud while thecomputations correspond to a simple area-average ofthe CFD results. Close agreement between CFD andmeasurements is seen. However, near the impeller

NASA TM-107515

4

trailing edge, the isentropic centrifugal static pressurerise is much higher than either that from CFD ormeasurements. This point will be addressed in a laterdiscussion.

The measured and computed spanwisedistributions of circumferentially-averaged totalpressure at a radius ratio of 1.1 are shown in Figure5 for the near design point operating condition. Alsoincluded for comparison, is the computed totalpressure distribution at a radius ratio of 1.18. A goodmatch is observed between CFD and measurements.Most of the discrepancies are near the shroudsuggesting perhaps less mixing in the CFD model ofthe clearance flow than is implied by themeasurements.

Figure 6 shows a comparison between thequasi-throughflow velocity distribution derived fromCFD and that measured with the laser anemometer onthree cross-flow planes (see Fig. 1) for the neardesign-point flow rate. The quasi-throughflow velocitydistribution was extracted from the velocity normal tothe spanwise grid lines employed for the CFD model.This velocity is normalized with the impeller tipspeed. The measurements were converted from theirraw form to a format similar to the CFD results. Itshould be noted that the laser probe has a restrictedrange of spatial coverage and is unable to survey theentire span or resolve the fine details near solidsurfaces. The quasi-throughflow velocity derived fromthe measurements represents data collected over theentire impeller circumference and then ensembleaveraged to yield the velocity distribution in a singleimpeller passage. Nevertheless, the intent here is toascertain whether or not the gross features of theimpeller internal flow structure are captured by theCFD model.

The CFD results of quasi-throughflowvelocity are presented for two different clearance gapdistributions: a constant tip gap of 0.203 mm (0.008inch), and the measured distribution previously given.As seen from Figure 6, the CFD results are in goodagreement with the measurements for the first twocross-sections presented. Near the splitter leading edgeat 30% chord, a small region of relatively lowerthroughflow velocity is observed on either side of thesplittered passage along the shroud. This is due toscraping of the leakage flow by the splitter leadingedge. At 52% chord, a distinctive low throughflowregion situated near the shroud of impeller passage 1is evident. The pitchwise location of the center of thislow throughflow region is clearly affected by theclearance gap as can be observed from the CFDresults. Proceeding to 96% chord, the CFD results,although acceptable in the large, differ from themeasurements in terms of fine details. These

differences may be due to deficiencies in theturbulence model or possibly numerical discretizationerrors. However, it is also possible that themeasurement uncertainties at this location are higherthan those of the CFD model. These issues will beclarified in the near future using more refinedmeasurement techniques and a higher fidelity CFDmodel.

The computed flow structure is very differentwithin the two sides of the splittered passage at 96%chord. In addition, a high throughflow region isobserved near both the suction and pressure surfacesof the leading side of the splittered passage (i.e.,passage 1). Although the classical jet-wake flowstructure is not evident, a structure dominated by theappearance of two large pools of low throughflowvelocity fluid is clearly observed. Most of theessential flow features are deemed adequatelyrepresented by the CFD model. Also, the present CFDresults using ADPAC are similar to those presentedby Skoch et al. (Ref. 7) using a commercial CFDcode.

DISCUSSIONHaving instituted a reasonable data match

between CFD and measurements, the question as tothe possibility for performance improvements is veryappropriate. Herein, this question is tackled by usingthe CFD model to explore the evolution ofirreversibilities and secondary flows within theimpeller. Only the CFD results using the actual shroudclearance distribution are interrogated. In addition, themeasurements are used to extract the overallperformance from inlet to the impeller trailing edge(i.e., separating impeller performance from measuredoverall performance) in terms of total pressure ratio,adiabatic efficiency, and wheel efficiency. Thisinformation is synthesized to establish the possibilityfor performance improvements by flow control.

Irreversibilities and Flow StructureThe principal losses in an unshrouded

impeller flow process are due to friction and mixinglinked to the dissipation of relative kinetic energy,shear work at the shroud, and clearance flow. Figure7 shows the development of the entropy field (s =[1/( -1)]ln(p ) ) within the impeller and vaneless-

diffuser discharge as derived from the CFD results atthe near design flow operating condition. Close to theimpeller leading edge, at 10% chord, the high entropyregion is small and confined to the solid surfaces.Near the splitter leading edge, at 30% chord, a highentropy region is beginning to accumulate along theshroud. Also evident is the almost isentropic hubendwall. Proceeding downstream to 52% chord,

NASA TM-107515

5

further accumulation of two high entropy cores can beobserved near the shroud. Note that the highestentropy region is situated near the juncture of theshroud and the full blade pressure surface.Subsequently, approaching the trailing edge at 70%,84%, and 96% chord, the high entropy regions nearthe shroud exhibit a rapid diffusion toward the centerof the passages. Referring to Figure 6, it can beobserved that the high entropy cores correspond topools of low throughflow velocities within theimpeller.

At the impeller discharge for a radius ratioof 1.01, high entropy regions are observed near thesplitter and full blade trailing edges. The thick trailingedges contribute to a dump loss. Also noted is therapid mixing between high and low entropy regionswhen moving downstream to higher radius ratios. Thisis further illustrated by the development of thecomputed mass-averaged entropy change (s/R )G

within the vaneless space presented in Figure 8. Avery rapid rate of entropy rise is observed from theimpeller discharge to a radius ratio of about 1.04which is consistent with a measure of the streamwiseimpeller wake decay reported by Skoch et al. (Ref. 7).Thereafter, a much milder rate of entropy rise is seen.Beyond a radius ratio of 1.18, the entropy field isnearly uniform.

The entropy distributions shown in Figure 7follow closely the secondary flow transport within theimpeller. It has been established by many investigators(see for example Ref. 6) that the main mechanism forthe accumulation of low momentum fluid within theimpeller is the spanwise transport of boundary layerfluid along the passage surfaces. The ultimate locationof pools of low momentum fluid results from abalance between secondary flows induced bystreamwise vorticity, corner vortices, and theclearance gap. An expression can be derived (seeZangeneh et al. [11]) from classical secondary flowtheory to describe the generation of impellersecondary flows. This expression is:

where W \ |W| . represents the local streamwiserel

component of relative vorticity (e.g., relative helicity)and is the rotational velocity. According to thisequation, secondary flows are generated when thereexists a component of acceleration due to eitherstreamline curvature (W.∇ W) or Coriolis force(2 xW) in the direction of relative vorticity ( ).rel

The first term is responsible for the passage vorticesdue to flow turning in either meridional or blade-to-blade planes, while the second term is due to Coriolisacceleration. Flow turning and streamline curvature inthe blade-to-blade plane generate secondary flows due

to vorticity in the endwall boundary layers. Meridionalcurvature induces secondary flows due to vorticity inblade surface boundary layers. The contribution fromCoriolis acceleration is effective if an axial boundarylayer gradient exists as is usually the case in the radialportion of the impeller. Other vortices having a localinfluence on the flow, such as the horseshoe, corner,and clearance vortices, are not described by the aboveexpression.

Secondary flow distributions were obtainedfrom the CFD results by first extracting a primaryflow defined along the local direction of thestreamwise oriented mesh lines and then calculating avector having components normal to this primary flowon several cross-flow planes. This is displayed inFigure 9 for the near design flow operating condition.Note that every other point has been removed forclarity. Also shown is the normalized relative helicitydistribution which gives a direct measure ofstreamwise vorticity. Near the impeller inlet, at 10%chord, there is some indication of

and the development ofa small scraping vortex at the shroud-pressure surfacecorner. Proceeding downstream to 30% chord, nearthe splitter leading edge, a large clockwise vortexgenerated by the meridional curvature can be observedalong the pressure surface of the full blade. Inaddition, a small leakage vortex interacting with thispressure surface vortex is noticed near the splittersuction surface similar to observations made byHathaway et al. (Ref. 6). In the suction surface part(passage 1) of the splittered passage, details of thesecondary flow structure are obscured by incidenceloading effects near the splitter leading edge. At 52%chord, strong blade vortices along both suction(counterclockwise vortex or negative helicity) andpressure (clockwise vortex or positive helicity)surfaces can be seen. There is a nearly symmetricpattern in impeller passage 1 (i.e., near full bladesuction surface) while in passage 2, the pressure sideof the blade surface vortex is reinforced by a growingshroud-side passage vortex (due to blade loading).The helicity chart indicates that the leakage flows(negative helicity) and the spanwise flows along thepressure surfaces (positive helicity) of the twopassages collide near the blade tip. This may explainthe existence of high entropy regions near the shroud-pressure side. Continuing to 70% and 84% chord,further development of the passage vortex and itsinteraction with the blade surface vortices and theleakage flow near the splitter suction surface can beobserved. In addition, between 84% and 96% chord,the shroud passage vortex, mainly contributed by theblade loading and augmented by the Coriolis vortex,

NASA TM-107515

6

is dominating. The leakage vortex can be observednear the shroud-suction side corner of the full blade.

From Figures 7, 8, and 9, an understandingof the generation and accumulation of low energyfluid within the impeller can be gained. The picturethat develops is one in which energy dissipationwithin blade surface boundary layers and shear workalong the shroud generate low relative kinetic energyfluid. This fluid is transported by the prevailingsecondary and leakage flows which results in thedevelopment of a pool of low relative kinetic energyfluid at the impeller exit. At the impeller discharge,the dump loss from the thick trailing edges along withthis pool of low relative kinetic energy fluid begin torapidly mix under the actions of turbulent viscousstresses and the residual secondary flows generatedwithin the impeller. Note that the role of unsteadyfluctuations (e.g., vortex shedding) in this mixingprocess is unclear and unaccounted for in the CFDmodel. Additional energy dissipation occurs due tothis mixing and frictional forces along the stationaryendwalls of the vaneless diffuser.

Blade Loading and Impeller Static Pressure RecoveryViscous dissipation in shear layers is

proportional to the wetted area and the cube of thelocal “free-stream” velocity. The free-stream velocityis related to the local surface static pressure or bladeloading. Figure 10 presents the loading distributionsderived from the CFD model at hub, mean, and tip.The static pressures are normalized with the inlet totalpressure. At the hub surface, the loading is nearlyzero over the first 30% of chord. From 30% chord tothe trailing edge, a gradual increase in loading can beseen in both parts of the splittered passage. Note thatthe loading distribution is similar in both parts of thesplittered passage except differences close to thesplitter leading edge. The loading diagram at mid-spanshows an almost uniform loading along the chordexcept for large variations locally near the splitterleading edge. This is consistent with the design intentfor this impeller. For the tip section, a nearly uniformloading distribution is also seen. Aft of the splitterleading edge, a noticeable difference is observed inthe loadings of the two sides of the splittered passage.This difference is due to the leakage flow.

Referring to Figure 4, the static pressure risedue to the centrifugal acceleration, assumed to bereversible, is higher at the impeller trailing edge thanthe circumferentially-averaged (area averaged) staticpressure obtained from either CFD or measurements.Assuming negligible impact of unsteady staticpressure fluctuations in the relative frame, Figure 4implies that inadequate (i.e., less than what is requiredto counterbalance losses along the shroud) relative

diffusion is achieved along the impeller shroud.Hence, the static pressure recovery potential of thisimpeller appears to be underdeveloped. Currently,most impeller design systems (see Japiske and Baines[12] for example) are structured similar to the wellknown jet-wake flow model first proposed by Dean[13] but have been further developed and extendedwith proprietary correlations derived from test data.This model assumes the flow to be partitioned intotwo zones at the impeller trailing edge: an isentropiccore or jet and a viscous wake. Impeller performanceis determined by a diffuser-like correlation definingthe impeller exit static pressure recovery as a functionof an effective measure of overall diffusion ratiosimilar to what has been reported by Schumann et al.[14]. However, this type of correlation does notaccount for the diffusion rate which is known to alsoplay a critical role in establishing the peak pressurerecovery. The isentropic assumption, the staticpressure recovery relationship, and a slip factor rulecompletely define the impeller exit jet aerodynamicstate. Ad-hoc modifications are made to account forthe presence of splitters. The wake is often assumedto have the same exit flow angle as the impeller exitmetal angle. This assumption along with the area andlosses allow a definition of the impeller exit wakeaerodynamic conditions. A mixing model for jet andwake is then used to arrive at the impeller exit mixedout aerodynamic state.

As shown in Figure 4, there is a substantialstatic enthalpy rise due to the centrifugal acceleration.This static enthalpy rise can be considered to occurreversibly. Thus, it is appropriate when consideringthe efficiency of the impeller to remove thecentrifugal enthalpy rise from consideration bydefining an intermediate state (U) and a wheelefficiency such that

where h is the isentropic static enthalpy rise at the2,is

impeller trailing edge. The wheel efficiency thusmeasures the quality and effectiveness of the relativediffusion (e.g., h - h ~ 0.5 (w - w ) ) process2 U 1 2

2 2

within the impeller. For the design speed, themeasured shroud static pressure at the impeller exitand the total temperature measured at a radius ratio of1.18 were used to estimate the impeller performanceat several corrected flows from choke to stall. Thiswas done using conservation of mass and energyassuming no aerodynamic blockage at the impellerexit. Thus, the impeller performance from inlet totrailing edge in terms of total pressure ratio, adiabaticefficiency (total-total), and wheel efficiency are shown

NASA TM-107515

7

in Figure 11. The scatter shown in Figure 11 is due tovariations in the rig inlet total pressure used fordetermining performance sensitivity to Reynoldsnumber changes. As originally noted by Vavra (Ref.4), negative wheel efficiencies are caused by very lowstatic pressure rise and do not imply negative entropyproduction. A peak impeller adiabatic efficiency ofnearly 94% is estimated, whereas the peak wheelefficiency is about 20%. These estimates werecorroborated by the CFD model which predicted animpeller adiabatic efficiency of 91% at zero wheelefficiency as compared to a value of nearly 92%shown in Figure 11 for the near design operatingpoint. Thus, there exists a possibility for significantperformance improvement through an aerodynamicredesign of this impeller. Such a redesign should beexecuted not only to increase the impeller efficiencyby reducing the entropy rise, but also to produce moreuniform flow conditions at the impeller discharge. Itmight then be possible to reduce mixing losses andenhance the effectiveness of the downstream diffusingelement.

Impeller Aerodynamic Redesign StrategyThe measured adiabatic efficiency at a radius

ratio of 1.18 near the design flow rate is 86.7% whileat the impeller discharge, a peak adiabatic efficiencyclose to 94% is inferred from the measurements at thedesign speed. Thus, it seems possible to achieve asignificant gain in efficiency at the same stall marginif the root causes of this efficiency deficit areattacked. An efficiency audit which accounts for aprojected increase in wheel efficiency, reducedclearance gap, and lower mixing losses is attemptedbased on the results presented. The results fromFigure 11 augmented with other data at variousshroud clearance levels are trans-plotted in Figure 12.This provides an estimate of the sensitivity of impellerefficiency to changes in wheel efficiency. A linearleast squares fit is shown going through most of thedata. Clearly if the wheel efficiency could beincreased by 40% to an achievable level of 60% (seeRef. 4), about a 2% gain in adiabatic efficiency maybe realized for this impeller. It seems reasonablebased on Figure 8 that another 2% could be gained byimproving flow uniformity to reduce mixing lossesdownstream of the impeller since the losses in thisregion are currently estimated to cost about 5% inoverall efficiency at the near design operating point.Hence, a net gain of 4% in adiabatic efficiency isestimated for this impeller at the aerodynamic designpoint.

A principal cause of stagnation pressurelosses is the failure of the impeller to achieve itsmaximum static pressure recovery, which inevitably

leads to stagnation pressure mixing losses after theimpeller. This is supported by the low estimatedwheel efficiency. Aerodynamic synthesis of theimpeller points to the following remedies leading toefficiency gains: better shroud static pressurerecovery, secondary flow control, and reduced leakageflows by reducing the shroud clearance gaps. Staticpressure recovery can be increased by using betterflow quality concepts. Improvements in both theamount and rate of internal diffusion, hence increasedstatic pressure recovery, may be obtained by properendwall contouring and the use of three-dimensionalor sculptured blades to control the flow. As previouslydiscussed, there exist strong blade surface secondaryflow vortices within the impeller. These secondaryflows can be controlled and possibly suppressed usingcarefully designed 3D blade geometries similar towhat has been done by Zangeneh et al. [15]. Inaddition, increased diffusion and 3D blades will leadto reduced viscous dissipation within the impelleritself.

Using the ADPAC code coupled to ageometry generation scheme for the impeller, asystematic parametric evaluation of the impact ofcertain impeller design variables on performance canbe executed. This will lead to a correlation betweenimpeller geometry, internal flow, and performance.Enabling inverse design and optimization techniquescan later be deployed.

SUMMARYSUMMARYA good match between CFD and

measurements was obtained for an unshroudedcentrifugal impeller of approximately 4:1 pressureratio. Significant discrepancies between the velocitymeasurements and CFD did not appear until thepurely radial part of the impeller where they areattributed to inadequate turbulence modeling,numerical discretization errors, and measurementuncertainties. Overall, the CFD gave a good predictionof the measured performance and resolved enough ofthe local flow details to accord it a prominent positionin a design optimization cycle.

Aerodynamic synthesis of CFD results andmeasurements using laser anemometry revealed poolsof low relative kinetic energy fluid within the impellerpassage. The origins of this fluid were deduced to befrom blade boundary layer material, leakage flow, andfluid having been subjected to shear work along thestationary shroud. Strong secondary and leakage flowsgenerated within the impeller carry this fluid withinthe blade passage to form the observed flow structure.

Although the peak impeller efficiency ofnearly 94% at the design speed was quite high, verylow wheel efficiencies on the order of 20% or less

NASA TM-107515

8

were estimated from the measurements. As defined,wheel efficiency gives a measure of the effectivenessand aerodynamic quality of the relative diffusionprocess. Thus, the impeller shroud static pressurerecovery potential was judged to be underdeveloped.A 2% increase in impeller efficiency is projected ifthe wheel efficiency were to be increased to a morereasonable value such as 60%. Additional gains canbe derived from a reduction of the discharge flowdistortion which will reduce mixing losses that areincurred downstream of the impeller. Given the lowinitial value of wheel efficiency in this impeller, acase was made for significant performanceimprovements through the use of flow controlconcepts such as 3D sculptured blades and endwallcontouring.

A follow-on parametric study of the impactof certain design variables on internal flow structureand performance of this impeller can be reliablyperformed using the ADPAC code. The ensuingcorrelation between geometry, flow structure, andperformance will facilitate the ultimate goal ofimproved impeller and stage aerodynamicperformance.

REFERENCES1. Dean R., “On the Unresolved Fluid Dynamics ofthe Centrifugal Compressor,” Advanced CentrifugalCompressors, 1971, ASME Publications.2. Kano F., Tazawa N., Fukao Y., “AerodynamicPerformance of Large Centrifugal Compressors,”ASME Paper 82-GT-17.3. Moore J., Moore J. G., Timmis, P. H.,“Performance Evaluation of Centrifugal CompressorImpellers Using Three-Dimensional Viscous FlowCalculations,” J. of Engineering for Gas Turbines andPower, Vol. 106, pp. 475-481(1984).4. Vavra , M. H., “Basic Elements of AdvancedDesign of Radial-Flow Compressors,” AGARDLecture Series No. 89 on “Advanced Compressors,”1970.5. Krain H., “A Study on Centrifugal Impeller andDiffuser Flow,” Transactions of the ASME, Vol. 103,pp. 688-697(1981).6. Hathaway M. J., Chriss R. M., Wood J. R.,Strazisar A. J., “Experimental and ComputationalInvestigation of the NASA Low-Speed CentrifugalCompressor Flow Field,” ASME J. ofTurbomachinery, Vol. 115, pp. 527-542 (1993).7. Skoch G. J., Prahst P. S., Wernet M. P., Wood J.R., Strazisar A. J., “Laser Anemometer Measurementsof The Flow Field in a 4:1 Pressure Ratio CentrifugalImpeller,” ASME Paper 97-GT-3428. Hirsch Ch., Kang S., Pointel G., “A NumericallySupported Investigation of The 3D Flow in

Centrifugal Impellers,” ASME Paper 96-GT-151, and,ASME Paper 96-GT-152.9. McKain T. F., Holbrook G. J., “Coordinates for aHigh Performance 4:1 Pressure Ratio CentrifugalCompressor,” NASA Contract NAS 3-23268 (1982),(to be published as a NASA CR).10. Hall E. J., Delaney R. A., "Investigation ofAdvanced Counterrotation Blade ConfigurationConcepts for High Speed Turboprop Systems: TaskVII - ADPAC User’s Manual," NASA CR 195472(1995).11. Zangeneh M., Dawes W. N., Hawthorne W. R.,“Three-Dimensional Flow in Radial-Inflow Turbines,”ASME Paper 88-GT-103.12. Japiske D., Baines N. C., Introduction toTurbomachinery, Concepts ETI and Oxford UniversityPress (1994).13. Dean R. C., “The Fluid Dynamic Design ofAdvanced Centrifugal Compressors,” Creare, TN-185(1974).14. Schumann L. F., Clark D. A., Wood J. R., "Effectof Area Ratio on the Performance of a 5.5:1 PressureRatio Centrifugal Impeller," NASA TM 87237 (1986).15. Zangeneh M., Goto A., Takemura T.,“Suppression of Secondary Flows in a Mixed-FlowPump Impeller by Application of 3D Inverse DesignMethod: Part 1-Design and Numerical Validation,”ASME Paper 94-G T-45.

NASA TM-107515

LE

TE

10%30%

52%

70%

84%

96%

R/R2 = 1.1

R/R2 = 1.18

R1s/R2 = 0.486

R1h/R1s = 0.395

b2/R2 = 0.079

Z / R = 0.759

(Sta. 1)

(Sta. 2)

Z

R

Fig. 1– Illustration of impeller blading, flowpath, and reporting stations

(splitter LE)

b2

R1s

R1hR2

= R2 R1h

9NASA TM-107515

(every 4th point removed)

shroud clearance region

Leading Edge Closeup

Splitter LE

Fig. 2– Computational mesh of impeller discharging into vaneless diffuser showing closeups of impeller leading and trailing edges

Trailing Edge Closeup

Full Blade LEVaneless Diffuser

10NASA TM-107515

Fig. 3– Overall performance: design speed characteristic from inlet to R/R2 = 1.18

(a) Total pressure ratio

(b) Adiabatic efficiency (total to total)

11NASA TM-107515

Fig. 4– Circumferentially averaged static pressure distribution along the shroud at design operation

12NASA TM-107515

Fig. 5– Impeller exit spanwise distribution of circumferentially averaged total pressure at design operation (measurements at R/R2 = 1.1)

R/R2 = 1.18

R/R2 = 1.10

13NASA TM-107515

30 % Chord

52 % Chord

96 % Chord

Measurements

Vqm/Utip

Variable Tip Gap Constant Tip Gap

SSPS

rotation

Fig. 6– Development of quasi throughflow velocity within the impeller at design condition

CFD

passage 1passage 2

–0.2 0.1 0.3 0.6

15NASA TM-107515

10% Chord 30% Chord 52% Chord

70% Chord 84% Chord

96% Chord

R/R2 = 1.01 R/R2 = 1.05

R/R2 = 1.1 R/R2 = 1.18

SSPS

ROTATION

Fig. 7– Computed entropy distribution within impeller and vaneless space at design operation

splitter

17NASA TM-107515

Fig. 8– Development of mass-averaged entropy change within the vaneless space as computed from CFD results at the near design operating condition

Radius Ratio

19NASA TM-107515

70% Chord

84% Chord

96% Chord

Fig. 9– Secondary flow development within the impeller at design operation

ps ss

52% Chord

Helicitypassage 1passage 2

rotation

10% Chord 30 % Chord

21NASA TM-107515

(c) ImpellerTip

(b) Mean

(a) HubFig. 10– Computed blade static pressure loading at near design operation

S.S.P. S.

23NASA TM-107515

Fig. 11– Impeller performance at design speed for various corrected flows (Wcor) as derived from measured trailing edge shroud static pressure and overall total temperature at R/R2 = 1.18

(c) Wheel efficiency

(b) Adiabatic efficiency (total)

(a) Total pressure ratio

24NASA TM-107515

Fig. 12– Variation of impeller adiabatic efficiency with wheel efficiency as derived from measured shroud static pressure and overall total temperature ratio

25NASA TM-107515

REPORT DOCUMENTATION PAGE

2. REPORT DATE

19. SECURITY CLASSIFICATION OF ABSTRACT

18. SECURITY CLASSIFICATION OF THIS PAGE

Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources,gathering and maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of thiscollection of information, including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 JeffersonDavis Highway, Suite 1204, Arlington, VA 22202-4302, and to the Office of Management and Budget, Paperwork Reduction Project (0704-0188), Washington, DC 20503.

NSN 7540-01-280-5500 Standard Form 298 (Rev. 2-89)Prescribed by ANSI Std. Z39-18298-102

Form Approved

OMB No. 0704-0188

12b. DISTRIBUTION CODE

10. SPONSORING/MONITORING AGENCY REPORT NUMBER

8. PERFORMING ORGANIZATION REPORT NUMBER

5. FUNDING NUMBERS

3. REPORT TYPE AND DATES COVERED

4. TITLE AND SUBTITLE

6. AUTHOR(S)

7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES)

9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES)

11. SUPPLEMENTARY NOTES

12a. DISTRIBUTION/AVAILABILITY STATEMENT

13. ABSTRACT (Maximum 200 words)

14. SUBJECT TERMS

17. SECURITY CLASSIFICATION OF REPORT

16. PRICE CODE

15. NUMBER OF PAGES

20. LIMITATION OF ABSTRACT

Unclassified Unclassified Unclassified

National Aeronautics and Space AdministrationWashington, DC 20546–0001andU.S. Army Research LaboratoryAdelphi, Maryland 20783–1145

1. AGENCY USE ONLY (Leave blank)

Technical Memorandum

NASA Lewis Research CenterCleveland, Ohio 44135–3191andU.S. Army Research LaboratoryCleveland, Ohio 44135–3191

This publication is available from the NASA Center for AeroSpace Information, (301) 621–0390.

A03

22

July 1997

WU–523–22–13

E–10818

NASA TM–107515ARL–TR–1461AIAA–97–2878

Unclassified -UnlimitedSubject Category 07

Aerodynamic Synthesis of a Centrifugal Impeller Using CFD and Measurements

L.M. Larosiliere, G.J. Skoch, and P.S. Prahst

Prepared for the 33rd Joint Propulsion Conference and Exhibit cosponsored by AIAA, ASME, SAE, and ASEE, Seattle, Washington,July 6–9, 1997. L.M. Larosiliere and G.J. Skoch, U.S. Army Research Laboratory, Lewis Research Center, Cleveland, Ohio 44135; P.S.Prahst, NYMA, Inc., 2001 Aerospace Parkway, Brook Park, Ohio 44142 (work funded by NASA Contract NAS3–27186). Responsibleperson, L.M. Larosiliere, organization code 5810, (216) 433–3403.

Turbomachinery; CFD; Centrifugal compressor

The performance and flow structure in an unshrouded impeller of approximately 4:1 pressure ratio is synthesized on thebasis of a detailed analysis of 3D viscous CFD results and aerodynamic measurements. A good data match was obtainedbetween CFD and measurements using laser anemometry and pneumatic probes. This solidified the role of the CFD modelas a reliable representation of the impeller internal flow structure and integrated performance. Results are presentedshowing the loss production and secondary flow structure in the impeller. The results indicate that while the overallimpeller efficiency is high, the impeller shroud static pressure recovery potential is underdeveloped leading to a perfor-mance degradation in the downstream diffusing element. Thus, a case is made for a follow-on impeller parametric designstudy to improve the flow quality. A strategy for aerodynamic performance enhancement is outlined and an estimate ofthe gain in overall impeller efficiency that might be realized through improvements to the relative diffusion process isprovided.