Experimental Thermal and Fluid Science 44 (2013) 301–311

Contents lists available at SciVerse ScienceDirect

Experimental Thermal and Fluid Science

journal homepage: www.elsevier .com/locate /et fs

The evolution of the flow topologies of 3D separations in the stator passageof an axial compressor stage

Xianjun Yu ⇑, Zhibo Zhang, Baojie LiuNational Key Laboratory of Science & Technology on Aero-Engine Aero-Thermodynamics, School of Energy and Power Engineering, Beijing University of Aeronautics & Astronautics,Beijing 100191, China

a r t i c l e i n f o a b s t r a c t

Article history:Received 18 January 2012Received in revised form 7 June 2012Accepted 7 July 2012Available online 20 July 2012

Keywords:Corner separationTopological analysesOil-flow visualizationStereoscopic PIVAxial compressors

0894-1777/$ - see front matter � 2012 Elsevier Inc. Ahttp://dx.doi.org/10.1016/j.expthermflusci.2012.07.00

⇑ Corresponding author. Tel.: +86 10 82316455; faxE-mail addresses: yuxj@buaa.edu.cn, y_xjun@126.c

Flow separations in compressor blade passages are common and can cause significant flow blockage andlosses production in some instances. Because of the measurement difficulties, most of the previous stud-ies concerning the compressor 3D separations were conducted in cascades facilities. In this paper, 3Dseparation flows were studied in the stator of a low-speed compressor test facility. In order to find theirevolutionary processes, oil flow visualizations were conducted at four compressor operating conditionsfrom the design to near-stall conditions. The results showed that the corner separations appeared at eventhe design condition; however, they were so weak causing very slight flow blockage until the maximumstatic pressure rise condition arrived. By using topological analyses methods, exact 3D flow structuresinside the stator passage were reconstructed and their flow mechanisms were analyzed. It was foundthat, as the mass flow rate decreased, the flow topologies of 3D separations became much more complexand the evolutionary processes of the 3D flows were significantly affected by both the flow–flow inter-actions inside the stator passage and the rotor–stator interactions between blade rows. However, thecomplicated 3D flow structures in the tested stator passage always consist with four basic types of flows:the corner vortex flow, the flow of the corner separation with/without the ring-like vortex, and the bladesurface separation flow. Finally, the results obtained based on the topological analyses of the oil-flowvisualization pictures were validated by using the measured results of stereoscopic particle imagevelocimetry.

� 2012 Elsevier Inc. All rights reserved.

1. Introduction

Three-dimensional separations in blade passages, especially forthe end-wall corner regions, are common in compressors, whichmay cause significant aerodynamic blockage and losses production,and can even affect the compressor stability [1–3]. Hence, in thepast few decades, it was always a key flow phenomenon concernedby compressor aerodynamicists. Great amount of researches wereconducted for clarifying the flow mechanisms and aerodynamicfeatures in turbomachinery by experimental, theoretical, andnumerical methods [4–10].

It is well known that there are two basic factors which affect theformation of three-dimensional separations in compressor bladepassages: (i) the adverse pressure gradient in the streamwise direc-tion and (ii) the secondary flow effects (skew of the endwall bound-ary layer, circumferential and radial flow migration in the near wallregions) in blade passages [11,12]. Considering that the three-dimensional separations in compressor test facilities or real engines

ll rights reserved.2

: +86 10 82316418.om (X. Yu).

are always very complex and are difficult to be investigated indetail, fortunately, the basic formation mechanisms of them canbe simulated well by cascades, most of the researches related tothe three-dimensional separations in turbomachinery were con-ducted in planar cascades [3,8,10,13,14]. However, because of theblade twists in radial direction (which can cause non-uniformityspanwise distribution of blade loadings or flows) and the strongsecondary flow effects and even the rotor–stator interaction phe-nomena, the separations in real compressors might have muchstronger three-dimensionalities than those in cascades [1,2].Therefore, the evolutionary processes and mechanisms of thethree-dimensional separations in compressors might appear muchdifferent with that in planar cascades and should be studied exper-imentally in detail, which is helpful for modeling of the compressorendwall flows, developing the flow control methods and validatingof the CFD schemes and codes.

In the present study, we investigated the three-dimensionalflow structures, particularly for the separation flows, in a low-speedlarge-scale compressor test facility by experimental methods andclarified their evolutionary processes in stator passages. In theexperiments, the oil-flow visualization technique was firstlyemployed and used to investigate the flow topologies of the limit-

Nomenclature

AbbreviationsSS blade suction surfacePS blade pressure surfaceLE blade leading edgeTE blade trailing edgeDE compressor design conditionMID compressor middle conditionMSPR compressor maximum static pressure rise conditionNS compressor near-stall conditionSL separation lineAL attachment lineF focus pointN node pointS saddle pointSPIV stereoscopic particle image velocimetry

NotationL rotor blade chord length at mid-spanVtip rotor tip speedq air densityl air dynamic viscosity coefficientRec Reynolds number=qVtipL/lxz out-of-plane vorticityX rotor rotational speedw out-of-plane velocityPin compressor inlet static pressurePout compressor outlet static pressureVaix axial flow velocityW static pressure rise coefficient, ðPout � PinÞ=ð1=2qV2

tipÞu mass flow coefficient, Vaix/Vtip

X, Y in-plane coordinates of SPIV measurement planesZ out-of-plane coordinates of SPIV measurement planes

302 X. Yu et al. / Experimental Thermal and Fluid Science 44 (2013) 301–311

ing streamlines on the endwall and blade surfaces at four compres-sor operating conditions along the compressor operating line. Andthen, based on the topological analyses, detailed 3D separationand vortex flow structures were reconstructed and their evolution-ary processes and flow mechanisms were analyzed. Finally, thereconstructed 3D flows based on the oil-flow visualization resultswere validated by the results obtained by using stereoscopic parti-cle image velocimetry (SPIV).

2. Experimental setup

2.1. Compressor facility

The test facility is a single-stage axial compressor with inletguide vanes (IGVs). A large contraction ratio bellmouth with a1.8 m outer diameter lemniscate profile equipped with flowstraightener is used to provide uniform and steady inlet flow. Therotor and stator blades with C4-series airfoil are designed in termsof the free vortex law. The stator blades are cantilevered with aclearance of about 1.5 mm. More detailed design parameters aresummarized in Table 1, which were also introduced by Yu and Liu[15]. In this study, measurements were conducted at the designcondition (DE, u = 0.58), the middle condition (MID, u = 0.51), themaximum static pressure rise condition (MSPR, u = 0.43) and thenear-stall condition (NS, u = 0.38), as shown in Fig. 1.

Table 1Representative design parameters of the compressor test facility.

Outer diameter (m) 1.0Hub-to-tip ratio 0.6Design speed (rpm) 1200Design mass flow rate (kg/s) 22.4Design flow coefficient, u 0.58Design static pressure rise coefficient, W 0.48Rec 6.6 � 105

IGV-rotor space (mm) 41.9Compressor configuration IGV + rotor + statorNumber of blades 36 + 17 + 20Blade camber angle (�) 17.4 + 26.5 + 49.1Blade stagger angle (�) 10.4 + 33.4 + 12.3Blade height (mm) 200 + 199 + 198Blade chord near hub (mm) 100 + 160 + 180Blade chord at mid-span (mm) 100 + 175 + 180Blade chord near casing (mm) 100 + 200 + 180Solidity at mid-span 1.43 + 1.18 + 1.43Rotor tip clearance (mm) 1.0Stator near hub clearance (mm) 1.5Rotor–stator space (mm) 48.7

2.2. Oil-flow visualization

The oil-flow visualization tests were conducted to qualitativelyindicate the flow patterns on the endwall and blade surfaces [16].The used oil is a mixture of silicon oil, pigment particles and mag-nesium oxide powder. Because the flow is very complex in thestator passage, particularly on the blade suction surface, surfaceflow visualization tests were run after many initial tests to opti-mize oil–paint mixture and thickness as well as the runtime. Gen-erally speaking, the places where the flow is regular with high flowvelocity should be painted with a relative dry oil mixture, while theplaces with low flow velocity, such as the reverse flow and stagna-tion flow regions, should be painted with a relative dilute oil mix-ture. In the presented experiments, the ratio of the oil and solidpower was chose between 1:4 and 1:2 and the ratio for the pig-ment and magnesium oxide was set as nearly 1:1. As a result,the runtime of each test was between 2 and 4 h.

2.3. SPIV measurements

A commercial SPIV system, developed by TSI Incorporation, wasemployed in the SPIV measurement. The light source is a dual cav-ity Nd:YAG laser, the maximum illumination energy is 150 mJ/pulse at a 15 Hz repetition rate. A pair of 1280 � 1024 pixels and12 bit frame-straddling based CCD cameras (PIVCAM 13-8) were

Fig. 1. Characteristics of the test compressor facility.

Fig. 2. Schematic layout of the measurement cross sections.

X. Yu et al. / Experimental Thermal and Fluid Science 44 (2013) 301–311 303

configured in the different sides configuration in Scheimpflüg con-dition. The fields of view inside the stator passages were about200 � 100 mm, which covered nearly half of the whole blade pas-sage. However, because of the combined effects of geometricrestriction of the blade passage and strong flare light at the hubsurface, the effective field of view covers only 50–70% blade spanheight. The layout of the measurement cross sections is shown inFig. 2. All of the 12 cross sections are nearly perpendicular to thelateral surfaces at the blade tip. Detailed experimental setup anddata processing methods have been introduced by Yu and Liu [15].

3. Results and discussion

3.1. Representative oil-flow results and their topologicalinterpretations

Considering that the flows near the blade suction surface arevery complex and the 3D flow separations in the stator passageare usually appear firstly herein, the pictures of the oil streaks onthe blade suction surface can visualize these flow structuresclearly. Representative oil-flow visualization results on the bladesuction surface at the tested four compressor operating conditionsare shown in Fig. 3. Based on the topological analyses of the oil pic-tures, including the suction surface, pressure surface, hub wall and

(a) design condition

(c) max. static pressure rise condition

Fig. 3. Representative oil-flow res

casing wall results (not shown here for brevity), the flow visualiza-tion results can be interpreted with detailed critical points.

According to the topological theory of three-dimensional steadyflow patterns introduced by Perry et al. [17,18] and Surana et al.[19,20], along a separation or attachment line, complex criticalpoints may appear, but the saddle type point and node type pointshould appear alternatively. The separation line, usually appearingas a converged limiting streamline, has four possible types in phys-ical fluid flows, i.e. (1) saddle–stable spiral connections; (2) saddle–stable node connections; (3) saddle–stable limit cycle connectionsand (4) stable limit cycles. The attachment line, usually appearingas a diverged limiting streamline, has also four possible types inphysical fluid flows, i.e. (1) unstable spiral–saddle connections;(2) unstable node–saddle connections; (3) unstable limit cycle–saddle connections and (4) unstable limit cycles.

For the oil flow results, some critical points may not be distin-guished clearly for the limited spatial resolution. Hence, the exacttopological interpretations of them should be deduced from the oilpatterns with the help of the above mentioned topological theory.Figs. 4–7 show the results of the topological interpretations of theoil-flow visualization photos obtained at different compressoroperating conditions. In these figures, the variation of the position,scale and shape of the 3D separations can be seen easily. Hence, theevolutionary processes of the 3D separations in the stator passagecan be established along the compressor operating line.

3.2. Evolution of three-dimensional separations along compressoroperating line

According to the topological interpretations of the oil-flow visu-alization results shown in Figs. 4–7, the evolutionary processes ofthe 3D separations, particularly for the variation of the scale ofthe separation patches in the streamwise, spanwise and circumfer-ential directions, in the tested stator passage can be establishedalong the compressor operating line.

At the DE condition, see in Figs. 3a and 4, corner separationflows appear at both the hub and casing endwall regions nearthe blade trailing edge. In Fig. 4, the separation lines SL1 and SL2bound the suction side near casing 3D corner separation and the

(b) middle condition

(d) near-stall condition

ults on blade suction surface.

Fig. 4. Topological interpretations of the oil-flow visualization results at the designcondition.

Fig. 5. Topological interpretations of the oil-flow visualization results at the middlecondition.

Fig. 6. Topological interpretations of the oil-flow visualization results at themaximum static pressure rise condition.

Fig. 7. Topological interpretations of the oil-flow visualization results at the near-stall condition.

304 X. Yu et al. / Experimental Thermal and Fluid Science 44 (2013) 301–311

separation lines SL3 and SL4 bound the suction side near casing 3Dcorner separation. It should be noted that because of the hub clear-ance of the cantilevered stator blade, the hub leakage flow inter-acts with the 3D hub corner separation flow, resulting complextopological flow patterns with the appearing of attachment lineAL1 and a saddle point S4. Similar limiting streamline flow pat-terns can also be seen in the computational results obtained byGbadebo et al. [21]. Moreover, it can be seen clearly in Fig. 4 thatthe suction side near casing 3D corner separation is just a very thinlayer adhere to the blade suction surface since SL2 is very close tothe blade suction surface. As for the near hub corner separation,because of the interaction of the hub tip leakage flows, SL4 is faraway from the blade suction surface. Hence, the scale of the nearhub corner separation in the circumferential direction cannot be

estimated base on the surface flow patterns. However, with theaid of the SPIV measurement results shown in Fig. 8, it can be con-cluded that both the two endwall corner separations are very weakand the blockage and losses caused by them can even be neglected.

As the mass-flow-rate decreases, at the MID condition, see in Figs.3b and 5, the two endwall corner separations enlarge and link to-gether, forming a whole-span trailing edge separation. The blademid-span separation results two separation lines SL5 and SL6 origi-nated from the same saddle point S5. Blade surface mid-spanseparation usually caused by the over limit blade loading or too largeblade incidence angle. For low speed condition, the flow separation

Fig. 8. Combined maps of ensemble-averaged measured results inside the stator passage at the design condition.

Fig. 9. Combined maps of ensemble-averaged measured results inside the stator passage at the near-stall condition.

X. Yu et al. / Experimental Thermal and Fluid Science 44 (2013) 301–311 305

usually grows gradually as the mass-flow-rate decreases. Thus tosay, at the MID condition, although the separation flows cover a largeportion of the blade suction surface, the flow blockage and lossescaused by them may still not very significant.

For the MSPR condition, see in Figs. 3c and 6, the topology ofseparation flow near the blade suction surface is much differentwith that at the previous two conditions. The separation region ini-tials much closer to the blade leading edge and appears as twopatches. One is a large patch spreading from the hub endwall tothe near casing side, bounded by SL4, SL10, SL6 and SL5 in Fig. 6.The other one is a small patch located at the blade leading edgenear casing corner, bounded by SL1 and SL7 in Fig. 6. The

separation line SL10 locates just before the leading edge of theblade near hub end, which indicates the spillage of the tip leakageflow and flow stall occurring herein, the same as that observed bySaathoff and Stark [22] from the oil flow pictures at the endwall ofa cascade. Thus to say, the first separation patch contains hub stallflows, the flow blockage and losses caused by it should be veryserious. The second separation patch bounded by SL1 and SL7starts from the near casing blade leading edge, indicating that itshould be caused by locally large incidence angle. Considering thatat small mass-flow-rate conditions the upstream rotor tip leakageflow and corner vortex are very strong and can cause a mass of lowmomentum flow stacking at the rotor tip region [15], the

Fig. 10. The evolution of the three-dimensional separation flows inside the stator passage along the compressor operating line.

306 X. Yu et al. / Experimental Thermal and Fluid Science 44 (2013) 301–311

significant increase of the local incidence angle just near the casingendwall is reasonable.

At the NS condition, see in Figs. 3d and 7 and compare Figs. 6and 7, the topology of separation flow is nearly the same as thatat the MSPR condition, however, the scale of the separation regionschanges a little. At the NS condition, the two separation patchesappearing at the MSPR condition nearly link together resulting inthe degradation of the two spiral nodes patterns, F1 and F4 shownin Fig. 6, to a single spiral node pattern, F1 shown in Fig. 7. The SPIVresult shown in Fig. 9 confirms this conjecture from the oil-flowtopology analysis. The near casing corner separation originates atabout 7% chord position and ends at about 50% chord position inthe streamwise direction. In the spanwise direction, it covers aboutthe near casing 20% span region. For the near hub large scale cornerstall, it covers about the after 50% chord in the streamwise direc-tion and the near hub 80% span region in the spanwise direction.

Based on the above analyses, the evolutionary process of the 3Dseparation flows in the tested stator passage can be drawn in thefigure shown in Fig. 10. Generally speaking, as the mass flow ratedecreases along the compressor operating line, the scales of the3D separations are enlarged in both the axial and circumferentialdirections. Although 3D separations are very common in compres-sors, they may cause no significant detrimental effects, such assignificant flow blockage and losses, unless flow stall occurs as thatconcluded by Lei et al. [13] and Yu and Liu [14]. At the DE and MIDconditions, the separation flows are mild and has no significantinfluence on the performance of the compressor. However, thehub corner separation augments suddenly and the hub corner stalloccurs since the MSPR condition, which cause significant detri-mental effects of the flow inside the stator passage. As mentionedabove, also see in Fig. 10, the topology of the flow separation isnearly the same at the MSPR condition and the NS condition; how-ever, comparing the Fig. 3c and d, one can see that the separationline of the near hub corner separation, i.e. the front edge of the re-verse flow region located at the after part of the blade suction

surface, is much clearer at the NS condition than that at the MSPRcondition. It may indicate that the large scale hub stall flows at theMSPR condition is unstable and its originating position wanderssignificantly. Hence, the hub corner stall at the MSPR condition isindicated as unsteady separation in Fig. 10.

3.3. Topological analyses of typical 3D flow structures

It is well known that the topological analysis of the limitingstreamlines is helpful for the reconstruction of the 3D flow struc-tures [17–20] and has been widely used for studying complexflows in turbomachinery [5,10,23,24]. In the above section we havediscussed the evolutionary process of 3D separation flows in thetested stator passage. In this section we will reconstruct the exact3D flow structures of the 3D separation flows, which is helpful forrealizing the evolution mechanics of them.

See in Figs. 4–7, the topology structures are very complex in thetested stator passage, particularly on the blade suction surface.Generally speaking, as the mass-flow-rate decreases the topologyof the 3D separation flows becomes much more complicated. How-ever, according to the regulars of the distribution of the criticalpoints and the development of the skin-friction lines, it can befound that the complex 3D flow structures in the stator passagecan be categorized into the following four types of flow:

(1) The flow of the corner vortex (flow type A, as shown inFig. 11). Here the corner vortex is a streamwise vortex. This typeof flow structures can usually be seen in turbomachines for thedeflection of the vorticity in endwall boundary layer in a curvedpassage [25]. Although, for axial turbomachinery, they oftenappear as the so called horseshoe vortices in turbine, it seems thatthey are rarely observed in compressors. However, in the presenttests, this type of flow can be seen at all of the four tested condi-tions owing to the existence of the hub leakage flow and theappearance of the near leading edge suction surface casing corner

Fig. 13. Flow topology of the corner separation without ring-like vortex (flowtype C).

Fig. 12. Flow topology of the corner separation with ring-like vortex (flow type B).

Fig. 11. Flow topology of the corner vortex (flow type A).

X. Yu et al. / Experimental Thermal and Fluid Science 44 (2013) 301–311 307

separation. The appearance of the limiting streamlines for the flowtype A can be seen clearly in Fig. 11. In this figure, see from thedownstream to the upstream, the corner vortex is a counter-clock-wise streamwise vortex, the limiting streamlines on the normalsurface have the tendency to approach the horizontal surface andthen they propagate away from the normal surface on the horizon-tal surface. In actual, there also should have an attachment line onthe normal surface and a separation line on the horizontal surface.It is obvious, if the corner vortex is a clockwise streamwise vortex,the evolution tendencies of the limiting streamlines on the twosurfaces are correspondingly contrary, and a separation line willappear on the normal surface and a reattachment line will takeplace on the horizontal surface. As can be seen in Fig. 4 andFig. 5, the group of limiting streamline of AL1 and SL4 indicatesthe hub leakage caused suction side near hub corner vortex andthe group of limiting streamline of AL3 and SL4 indicates the pres-sure side near hub corner vortex , which may induced by the suc-tion side near hub corner vortex. See in Figs. 6 and 7, the near hubtwo corner vortices still exist, however, because of the occurrenceof hub stall, their topological appearances become much compli-cated. The suction side near hub corner vortex is bounded byAL1, AL5, SL10, SL9 and SL4, and the pressure side one is boundedby SL9, SL4 and AL3. Moreover, for the near casing two corners,only one corner vortex, bounded by AL4 and SL8 and might beinduced by the suction side near casing leading edge separation,appears at the pressure side corner.

(2) The flow of the corner separation with ring-like vortex (flow typeB, as shown in Fig. 12). This three-dimensional flow structure forcorner separation was first derived by Schulz et al. [5] and has beenproved by other researchers by numerical simulations [3,7,13,26].The appearance of the limiting streamlines for the type B flowcan be seen in Fig. 12. The separation lines on the two surfacesof a corner start from the same saddle point and end with spiralnodes (focuses). For this type of corner flow separation, the separa-tion stream surface curls up as a ring vortex, which straddles onthe two surfaces at the spiral nodes. This type of flow appears atall the tested compressor operating conditions except the MID con-dition. At the DE condition, see in Fig. 4, at the suction side nearcasing corner, the topological interpretations of SL1-F1 and SL2-F2 indicate the existence of the flow type B herein. As for theMID condition, the topological interpretations of SL1-F4 and SL7-F2 shown in Fig. 6 also indicate that the suction side near casingleading edge separation should appear as flow type B. At the NScondition, see in Fig. 7, the suction side near casing leading edgeseparation still appear as flow type B, and its topological interpre-tations consist with SL1-F1 and SL7-F2.

(3) The flow of the corner separation without ring-like vortex (flowtype C, as shown in Fig. 13). This type of flow is the evolution offlow type B. Only one of the two separation lines ends with a spiralnode, another one flows downstream directly. In the experimentsconducted by Schulz et al. [5], they did not observe this type offlow. However, according to the numerical simulation results givenby Hah and Loellbach [7] and Weber et al. [26], the corner separa-tion of flow type C may occur when the back pressure decreasesbelow a certain value. Thus to say, the flow topology of the cornerseparation may appear as flow type B or C, depending on the com-pressor operating condition and the flow interactions inside theflow passage. See in Fig. 5, at the MID condition, because of theappearance of the mid-span separation, the end of SL1 spirals intothe mid-span separation region. As a result, the spiral flow origi-nated from F2 has to flow downstream directly, forming a suctionside near casing corner separation without ring-like vortex.

(4) The flow of the blade surface flow separation (flow type D, asshown in Fig. 14). The topologies of the limiting streamlines forthe 3D separations on blade surface (or near the mid-span region)always appear as the flow type D shown in Fig. 14. It can be seenthat the separation line of this type of flow always starts from a

Fig. 14. Flow topology of the blade surface flow separation (flow type D).

308 X. Yu et al. / Experimental Thermal and Fluid Science 44 (2013) 301–311

saddle point and ends at a spiral node. As can be seen in Figs. 5 and6, the flow separation on the blade suction surface occurs behindtwo separation lines, SL5 and SL6, starting from the same saddlepoint, S5, and terminating at two spiral nodes, F1 and F3 respec-tively. This type of flow separation is usually called as closed sep-aration and has very stable flow topology and 3D flow structure asshown in Fig. 14 [27]. The separation stream surface departs fromthe blade surface at the separation line and curls up into a stream-wise vortex incepting at the spiral node at the end of the separa-tion line. Thus to say, in Figs. 5 and 6, a pair of counter-rotatingstreamwise vortices should be observed downstream of the statornear the mid-span region. For the NS condition, see Fig. 7, becauseof the interaction of the blade surface flow and the near casingleading edge 3D separation, the two spiral nodes topology F1 andF4 degrades into the single spiral node topology F1, indicatingthe link of the two separation patches as mentioned in Section3.2. Hence, only a single streamwise vortex, originating from F3,can be observed downstream of the stator near the mid-spanregion.

3.4. Reconstruction of 3D flow structures inside the stator passage

According to the analyses in the above section, the 3D flowstructures in the stator passage can be reconstructed. The 3D flowstructures in the stator passage at the DE condition are depicted inFig. 15. It can be seen that according to the existence of the hub

Fig. 15. 3D flow structures at the design condition.

leakage flow, both the pressure side and the suction side hub cor-ner exist a corner vortex, i.e. flow type A. It seems that the forma-tion mechanisms of these two corner vortex are not the same asthat of the so called horseshoe vortex observed often in turbines.Because of the relative thin leading edge of the compressor blade,the horseshoe vortex formed at the leading edge of the compressorblade is so weak that dissipated soon after it propagates into theblade passage. However, due to the appearance of the hub leakageflow the suction side branch of the horseshoe vortex may mergewith the hub leakage vortex and thus a large scale suction side cor-ner vortex can even be seen at the outlet of the blade passage.Moreover, also owing to the hub leakage flow, the circumferentialmigration of the endwall secondary flow (flow from the passagepressure side to the suction side) may be blocked at about the mid-dle of the blade passage. As a result the endwall secondary flowwill uplift there and intensifies the pressure side branch of thehorseshoe vortex. This may be the reason for the formation ofthe pressure side near hub corner vortex. As also can be seen inFig. 15, 3D corner separation appears at both the near casing andnear hub suction side corners, however, the topologies of thesetwo separation structures are not the same. The corner separationnear the casing has a topology of flow type B, while it appears asflow type C at the near hub corner.

The reconstructed 3D flow structures at the MID condition areshown in Fig. 16. It can be seen that besides the flow structures ob-served at the DE condition a closed 3D separation (flow type D) ap-pears on the blade suction surface near the blade trailing edge,resulting in the degradation of the two corner separations at theblade trailing edge. It seems that the near hub corner separationdisappears (it may be still exists but very weak, since the topologyof the skin-friction lines here are not very clear in the oil-flow visu-alization results) and the near casing corner separation diminishesand changes from the flow type B to the flow type C.

At the MSPR condition, the 3D flow separations intensify con-tinually, as can be seen in Fig. 17. The closed separation on theblade suction surface enlarges in both the spanwise and thestreamwise directions. Because of its spanwise enlargement, thesqueeze effect causes the thorough disappearing of the trailingedge casing corner separation. As a result, the blade suction surfaceseparation extends from the near hub position to about 75% spanposition, forming a very large scale of suction side near hub cornerseparation at the aft 2/3 blade passage (see in Fig. 10). It should benoted that, as the trailing edge casing corner separation disappears,

Fig. 16. 3D flow structures at the middle condition.

Fig. 19. Ensemble-averaged SPIV results for the cross planes near the suction surface measured at the near-stall condition (cross-plane streamlines with the contour of thestreamwise velocity).

Fig. 18. 3D flow structures at the near-stall condition.Fig. 17. 3D flow structures at the maximum static pressure rise condition.

X. Yu et al. / Experimental Thermal and Fluid Science 44 (2013) 301–311 309

Fig. 20. Ensemble-averaged SPIV results for the cross planes near the pressure surface measured at the near-stall condition (cross-plane streamlines with the contour of thestreamwise velocity).

310 X. Yu et al. / Experimental Thermal and Fluid Science 44 (2013) 301–311

a newly formed corner separation appears at the suction side cas-ing corner near the blade leading edge and has the topology of flowtype B. As mentioned above, this flow separation may be caused bythe accumulating of significant low momentum flows in the up-stream rotor near the casing and the corresponding increasing ofthe local incidence angle of the stator blade. Obviously, as the airflows downstream of the leading edge casing corner separation,the ring-like vortex will be tipped and stretched. Since the flowvelocity is lower near the casing wall than that away from casingwall, the ring-like vortex will finally change into a clockwisestreamwise vortex (see from the outlet to the inlet) and form thesuction side near casing corner vortex, which can be seen at theaft of the blade passage as shown in Fig. 17. Meanwhile, becauseof theory of the conservation of the circulation, a counter-rotatingcorner vortex should appear at the pressure side casing corner.

Fig. 18 shows the 3D flow structures in the stator passage at theNS condition. As mentioned before, it seems that the flowtopologies of the 3D flow structures at the NS condition are nearlythe same as that at the MSPR condition. However, because of theinteraction of the blade surface flow and the near casing leadingedge 3D separation and the degradation of the critical pointsaround spiral node F1, only a single streamwise suction surfaceseparation vortex close to the hub of the blade passage canbe seen.

3.5. Validation of the 3D flow structures by SPIV results

In order to validate the above analyses for the reconstruction ofthe 3D flow structures inside the stator passage, the measurementresults obtained by SPIV at the NS condition are shown in Figs. 19and 20. As can be seen in these figures, the leading edge suctionside near casing corner separation and the large scale suction sidenear hub corner separation can be seen clearly. Obviously, thescales, positions and configurations of these separation flows arenearly the same as that deduced from the oil-flow visualization re-sults analyzed in Section 3.2. Moreover, the detailed 3D vortex flowstructures of the suction side near casing corner vortex, pressureside near casing corner vortex and the suction surface separationvortex deduced based on the topological interpretations in Section3.3 are also can be seen clearly in the SPIV results. Thus to say, theanalyses in the above several sections should be reasonable andreliable.

4. Conclusions

By using the oil-flow visualization technique and the topologi-cal analysis, the three-dimensional flow structures, particularlyfor the separation flows, are investigated in a low-speed large-scalecompressor test facility at four operating conditions with the mass

X. Yu et al. / Experimental Thermal and Fluid Science 44 (2013) 301–311 311

flow coefficient of 0.58 (design condition), 0.51 (middle condition),0.45 (maximum static pressure rise condition), and 0.39 (near-stallcondition). The evolutionary processes of the 3D separation flowsalong the compressor operating line are established and the exact3D flow structures are reconstructed and their flow mechanismsare analyzed. Finally, in order to prove the conclusions obtainingbased on the topological analyses of the oil-flow visualization re-sults, detailed results measured by SPIV at the NS condition areused for the validation. From the above studies, the following con-clusions can be drawn:

(1) Corner separations can even be seen at the design conditionin the stator passage of the tested compressor, however, theflow blockage caused by them are very weak. As the massflow rate decreases, the flow separations in the tested statorpassage become much more significant and complex. At themaximum static pressure rise condition and the near-stallcondition, the large scale near hub corner separation/stallappears, and cause much more significant flow blockage.

(2) There are four basic types of flows in the tested stator pas-sage: the corner vortex flow; the corner separation withring-like vortex; the corner separation without ring-like vor-tex; and the blade surface separation flow (usually appear asa closed separation). It seems that no matter how complexthe flows in the tested stator passage are, they always con-sist with part or all of these four types of flows.

(3) The hub leakage flow dominates the flows near the hub of thetested stator passage and may cause the formation of the twohub corner vortices. However, the influences of the hub leak-age flow to the near hub secondary flows and the corner sep-arations are still not very clear and needed further studies.

(4) The rotor–stator interactions also markedly affect the evolu-tionary processes of the 3D flow structures in the tested sta-tor. The accumulation of significant low momentum flows atthe upstream rotor tip after the maximum static pressurerise condition causes a locally increase of the incidence angleat the stator blade tip and results in the leading edge suctionside near casing corner separation.

(5) The measured results at the near-stall condition obtained bySPIV were used for validating the topological analyses basedon the oil-flow visualization results in the paper. The resultsshow that the topological analyses and the reconstructed 3Dflow structures in the tested stator passage are reliable.

Acknowledgments

The authors would like to acknowledge the support of NationalScience Foundation of China, Grant Nos. 50976009 and 51006007.

References

[1] D.H. Joslyn, R.P. Dring, Axial compressor stator aerodynamics, ASME Journal ofEngineering for Gas Turbines and Power 107 (1985) 485–493.

[2] N.M. McDougal, A comparison between the design point and near-stallperformance of an axial compressor, Journal of Turbomachinery 112 (1990)109–115.

[3] V.-M. Lei, A Simple Criterion for Three-dimensional Flow Separation in AxialCompressors, PhD Thesis, Department of Aeronautics and Astronautics, MIT,2006.

[4] H.D. Schulz, H.E. Gallus, Experimental investigations of the three-dimensionalflow in an annular compressor cascade, Journal of Turbomachinery 110 (1988)467–478.

[5] H.D. Schulz, H.E. Gallus, B. Lakshminarayana, Three-dimensional separatedflow field in the endwall region of an annular compressor cascade in thepresence of rotor–stator interaction. Part I. Quasi-steady flow field andcomparison with steady-state data, Journal of Turbomachinery 112 (1990)669–678.

[6] H.D. Schulz, H.E. Gallus, B. Lakshminarayana, Three-dimensional separatedflow field in the endwall region of an annular compressor cascade in thepresence of rotor–stator interaction. Part II. Unsteady flow and pressure field,Journal of Turbomachinery 112 (1990) 679–690.

[7] C. Hah, J. Loellbach, Development of hub corner stall and its influence on theperformance of axial compressor blade rows, Journal of Turbomachinery 121(1999) 67–77.

[8] S.A. Gbadebo, N.A. Cumpsty, T.P. Hynes, Three-dimensional separations in axialcompressors, Journal of Turbomachinery 127 (2005) 331–339.

[9] M. Choi, J.H. Baek, Role of the hub-corner-separation on the rotating stall, in:42nd AIAA/ASME/SAE/ASEE Joint Propulsion Conference & Exhibit,Sacramento, California, 2006.

[10] H.L. Zhang, S.T. Wang, Z.Q. Wang, Variation of vortex structure in a compressorcascade at different incidences, Journal of Propulsion and Power 23 (2007)221–226.

[11] C. Mertz, A Study of the Effect of Boundary Layer Control on an Axial FlowCompressor Stage, in, Aeronautical Engineer Thesis, California Institute ofTechnology, 1954.

[12] J.H. Horlock, P.M.E. Percival, J.F. Louis, B. Lakshiminarayana, Wall Stall inCompressor Cascades, in: ASME paper 61-WA/FE-29, 1961.

[13] V.-M. Lei, Z.S. Spakovszky, E.M. Greitzer, A criterion for axial compressorhub-corner stall, Journal of Turbomachinery 130 (2008) 031006 (031010pages).

[14] X.J. Yu, B.J. Liu, A prediction model for corner separation/stall in axialcompressors, in: Proceedings of ASME Turbo Expo 2010: Power for Land, Seaand Air, Glasgow, UK, 2010.

[15] X.J. Yu, B.J. Liu, Stereoscopic PIV measurement of unsteady flows in an axialcompressor stage, Experimental Thermal and Fluid Science 31 (2007) 1049–1060.

[16] L.C. Squire, The motion of a thin oil sheet under the steady boundary layer on abody, Journal of Fluid Mechanics 11 (1961) 161–179.

[17] A.E. Perry, M.S. Chong, A description of eddying motions and flow patternsusing critical point concepts, Annual Review of Fluid Mechanics 19 (1987)125–155.

[18] J.M. Delery, Robert Legendre, Henry Werle, Toward the elucidation of three-dimensional separation, Annual Review of Fluid Mechanics 33 (2001) 129–154.

[19] A. Surana, O. Grunberg, G. Haller, Exact theory of three-dimensional flowseparation. Part 1: steady separation., Journal of Fluid Mechanics 564 (2006)57–106.

[20] A. Surana, G.B. Jacobs, G. Haller, Extraction of separation and attachmentsurfaces from three dimensional steady shear flows, AIAA Journal 45 (2007)1290–1302.

[21] S.A. Gbadebo, N.A. Cumpsty, T.P. Hynes, Interaction of tip clearance flow andthree-dimensional separations in axial compressors, Journal ofTurbomachinery 129 (2007) 679–685.

[22] H. Saathoff, U. Stark, Tip clearance flow induced endwall boundary layerseparation in a single-stage axial-flow low-speed compressor, in: Proceedingsof ASME TURBOEXPO 2000, Munich, Germany, 2000.

[23] S. Kang, C.H. Hirsch, Experimental study on the three-dimensional flow withina compressor cascade with tip clearance. Part I – velocity and pressure fields,Journal of Turbomachinery 115 (1993) 435–443.

[24] D. Nerger, H. Saathoff, R. Radespiel, V. Gummer, C. Clemen, Experimentalinvestigation of endwall and suction side blowing in a highly loadedcompressor stator cascade, Journal of Turbomachinery 134 (2012) 021010.

[25] E.M. Greitzer, C.S. Tan, M.B. Graf, Internal Flow: Concepts and Applications,Cambridge University Press, Cambridge, UK, 2004.

[26] A. Weber, H.-A. Schreiber, R. Fuchs, W. Steinert, 3-D transonic flow in acompressor cascade with shock-induced corner stall, Journal ofTurbomachinery 124 (2002) 358–366.

[27] J.Z. Wu, H.Y. Ma, M.D. Zhou, Vorticity and Vortex Dynamics, Springer-Verlag,Berlin, 2006.