I .

AIAA 92-3313 A NAVIER-STOKES ANALYSIS OF A CONTROLLED-DIFFUSION COMPRESSOR CASCADE AT INCREASING FLOW ANGLES G. V. Hobson Naval Postgraduate School Monterey, CA

AI AAIS A El ASM El AS E E 28th Joint Propulsion

Conference and Exhibit Julv 6-8, 1992 / Nashville, TN

For permlsslon lo copy or republlsh, contact the Amerlcan lnstltule of Aeronautlcs and Astronautics 370 L'Enfam Promenade, S.W., Washlngton, D.C. 20024

A Navicr-Stokcs Analysis of a Controlled-1)iffusion Compressor Cascadc

a t 111 c rcasi ng 111 Icl - Flow An g Ics

Garth V. Hobson*

Turbopropuls ion Laboratory Naval Postgraduate Scliool

Montcrey, CA 93943

d is s i 11 ; i t i 0 I I rat c of t u rhul en ce A b s t r a c t

'l'lic cxtensivc sct or expcriincntal data, which was measnrcd primalily for codc validation. was used for comparison with a finite volume, incompressible , Navicr-Stokcs solution of the viscous flow through a set of controllctl-difrusion compressor bladcs. This was done at increasing inlet-flow angles, and tlic solution was optimized for tlic dcsign irilct-[low angle. The main ohjcctivc o f this paper is to predict the losses through thc cascade, howcvcr; another ohjcctivc is to predict the turbulcnce field t rp~trcain, tlirougli and aft of the blade row. Exccllcnt comparison!: were achicvcd at dcsign incidence, however; a t increasing flow angles tlic losses were not accuratcly prctlicrcd with tlic codc.

N o in e n c I a t u r e

skin friction cocfficicnt pressure cocff ic icnt

- empirical constants i n tlic t urbu I en t d i ss i p at i on cq u at i on

- empirical constant in the algebraic eddy viscosity model

- empirical functions in the turbulent dissipation equation

- empirical function in tlic algebraic eddy viscosity model turbulent kinetic energy inlct turbulence length scale p r c s s u r e t u rb it I en c e axial velocity tangential veloci ty total velocity componcnt axial dircction

in t en sit y

- tangent ia l direct ion

* Mcmbcr AIAA: Associate Professor

1

rnolccul :I r viscosity tnthulcnt cdtly viscosity incan strain rate squared d cn s i t y cinp i rica I c o i l st a n t s i n turbulcncc model wall shear strcss mass avcragctl total prcssnrc I oss coc f Ti c i c ti t

h U L L S U I y I

11

S u b s c r i t i t s

i o r l - i n Ict cond i t i ~ n s (I o r 2 - outlct valucs 1

i t c r a t i on c v 11 n t c r

-

s t ngn at i 0 11 q ail t i t ic s

I n t rod u c t i o n

The succcssful prediction or cascadc and turbomachincry rlows should include an adequate resolution of tlic turbulence licld throughout the blade passagc. II this is not accurately prcclicted thcn the growth of the bladc surface boundary laycis will not be correctly captured as well as the wake distributions. The wakc profile a n d static pressurc rccovcry aft of tlic hladcs both need to hc coinputcd i f tlic losscs arc to be accuratcly prcdictctl. This paper is an attempt to resolve thcsc phcnoincna as well as tlic increase of inlet Ticcstrearn turbulence intensity as tlic flow approachcs a cascade of controllcd-difTiision compressor b I ad c s .

Recently, Davis ct a l . ’ , pcrformcd Navicr- StokcT predictions of comprcssor cascade pcrformancc with good overall results. Howcvcr they did not show comparisons, bctwccn the computational and cxpcrimcntal results, of the finc-scale details of the flowficld through thc cascades. They used an algebraic eddy-viscosity turbulcncc modcl, proposed by Baldwin and Lomax, which was modificd for tlic calculation of separated flows. Transition of thc flow in the blade surface boundary laycr is usually specified empirically with little o r no physical cffccts takcn into account. Davis ct al. did not mention how transition was modcllcd in their calculations.

I l o et a1.2 uscd a onc-cquation modcl which can account for frccstrcam turbulcncc cffccts on tlic hound;i~y laycr growth. I f ihc turbulent kiiictic cncrgy cquation is intcgratcd to the wall thcn thc cffcct of frccstrcam turbulcncc intcnsity on transition can bc accountcd for, as was nicntioned in their paper. Thcy showcd good overall results for thc pcrformancc prcdiction for the same set of controllcd- diffusion compressor bladcs, but at the low

v

w incidcncc anglcs.

A finitc-volume Navier-Stokes solution of thc viscous flow through a comprcssor cascade of controlled-diffusion blade profilcs at various incidcncc anglcs is prcscntcd. The ability of thc code to prcdict fully turbulcnt flow scparation has bccn succcssfully dcmonst ra tcd l with a low- Rcynolds-number two-cquat ion turbulcncc modcl as proposcd by Lam and Brcmhorst4 This modcl is uscd to prcdict thc laminar Icading-cdgc scparation bubblc, which dcvclops 011 tlic suction sidc of the bladcs i n tlic prcscncc of high inlct frccstrcam turbulcncc. Thc incrcasc in tnrbulcncc intcnsity, which has bccn nicasurcd5, must be computed if thc lcading edgc scparation bubblc and subscqucnt growth of thc suction-sidc boundary laycr arc to bc accuratcly prcdictcd.

Thc controlled-diffusion bladcs considcrcd arc thosc designed by Sangcr6, and cxpcrimcntally tested at various incidcncc

v anglcs by Elazar7. Good predictions of blade

2

surfacc prcswrc distribution and losses havc bccn achicvcd at tlic low incidcricc anglcs considcrcd by Elazar. Ilowcvcr; no dctailcd mcasurcmcnts wcrc takcn in the Icading-cdgc scparation bubblc rcgion at thcsc incidcnccs, hcncc no comparisons could he madc with thc cxpcrimcntal data.

An expcrimcntal program has hccn carried out5 to measure tIic fIowficId in the cascade at an inlct flow anglc o f 48 dcgrccs which has a largcr scparation bubhlc. ‘Thc flowficld upstrcam of thc cascadc has bccn fully characterized, particularly thc turbulent flow quantitics. Thc suction sidc boundary laycr docs not scparatc at this high iricidcncc anglc. This i s duc to tlic high turbulcncc produccd in thc lcading cdgc rcgion around the scparation bubblc.

Tnrhulcncc Motlcl

Thc Lam and Brcnihors14 k-e turbulcncc modcl is uscd, which includcs tlic cffcct of frccstrcam turbulcncc intcrisity on the boundary laycr transition. Thc two- dimcnsional form of thc cquations arc as fol lows;

where Cp: C E I ; Cez; q; and oE, arc constants and fp; f l ; and f2 arc damping functions and wherc all uscd as rccommendcd in ref. 4.

During thc relaxation proccdurc tlic turbulent viscosity is undcrrclaxcd with thc

whcrc thc superscript n denotes the itcration level. As a valuc of . I % was used for the relaxation parametcr throughout this study, then an initial guess of pt equal to zcro gavc production terms in the k and E equations that were insignifant. Hence no turbulcncc was produced in the shcar layers, thus initially pt was sct cqual to the laminar viscosity.

Boundarv Co nditions

Thc inlct flow angle and velocity wcrc spccificd at thc upstrcam boundary of thc 160x242 "11-type" grid uscd, an examplc of which is shown in figure 1. All tlic computational tcst cases were run at a Rcynolds number of 700,000, based on chord length, which corrcspondcd to thc experimental test condition. For each test case, which had varying inlet-flow anglc, a ncw grid was generatcd with the grid inlct anglc cqual to thc inlct flow anglc.

The inlct frecstrcam turbulence intensity was set equal to 1.5% which was cxpcrimcntally measured using a two- component lascr Doppler v c l o c i m c t c r ~ ~ ~ . The inlct turbulent kinctic encrgy is then dctermincd from thc following:

ki =Z(TuiUi)Z 2 161

Tlic lcvcl of turbulence intensity was also verified with a singlc wire hotfilm probc, whose autocorrelation gavc a mcasurc of tlic integral scale of the incoming turbulcncc5. The intcgral scale was found to be approximately 1" which corrcspondcd to thc inlct guidc vane spacing, and this valuc did not vary with incidcncc. Thus a lcngth scale of 0.2 (non-dimcnsionalized with respect to thc chord length which is 5 " ) was used as input to thc program to determine thc inlct frecstream lcvcl of turbulence dissipation ratc, as spccificd by thc following equation;

paramctcr wp, as Follows;

~

3

If equation 171 is siibstitutcd i n t o cquation [3] , then tlic wcll known Prandtl- Kolrnogorov relation is rccovcrcd;

pti = fpCpki l /*Li [ 8 1

Now i f [6] is substitutcd inlo [8] then thc following rcsult is obtained;

I /' pti = fPC@ Tui Ui Li [91

which shows that thc inlct turbulcnt cddy viscosity is dircctly proportional to ttic inlct turbulence intensity and length scale. 'I'hc correct spccification of tlicsc two paramctcrs is crucial not only for thc initialization of thc turbulcncc ficld, but also to corrcctly spccify the cffcctivc Reynolds number which is uscd by thc codc as an upwind/ccntral diffcrencc switch.

A periodic solver was uscd upstream and downstrcani of tlic cascade which cnlorccd pcriodicity of all ttic flow variablcs ahcad of and downstrcam of ttic bladcs.

No-slip vclocity boundary conditions wcrc used on tlic bladc surfaccs, and thc normal momentum equation was uscd to cxtrapolatc thc static prcssurc toward thc bladc surrace. On ttic bladc surfacc the turbulcnt kinctic energy was set equal to zcro and thc normal derivative of the dissipation ratc was also sct cqual to zcro. For t h c low-Rcynolds-number form of thc k-E modcl to work SuccssFuIly a fine cnough grid nccds to bc uscd, such that at lcast fivc grid points arc within the laminar sublaycr. This was satisricd by ensuring that [tic first grid point off the bladc surfacc was at a y+ of lcss than one.

At the exit plane all flow variablcs wcrc extrapolated outwards, with a zcro gradient in thc dircction of thc frccstrcam vclocity. This esscntially satisfied thc assumption that Ihe streamwise dilruusion at this location was n e g l i g i b l c .

Resu l t s

Tlic prcdictcd bladc surfacc prcssurc distribution, and its comparison with tlic cxpcrimcntally mcasurcd values*, is sliown in figurc 2, for the dcsign inlct flow anglc of 40 degrees. The agreement is cxcellcnt, and some features of the pressure distribution arc worth noting. On the prcssure side (upper curve) the pressure distribution i s rclativcly flat, and thc slight variation is due to thc local curvature of the hladc. On thc suction sidc of the hladc the lcading cdgc radius produccs a suction spikc within the first 5% of thc bladc chord. Thc cxtrcme advcrsc prcssurc gradicnt on thc rcar of this spikc produces a lcading cdgc scparation bubble which has bccn obscrvcd cxpcrimcntally with a china-clay surfacc flow visualization tcchniquc9. After this initial suction prcssurc thc bladc surfacc prcssure dccrcascs until 30% chord duc to thc local curvature of thc bladc. Thcn thc final diffusion takcs placc on tlic rcar of tlic hladc surfacc at a ncar constant rate. As mcntioncd carlicr to accuratcly predict the Iosscs, thc static pressure rccovcry in thc wakc nccds to bc computcd. A valuc of Cp2 = 0.386 was computed, which comparcs

valucs of 0.34 (Shrccvc ct. aI . lO).

The prcdictcd nican flow vclocity distribution (Fig. 3) bctwccn tlic blatlcs, at stations 2 through 8, are comparcd to thc cxpcrimcntally mcasurcd LDV data7. From the lcading cdgc to 41.7% axial chord location the prcdiction of the frccstrcnm vclocity lcvcl is vcry good. Both the suction- and pressure-side boundary layer profilcs are well prcdictcd as wcll as thc frecstrcam velocity Icvcls.

Howcvcr the comparison in the rcar part of the passage (Fig. 4) is not as good as thc front part. The codc i s strictly two dimensional and cannot account for thc axial vclocity dcnsity ratio (AVDR), licncc the predictcd frcestream velocity diffcrs from the experimental data as the flow approaches the trailing cdgc. On the suction sidc whcre the growth of the boundary layer is significant the comparison is good, howcvcr; the actual shapc of the vclocity profilc is not accuratcly capturcd. On the prcssurc sidc whcre there is vcry littlc

v

- favorably with thc cxpcrimcntally mcasurcd

4

boundary layer growth tlic agrccmcnt is good. At the 99% chord location tlic cxpcrimcntal data shows an ovcrshoot of thc frccstrcam vclocity on tlic prcssurc side. Although tlic lcvcls arc off, tlic computations also sliowcd a slight ovcrshoot or wall jct vclocity profilc. This ovcrshoot is due to thc bulging of thc hladc profilc at thc trailing cdgc, which is charactcrislic of a controllcd-diffusion comprcssor hladc.

Figure 5 shows thc prcdiction of thc wakc profiles downstrcam of the cascadc. At the first two stations (5.5% and 7.6% axial chord downstrcam of thc hladcs) tlic codc prcdictcd rcvcrsc flow at the wakc cctitcr linc. The flow rcvcrsal manifcstcd itsclf as two rccirculation regions on either sidc of thc trailing cdgc. The wakc ccntcr line also coincided with tlic first grid linc off the hladc surfacc in tlic downstrcam dircclion, howcvcr; by the third profilc (14.1%) thc wakc ccntcrlinc has movcd off tlic first grid linc arid thc prcdictcd wakc piofilc has only positivc vclocity valucs. By tlic final station (22.2%) tlic prcdictcd wakc profilc is quitc good, as tlic wakc deficit is almost prcdictcd. There is still somc cvidcucc that tlic first grid linc and tlic fine grid clustcring on eithcr sidc of this linc is contaminating thc solution, as tlic "k ink" in tlic prcdictcd profilc is on this linc. All the prcdictcd wakc profilcs do not match the data in the frccstrcam, this once again is an indication that tlic codc cannot accourit for the accclcration of the frccsrrcam due to stream tub c con t rac t ion.

Fig. 6 shows tlic prcdictcd turbulcncc intcnsity, which was calculatcd from the prcdictcd turhulcnt kinctic cncrgy, and its comparison with thc cxpcrimcritally mcasurcd valucs across the wake at thc samc axial locations downstrcam of thc trailing cdgcs. Although thc agrccmcrit is not good [or all tlic profilcs, tlicrc arc sonic fcarurcs which are qualitativcly similar. Thc first two prcdictcd profilcs (5 .S% and 7.6%) give good frccstrcam valucs, howcvcr; tlicrc is too much diffusion in tlic turbulent kirictic cncrgy equation as the profilc starts to approach the first rclativc maximum to soon on the prcssure side. Thc maximum valuc of approxirnatcly 14% turbulcncc intensity i s not prcdictcd, as thc computed distribution actually shows scvcrc grid dcpcndcncc, and only produccd a maximum of 8%. Tlic

prediction shows the doublc maximum on tlic suction side (right) o f the wakc, which is also cvidcnt in thc cxpcrimcntal data. By tlic last two profilcs (14.1% and 22%) the profile shapc is orf as i t shows a maximum on thc suction sidc and not on the prcssurc side as is evidcnt in thc cxperimcntal data. Once again the maximum prcdictcd lcvcl of turbulcncc intensity i s approximatcly 40% off, howcvcr the width of the wakc intcraction is wcll prcdictcd. Thus ovcrall the two-cquation turbulence model cannot prcdict the high lcvels of turbulcnce intensity gcncrated in this flowficld which has a rclativcly high inlct freestream turbulcncc intensity. Tlic ability of the niodcl to accuratcly prcdict the turbulcncc field is dcpcndcnt on thc grid rcsolution and grid stretching across thc wakc. A morc uniformily spaced grid was also invcstigatcd. which gavc bcttcr wakc profilcs. but prcdictcd flow scparation on the bladc suction surfacc at this incidence an g 1 c.

The prcdictcd suction-sidc skin-friction distribution is shown in Fig. 7. The ncgativc skin friction shows the cxtcnt of thc separation rcgion. and the rcst o f the bladc surfacc has an attached boundary layer. The vcry low valucs of skin friction on tlic rear of the blade, particularly aft of 40% chord, was an indication that separation may occur in this rcgion. Computed velocity vcctors around thc lcading edge rcgion (Fig. 8), show thc reverse flow on the uppcr surfacc.

Turning now to the orf-dcsign pcrformancc prediction, thc code was run with increasing inlct flow anglc at tlic same Rcynolds number of 700,000. Fig. 9 shows that tlic predictcd blade surfacc prcssure distribution, at 43 dcgrccs inlct flow anglc, which compares wcll with the cxpcrimcntally mcasurcd values*. Tlicrc is a slight discrcpancy around thc Icading cdgc on thc prcssurc side, and tlic diffusion on the suction sidc aft of 30% chord is also slightly off. As can bc cxpcctcd tlic lcading cdgc separation bubblc is larger for this inlet-flow anglc, as shown by thc suction- sidc skin-friction distribution in Fig. I O . Tlic negative lcvels of skin friction sccni t o bc unrealistically high as thcy excecd the lcvcls further downstream whcrc the flow is still attached. The lcvcls of skin friction on

thc attached portion o f tlic blade arc lowcr than thc design inlct flow anglc casc.

Tlic prcdictcd bladc-surfecc ptcssurc distribution, at an inlct flow angle of 46 dcgrecs is shown in Fig. 1 I . These computations are comparcd to thc bladc- surfacc static prcssurc incasurcmcnts pcrformcd by Koyuncul 1 , who pcrfromcd wakc survcys for varying inlct-flow angles from 24 to 46 dcgrecs. Now tlic comparison on thc prcswrc sidc is vcry good, cxccpt at the trailing cdge. On thc suction sidc a doublc minimum prcssurc is prcdictcd, which docs not coincide with tlic expcrimcnt, The diffusion o n thc rcst of tlic bladc is howcvcr cayturcd, except for the trailing edge, whcrc t h c cxpcrimcntnl data shows somc irrcgularity. Tlic lcading cdgc separation bubble, now shown as a scrics of niultiplc spikcs on the skio-friction distribution (Fig. 12), has grown significantly. This flow rcvcrsal most probably causcd the lack of :igrccmcnt o f tlic prcssurc distribution at tlic Icading cdgc (Fig. 11). Thc vclocity vcctors for this casc arc shown i n Fig. I ? , which clearly show thc incrcascd Icading cdgc bubblc.

Prcliminary computations wcrc run at 48 dcgrecs i n k - f l o w anglc. Thc prcdictcd bladc-surface prcssurc distribution did not agree with thc cxpcrimcnt at all, and thc skin-friction distribution sliowcd a largc lcading cdgc separation rcgion with ncar zcro shcar on the rcst of the suction surfacc. Thc prcdictcd m a s s avcragcd losscs dcfincd as follows;

1101 G-Fll - F,2 Fll - I.1

arc cornparcd with both Koyuncu'sl ' a n d Drcon ' s8 cxpcrimcnt:illy incasurcd valucs. Qualitativc agrcciiiciit is acliicvcd bclwccri tlic dcsign inlct-flow anglc and 46 dcgrccs, howcvcr at 48 dcgrccs tlic code was not ablc to prcdict tlic continucd incrcasc in losscs.

Con c 111 si o n s

By correctly spccifying tlic inlet boundary conditions. particularly tlic turbulcncc intcnsity and Icngth scale, a good prcdiction o f the cascadc pcrformancc was achicvcd at

tlic dcsign inlct flow angle. Tlic loss prediction was of f by 30% wliich is citlicr dues 10 [lie inability of tlic turbulcncc inodcl

intcnsity in thc wakc or the inability of tlic codc to account lor thc AVDR.

A significant feature of tlicsc computations is tlic ability of thc code to computc tlic leading edge separation bubblc and thcn remain attached downstream to the trailing cdgc for most of the inlct flow anglcs considcred, except thc highcst. The far downstream static pressure recovery was also well predicted,

At thc dcsign inlct flow angle the turbulcncc field, particularly in thc wakc, was qualitatively predicted. Qualitative agrccmcnt of thc loss distribution was also prcdiclcd if thc flow rcniaincd attachcd on the suction side of thc bladc.

R e f c r c n ccs

- to predict the high lcvcls o f turbulcncc

I . Davis, R. L., Ilobbs, D. E. and Wcingold, 11. D., "Prediction of Compressor Cascadc Performance Using a Navicr-Stokes Tecliniquc," ASME Journal of Turbornachinc-ry. Vol. 1 10, pp 520 - 531. October 1988.

2. Ho, Y. K., Walker. G. J. and Stow, P.. "Boundarv Laver and Navier-Stokes Analvsis of a NASA Controlled-Diffusion Compressor Bladc," ASME 90-GT-236.

3. Hobson, G. V. and Lakshminarayana, B., "Prcdiction of Cascade Pcrformancc Using an Incompressible Navier-Stokes Technique," ASME Journal of Turbornachinery, Vol. 113, pp 561 - 572, Oct 1991.

4. Lam, C. K. G. and Bremhorst, K., "A Modified form of tbc k-epsilon Modcl lor Predicting Wall Turbulence," Journal of Fluids Engineering, Vol. 103, pp. 456 - 460, 1981.

5. Hobson, G. V. and Shrcevc, R. P. S., "Inlct Turbulcncc Distortion and Viscous Flow Development in a Controlled-Diffusioii Compressor Cascade at Very High Incidcncc," AIAA-91-2004 (To be published in the AIAA Journal f o r Propulsion and

P o w e r )

6. Sangcr, N. L., "Tbc Usc of Optimization Tcchniqucs to Design Controlled-Diffusion Comprcssor Blading," ASME Joicrrinl of Engineering f o r Power, Vol. 105, pp. 256 - 264, April 1983.

7. Elazar, Y., " A Mapping of ttic Viscous Flow Behavior i n a Controllcd Diffusion Compressor Cascadc and Prcliniinary Evaluation of Codcs for tlic Prcdiction of Stall," P1i.D. T h i s and Tcclinical Report NPS67-88-001, Naval Postgraduatc Sctiool, March 1988.

8. Drcon, J. W., "Controllcd Diffusion Comprcssor Bladc Wakc Mcasurcmcnts," M.Sc. Thcsis, Naval Postgraduate School, Scptcmbcr, 1986.

9. Sangcr, N. L., and Shrccvc, R. P., "Comparison of Calculatcd and Expcrimcntal Cascadc Pcrrormancc for Controlled- Diffusion Coniprcssor StaIor Blading," ASME Journal of Turhomachinery, Vol. 108, pp 42 - SO, July 1986.

IO. Shrcevc, R. P., Elezar, Y., Drcon, J. W., and Baydar, A,, "Wakc Measurcnicnts and Loss Evaluation in a Controlled Diffusion Comprcssor Cascadc," ASME Journal of Turboinachinery, Val. 113, pp. 591 - 599, Oct. 1991.

11. Koyuncu, Y., "Report of l'csts of a Comprcssor Configuration of CD Blading," M.Sc. Thcsis, Naval Postgraduatc School, March. 1984.

G

Y

Fig. I Computational "11-type" grid

0.0 0.2 0.6 0.R I .o X/C:llortl

Fig. 2 Cotlipatison of the prcdictcd and nicasurcd blade surface pressure distribution at design.

7

15.1% ?

Fig. 3 Comparison of the computed and mcasurcd (LDV) pitchwise velocity distributious i u the forward section of tlic bladc passage.

8

Fig. 4 Comparison o f tlie computed and measured (LDV) pitchwise velocity distributions in the rearward section of [lie blade passage.

9

1.6%

5 3 %

Fig. 5 Comparison of the con~putcd and measured (LDV) wake prorilcs.

10

Fig. 6 Comparison of the computed and measured (LDV) turbulence intensity distribution across thc wakes.

1 1

d

L

W

11.11 11.2 11bl 11.6 11 x I I1

X I ( ' 1 1 0 , O

suet i o i i -si de Fig. '7 Corn pu tcd ski ii - rr i ci i 011

coerficiciii tlistribuiioii a i tlic desigii iiilct flow aiiglc.

X

Fig. 8 Cotiil~iiictl vclociiy vcctois ar(irttitl ilic lcatliiig ctlgc a i design.

12

Fig. 9 Coniparisott of tlic prctlictctl niid nicasurcd blndc sud'acc prcssnrc disiribution at an inlct Ilow nnglc of 43 tlcgrccs.

Pig. I 1 Coinp~irisoti o f tlic prcdictcd and mcnsurcd hlndc surface pressure distribution ai nn inlct f low anglc o r 46 tlrgrccs.

l'i[:. I 2 C'onipulcd S K I io11 -sick s kit1 -TI iciioti distril)ufiori at ai inlci f low ntiglc c>f 4G dcgrccs.

Fig. I O Cornpotcd suction-sidc skiti-rriction disirihution at an irtlci fforv nnglc o f 43 dcgrccs.

X

Fig. 13 Computed velocity vectors around the leading edge at an inlet flow angle of 4 6 degrees.

Fig. 14 Comparison of the computcd an measure loss distribution for the control led-diffusion compressor blades.

14