Impact of Operating Conditions and Axial Casing Grooves on the Evolution of Flow Structure Across Blade Rows in an Axial Compressor

The past few decades have seen growing attention in the turbomachinery community to improvements in the performance and efficiency of axial compressors. One central focus of the research has been understanding the mechanisms involved with the onset of stall [1,2] and the development of methods to extend the stall margin [3]. Numerous studies, experimental as well as computational, have identified flow structures and phenomena that are linked with flow instabilities when the machine operates near its stall point [4–12]. For example, März et al. [4] show that a “rotating instability vortex” can form near the leading edge (LE) plane of a rotor and could exist even when the compressor is stable. Inoue et al. [5] propose a model involving a tornado-like separation vortex that coexists with the tip leakage vortex. Hoying et al. [6] suggest that the tip clearance vortex moves forward of the blade row as the mechanism for the development of short-length scale disturbances. Vo et al. [7] claim that spillage of tip clearance flow ahead of rotor LE is one of the necessary criteria for the initiation of spike-type stall. Similar phenomena are described by Pullan et al. [8], Hewkin-Smith et al. [9], and Cameron et al. [12]. In a recent study, Eck et al. [10] argue that the pre-stall disturbances seen in their compressor exist at all flowrates as rotating instabilities and that they are not caused by LE separation. Chen et al. [11] show that at pre-stall, secondary vortical structures such as backflow vortices (BFVs) are formed under the tip leakage vortex (TLV) due to the interaction of tip leakage flow with the main passage flow. These structures extend diagonally upstream toward the LE of the next blade and occasionally spill over in the next passage either by circumventing the leading edge or by going through the tip gap.

With this wide yet incomplete understanding of phenomena that could lead to the onset of stall, one approach that has gained significant traction for delaying the onset of stall is passive flow control using endwall casing grooves [13–15]. Included are two classes of configurations, namely circumferentially continuous [16–18] and discrete axial casing grooves (ACG). The common understanding is that the grooves improve the performance by, e.g., reducing the extent of low axial momentum near the casing either by transporting high momentum fluid or by engulfing flow structures that cause blockage. Out of the two designs, the axial casing grooves (ACGs) have shown greater improvements in stall margin, but this gain often comes with a penalty in peak efficiency at higher flowrates [19–21]. Hence, considerable research has been carried out to understand the impact of casing grooves as well as to optimize their location with respect to the blades, their relative size, and shape [16,18,21].

Despite the comprehensive literature, for several reasons, the knowledge about the flow field at near-stall flowrate and the flow modulation by the grooves is still limited. In numerical simulations, Reynolds-averaged Navier–Stokes (RANS) equations inherently do not resolve the turbulence and associated instabilities of near-stall flow, especially for large tip gaps [22–24]. Improved predictions are likely to be obtained using large Eddy simulations (LES) [25]. Both approaches require experimental validation data. Because of limited access, experimental studies have mostly consisted of measurements of performance, wall pressure distributions [10], velocity in places where probes could be inserted [26], and recently, particle image velocimetry (PIV) through windows installed in the casing [27,28].

The Johns Hopkins University (JHU) refractive index-matched facility provides the unique ability to perform PIV measurements with unobstructed optical access. Past studies in a one-and-half-stage axial compressor have examined the flow under multiple operating conditions, with and without casing treatments of varying geometries. These studies reveal the role of BFVs in the onset of stall [11], follow the evolution of TLV, characterize the turbulence in multiple machines, and study the effect of ACGs on the flow structures [29,30]. However, the focus has been predominantly on the tip region of the rotor. This paper builds upon the previous studies through characterization of flow structures in the entire span and blade passage of the machine and tracks their evolution axially across two blade rows. The measurements have been performed at two operating conditions, namely, at the pre-stall flowrate and near the best efficiency point (BEP) of the untreated casing. Comparison with the meridional plane data from previous studies enables us to identify the role of the leakage flow, as well as backflow and tip vortices in generating reverse flow in the tip region. Measurements performed between and within blade rows show the radial expansion of the blockage with axial distance in the rotor and demonstrate that the stator acts as a flow homogenizer. The effect of semicircular ACGs on the flow structure is studied under both operating conditions. A previously unobserved mechanism generated by the ACGs that enhances the mixing of near-casing slow flow with the higher momentum fluid in the middle of the span, which reduces the tip blockage, is elucidated. Furthermore, passage-averaged velocity profiles highlight that the grooves modify the flow over the entire span, not only in the tip region. The setup and procedures are presented in Sec. 2. The results are shown next, followed by a discussion and conclusions highlighting the present findings.

The experiments are performed in a one-and-half-stage axial compressor which is derived from the low-speed axial compressor (LSAC) present at NASA Glenn Research Center. The compressor, consisting of a 20-blade inlet guide vanes (IGV), a 15-blade rotor, and a 20-blade stator, is installed in the JHU refractive index-matched facility. The top nine blades of the IGV and the stator row and all 15 rotor blades are made from acrylic whose refractive index is matched with the aqueous Sodium iodide solution that is used as the working fluid. The facility provides us with unobstructed optical access for PIV and flow visualizations. The relevant geometric and kinematic parameters of the compressor and the facility are provided in Table 1. More information regarding the facility can be found in Refs. [31–33].

Recently, we have recalibrated the performance curve of the compressor with more accurate flowrate measurements in the facility [34]. The updated performance curve is shown in Fig. 1. The pre-stall point of the compressor with untreated casing is at φ = 0.28, where the flow coefficient is defined as φ = V_z/U_T. It is the lowest flowrate at which the compressor can be operated in stable condition, and at lower flowrates, the pressure rise across the machine starts to drop and its operation is accompanied by vibration and noise. The BEP of the compressor is at φ = 0.38.

Past studies on this compressor have found that installing semicircular ACGs around the leading edge of the rotor increases the stall margin and peak pressure rise by 40% and 32%, respectively. However, they cause a reduction of 2.4% in peak efficiency [30]. The rotor-groove configuration and the dimensions are highlighted in Fig. 2. The downstream one-third end of the grooves overlaps with the rotor tip. There are in total 60 grooves (four per rotor-blade passage) and are skewed at 45 deg in the direction of rotation of the rotor to facilitate flow into the grooves.

To study the evolution of the flow structures throughout the machine, stereo-PIV (SPIV) measurements have been performed in a series of axially aligned planes that cover the entire passage and span and are distributed throughout the machine. The location of the planes is provided in Table 2. The SPIV setup is shown in Fig. 3 along with the axial planes marked on a sketch of the compressor blade rows. We have used a dual pulse, 200 mJ/pulse Nd:YAG laser for illumination, and a pair of 29 MP (6600 × 4400 pixel array) Imperx B6640 charge-coupled device (CCD) cameras for imaging. The entire system is synchronized with a shaft encoder which allows for data to be acquired at any desired rotor-blade orientation. The time delay between the two exposures is kept between 40 and 60 µs depending on the operating condition and measurement location. The time delay is chosen to restrict the motion of the particles to below 40% of the window size. For seeding, silver-coated hollow glass spheres are used whose mean diameter is 13 µm. In addition to multiple planes, data are acquired at multiple rotor-blade orientations for each case. Finally, phase-locked, ensemble-averaged velocity components are calculated over 2000 instantaneous realizations for planes in between the blade rows and 1000 instantaneous realizations for planes cutting through the rotor and the stator. The camera system is mounted on a motorized slide and is calibrated in a two-step process. First, the entire system is traversed upward to image a dotted target kept in a box filled with the same solution as the facility for a coarser calibration. This is done to mimic the actual imaging conditions. Next, the PIV data itself are used for self-calibration on particle images [35] for finer calibration. The PIV images are preprocessed with background subtraction and a modified histogram equalization algorithm as described in Ref. [36]. PIV processing has been done using the commercial software package LaVision DaVis^TM. The velocities are calculated using multi-pass cross-correlation, starting with an interrogation area of 64 × 64 pixels, and ending with 32 × 32 pixels, which corresponds to 1.1 × 1.1 to 1.2 × 1.2 mm², varying slightly across axial planes. With a 50% overlap between adjacent windows, the vector spacing is 0.53–0.64 mm. Consequently, the present measurements do not resolve the velocity gradients in the inner part of boundary layers on surfaces, and the first vector shown near boundaries represents the spatially averaged velocity up to a distance of 1.1–1.2 mm away from the wall. For meridional plane data focusing on the tip region, shown for comparison, the vector spacing is 0.16 mm. The universal outlier detection algorithm described in Ref. [37] is used during vector post-processing to get rid of spurious vectors.

In the return line of the facility, the velocity profile in a horizontal and a vertical plane is measured by translating a pitot tube across the entire diameter of the pipe. The flowrate is then determined by interpolating the profile over the entire area. The uncertainty in this estimation is within 2%, as estimated by comparing several interpolation methods [34]. The uncertainty in the instantaneous velocity measurement is 0.1–0.2 pixels provided there are 4–5 particles in most of the interrogation windows [38]. The typical particle displacement is 10–12 pixels, resulting in uncertainty of 2%. The uncertainty in the ensemble-averaged velocities calculated using 1000–2000 realizations is an order of magnitude lower.

The axial velocity is normalized by annular area-averaged axial velocity (V_z) of that operating condition. V_z can also be directly calculated as flow coefficient times the blade tip speed. The choice of this normalization enables us to compare the blockage in the compressor under various operating conditions. On the other hand, the circumferential and radial velocities are normalized by the blade tip speed. Measurements in the A1 plane are used to characterize the inflow of the machine. At both operating conditions, the inflow is uniform in the mid-passage region with a thick boundary layer already developed on both casing walls [34]. In the A1 plane, the inlet flow turbulence intensity outside of the casing boundary layer is 4% of V_z at pre-stall and 3% near BEP.

This sub-section discusses the flow distribution in the entire machine at the pre-stall condition of the untreated casing (φ = 0.28). Starting with the flow upstream of the rotor (A2 plane), Fig. 4 compares the distributions of normalized ensemble-averaged axial velocity, circumferential velocity, and axial vorticity for the untreated casing and the ACGs for rotor-blade orientation Θ = 0 deg. For discussion, the circumferential position of the rotor LE, which is located downstream of this plane, is marked on the figure. We begin by identifying the flow features for the untreated casing (first row, Fig. 4). The axial velocity contours show that the wake of the IGVs, the entire region near the casing, and the entire span in front of the blade has low axial momentum whereas the mid-span region between the rotor blades has higher axial velocity. Owing to its proximity, the blade causes the blockage in front of it and this particular region moves circumferentially with the rotor blade. The circumferential velocity is higher on the pressure side (PS) of the rotor blade than on the suction side (SS), peaking near the hub on the pressure side. The latter region is observed for all four orientations recorded and moves with the rotor blade. However, it becomes distorted when the blade is in the vicinity of the IGV wake, such as for the shown orientation. The distribution of axial vorticity also has the in-plane velocity vectors superimposed on the color contours. These vectors are diluted 2:1 in the radial direction for clarity. The notable features are the clear signature of the IGV wake with the high negative peak in the middle of the passage, and slight negative layers near the casing and the hub.

The second row shows the same distributions with the semicircular grooves installed. In this case, the A2 plane cuts the grooves near their upstream end as shown in Fig. 2. Previous studies [29,30] on this compressor involving SPIV measurements in multiple meridional and radial planes have revealed how the flow structures change in the tip region when the semicircular ACGs are installed. At φ = 0.28, the grooves ingest a part of the TLV from their downstream end when the pressure side of the rotor blade is aligned with the groove inlet. Consequently, they also inhibit the formation of BFVs and curtail their movement from one blade to the next. Figure 4(d) shows that the axial velocity distribution inside the grooves in the illuminated plane is mostly negative, as expected. Near the casing, in the region where the groove outflow occurs, the axial velocity is low, but it is high in the space between the grooves. As the compressor is operated at the same flow condition, the negative axial flow inside the grooves is accompanied by an elevated axial velocity deep in the passage in comparison to that of the untreated casing. Owing to the groove orientation, the outflow occurs in the negative circumferential direction, creating a layer of negative circumferential velocity (U_θ < 0) along the tip region (top 20% of the passage). However, U_θ in the rest of the span remains unaffected. Since the main passage flow below this layer has a circumferential momentum in the direction of the blade rotation, there are significant gradients of circumferential velocity in this region; hence, a layer with elevated negative vorticity exists. In addition, the flow jetting out of the grooves generates a pair of axially aligned vortices on the corner of the grooves. While the vortex pair always appear irrespective of the rotor-blade orientation, the strength of these vortices does vary with the location of the blade with respect to the grooves. The positive radial velocity induced by these vortices in the space between the grooves entrains high axial momentum fluid from the inner parts of the passage to the tip region, thus reducing the blockage. This particular effect of semicircular casing grooves has not been recognized in our earlier studies based on observations in meridional planes.

While the ensemble-averaged axial velocity for the untreated casing is positive over the entire plane upstream of the rotor, the instantaneous realizations show occurrence of large regions with negative axial velocity. They are almost always surrounded by regions of large (negative as well as positive) circumferential velocity i.e., fast outward flow away from their center (resembling a source), giving an impression of a flow circumventing a region with high adverse pressure gradients. One such event is shown in Fig. 5(a). Figures 5(b) and 5(c) provide statistics on the circumferential and radial locations of such negative axial flow events, for four rotor-blade orientations based on the analysis of 2000 instantaneous realizations. Realizations having at least 2.56 mm² region of negative axial velocity in the flow field are used for analysis. Although this choice is arbitrary, the trends are not qualitatively sensitive to it. The radial distribution (Fig. 5(c)) shows that such events are restricted to the top 30% of the passage with about 50% of them occurring in the tip gap region, hence showing that these events are related to the tip leakage flow. Moreover, although these events occur at all circumferential locations, the distribution is preferentially located in front of the blade leading edge presumably due to the effect of blade-induced pressure gradients. The preferred presence on the PS, irrespective of blade orientation, is consistent with the occurrence of BFVs, which extend diagonally upstream from the suction side mid chord of one blade to the leading edge pressure side of the next blade (see Fig. 6(c)). As the BFVs circumvent the leading edge of the next blade from its pressure side to the suction side and penetrate into the following passage, they induce negative axial velocity in front of it. On the other hand, less frequent events near the suction side could occur due to the velocities induced by the tip leakage vortex. Furthermore, the impact of the IGV wake on the occurrence of these events can be seen in Fig. 5(d). This figure compares the area fraction of these events calculated over 2000 realizations to the total flow field area interrogated. The background is the color contour of axial vorticity from Fig. 4(c). It is evident that when the IGV wake is in the vicinity of the pressure side of the blade, such as at Θ = 0 deg and 4.8 deg, the area fraction is 75% larger in comparison to when the blade is far from the wake.

To further highlight the role of BFVs in inducing intermittent reverse flow upstream of the rotor, Fig. 6 shows color contours of an instantaneous realization of axial velocity (Fig. 6(a)) and circumferential vorticity (Fig. 6(b)) in a meridional plane shortly before the rotor-blade tip leading edge reaches the sample plane, at s/c = −0.11 where s refers to the rotor-blade chordwise coordinate and c is the blade chord. The instantaneous in-plane velocity vectors are also superimposed on the figure, and the black dotted line shows where the blade tip leading edge (z/c_A = 0) is located. The sample area contains a large vortex whose center is marked with a black arrow, extending upstream of the leading edge plane, and inducing negative axial velocity above its center up to z/c_A = −0.13. The location, orientation, and negative velocity induced by this vortex on the pressure side of the blade are consistent with that of a BFV, which is shown in a cavitation flow visualization image in Fig. 6(c). The vorticity direction of the BFV is shown with a yellow arc and it can be inferred that these structures have a signature of positive circumferential and axial vorticity. Flow visualization in previous studies of this machine [11] has shown that multiple such structures can form in the passage with an untreated casing, and they can induce reverse flow at multiple locations, preferentially closer to the pressure side of the rotor blade.

The evolution of the ensemble-averaged flow inside the untreated rotor passage for blade orientation Θ = 0 deg is shown in Fig. 7 using planes A3–A5. Plane A3 cuts the rotor blade at z/c_A = 0.28, A4 cuts it at z/c_A = 0.89, and A5 is located in the rotor–stator gap. At z/c_A = 0.28 (first row), the blockage region with low U_z and high U_θ (peaking as high as 80% of the blade tip speed) covers the upper 30% of the passage. Here, the axial velocity is negative near the casing along the entire circumference, presumably under the influence of the TLV, BFVs, and reverse leakage flow. At z/c_A = 0.89, the blockage region expands to about 40% of the passage, but the negative axial velocity region is restricted mostly to the blade tip and the suction side of the rotor. Downstream of the blade, the low U_z, and elevated U_θ region expand to cover the upper 50% of the span.

The axial vorticity is negative in thin layers along the outer casing and positive over a broad area in the middle of the span. Both layers are predominantly associated with radial gradients in U_θ. To explain how these layers are related to the tip region flow structure, we refer to the results of prior measurements in a meridional plane of the same machine and flow conditions [11]. Figure 8(a) shows the axial vorticity distribution, and Fig. 8(b) illustrates the multi-layer vortical structure around the TLV center, overlaid on the circumferential velocity distribution, in a meridional plane. They show that the tip leakage flow, the TLV, and the flow surrounding it appear like a swirling jet that rotates in the same direction as the blade, but at a lower velocity than the tip speed. As the entrained leakage flow with elevated U_θ swirls around the TLV center, the negative axial vorticity layer (where ∂U_θ/∂r < 0) flowing across the tip gap from the pressure side (layer 1) changes to a radial vorticity layer and then to a broad positive axial vorticity layer (∂U_θ/∂r > 0) below the TLV center (layer 4). The BFVs form along the latter layer at the interface of the high U_θ region and the passage flow underneath it. The signatures of these two layers are evident in the axial planes depicted in Fig. 7. We do not observe layer 2 in the current measurements (Fig. 7(c)), presumably since this layer is too thin at z/c_A = 0.28 for the resolution of the axial planes, whereas the meridional plane measurements have been performed at a higher spatial resolution and smaller sample area. Hence, layers 1 and 3 (Fig. 8(b)) appear as one negative <ω_z> layer in the axial plane data. Deep in the passage (z/c_A = 0.89), the generation of secondary vortical structures persists. They cause TLV breakdown [33], and the resulting blockage expands radially inward to occupy a substantial fraction of the passage span. Consequently, the positive axial vorticity layer becomes broad while the negative layer remains confined to the tip region. This positive layer also persists downstream of the rotor.

As mentioned, the high circumferential velocity observed in the upper part of the span throughout the rotor passage occurs underneath the TLV core, and the secondary flow structures, most prominently BFVs, are formed at the interface of this region and the main passage flow below it. In previous studies of this compressor, data in a series of meridional planes have shown that, at pre-stall, the roll-up of leakage flow around the TLV is characterized by high circumferential velocity. A comparison of the U_θ distribution for z/c_A = 0.28 (A3 plane) and three meridional planes focusing only on the tip region is shown in Fig. 9. For reference, the location of the A3 plane is marked on the meridional plane velocity fields and the location of all the planes relative to the blade is shown in Fig. 9(c). At s/c = 0.22, the high U_θ is caused by the roll-up of the leakage flow associated with the previous blade (not in the field of view). At s/c = 0.55, the leakage flow associated with the blade in the field of view begins rolling up around the TLV center and already has a high U_θ. Finally, at s/c = 0.77, the circumferential velocity magnitude of the leakage flow becomes 0.6–0.7U_T in the upper 15% of the passage.

The information provided in Figs. 5–9 along with prior measurements in meridional planes focusing on the blade tip, and flow visualizations using cavitation [11], demonstrate that the onset of stall is associated with the diagonally upstream propagation of BFVs (Fig. 6(c)). They penetrate into the next passage, either by circumventing the leading edge of the next blade (evident from Figs. 5 and 6) or through the tip gap of the next blade, trigger formation of similar structures there, and so on [11]. As discussed before (Figs. 7 and 8(b) [11]), being associated with the high radial gradients in circumferential velocity under the TLV, these vortices are an inherent part of the tip leakage flow structure. This high U_θ originates from the entrainment of flow from the pressure side of the blade across the tip gap (Fig. 9). As the TLV and this high U_θ region surrounding it migrate away from the blade and approach the next blade, they become the source of the high U_θ flow that is entrained across the tip gap to the next passage. The migration rate of the TLV and surrounding flow toward the next blade increase with decreasing flowrate [39]. Consequently, while BFVs form at all flowrates, the likelihood that they will propagate around the leading edge (or across the tip gap) to the next passage increases with decreasing flowrate.

Next, it would be worthwhile to demonstrate the striking level of flow instability in the rotor passage under pre-stall conditions. Figure 10 shows sample instantaneous velocity and vorticity distributions at z/c_A = 0.28. The region with u_z < 0 extends to the mid-span, the peaks in u_θ exceed the blade tip speed, and the positive axial vorticity with distinct peaks associated with the BFVs and other secondary structures is distributed across the entire width of the passage. The negative vorticity above this layer and near the casing is presumably associated with the TLV. Further downstream (not shown), as the positive vorticity layer expands, some of these vortices intermittently reach the rotor hub. The associated turbulence levels are extremely high and are discussed in a parallel paper [40].

The flow structure downstream of the rotor changes drastically when the axial casing grooves are installed. The velocity and axial vorticity contours are shown in Fig. 11. As is evident, with the grooves installed, the size of the region with low axial momentum and elevated circumferential velocity in the tip region decreases significantly due to significantly less blockage upstream of the rotor (Fig. 4(a) vs. Fig. 4(d)). Consequently, the axial velocity magnitude in the lower half of the passage is smaller in comparison to the untreated casing. Apart from the blade wake, the low momentum region is now mostly centered near the pressure side of the blade. This entire pattern rotates with the rotor and is affiliated with the TLV and BFVs generated by the previous blade (to the left of the shown field of view) that has migrated to the pressure side of the next blade.

While a part of the TLV is entrained into the groove, and generation of BFVs is prevented near the leading edge, both vortices continue to form downstream of the groove, but to a limited extent [29]. Hence, their scales and impact are significantly weaker than those occurring with the untreated casing. Near the casing, there are two separate circumferentially aligned layers of positive vorticity on either side of the wake. On the pressure side, the layer is located under the TLV and corresponds to layer 4 of the untreated casing shown in Fig. 8(b). On the suction side, the layer that appears merged with the blade corresponds to layer 2 in Fig. 8(b). Next, near the hub on the suction side, there is a region of high positive vorticity (marked as A in Fig. 11(c)), with a mild negative region above it (marked as B). While we cannot pinpoint the origin of these regions, it is possible that the positive axial vorticity peak is associated with the hub/horseshoe vortex leg propagating along the suction side of the blade in the field of view, and the broad negative region is the hub vortex leg originated from the pressure side of the next blade. Finally, the two nearly circular blobs of negative vorticity in the middle of the passage are the signatures of the IGV wake (see Fig. 4(f)), implying that the IGV-induced flow non-uniformities persist downstream of the rotor. For the untreated casing at pre-stall conditions, these signatures can be seen at z/c_A = 0.28, but not further downstream, presumably since they are scrambled by the high turbulence. The wake signature of the rotor blade, with grooves installed, is also more distinct. It has a positive axial vorticity signature due to circumferential gradients of radial velocity whose distribution is shown in Fig. 12.

The evolution of flow in the stator passage for the untreated casing is presented in Fig. 13. Note that the color scales here are different from those in the rotor passage and in some cases vary across planes. The broad layer with low axial momentum downstream of the rotor extends into the beginning of the stator (first row, z/c_A,S = 0.1), and U_θ is still elevated in the outer parts (Fig. 13(b)), but not to the same extent as in the rotor passage. Deeper into the stator passage (second row, z/c_A,S = 0.71), the axial velocity distribution becomes more homogenized, the elevated circumferential velocity region shrinks, and the elevated positive vorticity layer fades. Downstream of the stator, U_z (Fig. 13(g)) shows limited signs of the earlier momentum deficit in the outer layer, but there is a broad area with low axial momentum and nearly zero circumferential velocity in the stator blade wake. While the mean boundary layer on the stator blades remains attached in the aft part of the stator passage (U_z remains positive in Fig. 13(d)), instantaneous realizations (not shown) indicate that boundary layer separation occurs intermittently in lower sections of the aft part of the stator suction side, resulting in negative axial flow and, inherently, an increase in the turbulence level. This process is presumably caused by the adverse pressure gradients in the stator, as evidenced by the rapid decrease in U_z from U_z/V_z > 1.5 at z/c_A,S = 0.1 to ∼0.4 at z/c_A,S = 0.71 along the inner part of the suction side boundary layer. Consequently, the stator blade wake is broad and has low axial and circumferential velocity throughout the span. Finally, the stator in this machine has a rotating hub, and the blades are cantilevered with a measured tip gap of about 0.6 mm, i.e., it is much narrower than the rotor tip gap. Hence, we do not resolve the details of the hub leakage flow at the present magnification. However, the flow near the hub, consisting of low axial velocity and elevated circumferential velocity, is consistent with trends observed for the rotor tip leakage.

The velocity distributions downstream of the stator with the axial casing groove installed are shown in Fig. 14. In comparison to the results for the untreated casing, the axial velocity distribution is more homogenous, and the stator wake is thinner. The thinner wake should be an expected outcome of the reduced U_z at the entrance to the stator in the inner part of the passage. Consequently, the flow deceleration, hence the adverse pressure gradients in the inner part of the stator passage are expected to be milder. In the vorticity distributions (Fig. 14(c)), the signatures of stator blade wakes are more distinct along with the hub flow structures. Near the hub, the negative vorticity on the suction side and positive peak on the pressure side could be related to hub vortices, but the flow there is also affected by leakage.

Starting with the flow upstream of the rotor (plane A2), the axial and circumferential velocity distributions for untreated casing near BEP are shown in the first row of Fig. 15. The axial momentum is low near the casing, the hub, in front of the blade leading edge, and at the intersection of the IGV wakes and rotor blade with the hub. However, in comparison to the distributions at pre-stall (Fig. 4(a)), the deficit is considerably milder, especially in the tip region. Consequently, the axial velocity in the mid-passage region is also lower. In contrast, the circumferential velocity near BEP is higher than that at pre-stall over the entire passage although their spatial distributions appear to be similar. Unlike the pre-stall condition, there are no regions/events of negative axial velocity in the instantaneous realizations since a significant fraction of TLV roll-up occurs further downstream in the rotor passage, the leakage flow is weaker, and the BFVs are less frequent and do not extend to the pressure side of the next blade [39]. With the ACGs installed (Figs. 15(c) and 15(d)), the axial and circumferential velocity components in the grooves are negative, as expected, but their magnitudes are lower than those at pre-stall. The flow still jets out of the grooves and generates a pair of axial vortices along the sides of each groove (not shown), but the velocities involved are lower; hence, their influence on the flow in the tip region is reduced.

The velocity and axial vorticity distributions across the rotor passage are presented in Fig. 16. Clearly, the axial velocity deficits are significantly smaller than those at pre-stall and mostly confined to the tip region (Figs. 16(a), 16(d), and 16(g)). The elevated U_θ region is also narrower and does not extend circumferentially to the entire passage (Figs. 16(b), 16(e), and 16(h)). At z/c_A = 0.28, the signature of the leakage flow near the suction side of the blade tip with negative axial and circumferential velocity components is evident. The vorticity contour (Fig. 16(c)) shows multiple negative peaks, the most prominent being the signatures of the TLV of the shown blade near the tip suction side, and that of a previous blade on the left side. Other regions include the pressure side corner of the blade tip, which is associated with the leakage flow entering the tip gap, two mid-passage peaks originating from the IGV wakes, and two structures near the hub. One of them is presumably associated with the rotor-blade hub vortex on the pressure side, and the other is likely to originate from the IGV. Positive axial vorticity peaks are present in two distinct regions. The first extends from the blade tip to the TLV and is presumably associated with layer 2 illustrated in the meridional view (Fig. 8(b)). The second region is located to the right of the TLV of the previous blade and is associated with the radial gradients of U_θ on the outer side of the TLV (upstream of the TLV in the meridional view (Fig. 9(b), s/c = 0.55)).

Deeper in the passage, the TLV, the leakage flow, and the region with elevated U_θ extend away from the suction side of its originating blade toward the pressure side of the next blade [39]. At z/c_A = 0.89 (plane A4), the TLV is located at x/L∼0.7, near the leftmost point of the layer with negative axial vorticity (layer 3). It is bounded by two layers of positive vorticity, a narrow layer close to the tip, and a broader layer, where U_θ is high, deeper in the passage (where BFVs roll-up). These layers are associated with radial gradients in U_θ and are identified as layers 2, 3, and 4 in the meridional plane shown in Fig. 8. The negative vorticity layer extends diagonally to the outer casing, where it persists all the way to the pressure side of the next blade, as seen on the left side for a previous blade. The latter corresponds to layer 1 in Fig. 8(b), which becomes thicker as the endwall boundary layer separates at the point where the axially backward leakage flow meets the passage flow. Note that the meridional plane data show that, owing to its alignment, the circumferential vorticity component in the TLV center is an order of magnitude higher than the axial and radial components [39]. Hence, it is difficult to decipher the TLV in an axial plane. In addition, several negative vorticity peaks appear in the inner part of the passage, including the signature of the IGV wakes in the middle and multiple structures along the hub region. Downstream of the rotor (third row of Fig. 16), the remnants of positive vorticity layer 2 appear merged with the rotor wake, and those of positive layer 4 associated with the previous blade are visible on the pressure side of the wake. A thin region of negative vorticity along the casing appears to be remnants of layer 1. A faint peak in between is presumably the leftover of layer 3 and the TLV. With the ACGs installed (Fig. 17), for the most part, the axial and circumferential velocity distributions downstream of the rotor appear to be very similar to those of the untreated casing (compared to Figs. 16(g) and 16(h)). However, the region with reduced axial velocity and increased circumferential velocity appears to be more circumferentially distributed. The vorticity distributions are quite similar, although the TLV signature at the left end of the negative vorticity layer is less distinct. As discussed in Ref. [30], at high flowrates, the TLV entrains secondary structures from within the grooves into the passage and becomes distributed over a broader area owing to the interaction. Finally, with such similar flow structures at the exit of the rotor, the velocity and vorticity distributions downstream of the stator with casing grooves installed are very similar to those of the untreated casing. The axial velocity distribution for the untreated casing is shown in Fig. 18. Except for deficits near the hub, the narrow stator blade wakes, and very close to the casing, the streamwise momentum distribution is uniform. The circumferential velocity distribution (not shown) is also quite uniform, with magnitudes ranging from 0.15U_T to 0.3 U_T.

In the previous section, the important flow features are identified in the entire compressor. To further comment on the flow, the data have been integrated circumferentially to obtain the passage-averaged velocity profiles for all four rotor-blade orientations. Figure 19 presents the profiles upstream of the rotor, in the rotor–stator gap, and downstream of the stator. Here, the axial velocity profiles are normalized by the passage-averaged axial velocity obtained in the area being averaged, V_zp. It differs from the annual area mean velocity V_z by 2–3%, 5–6%, and 2–3% upstream of the rotor, in the rotor–stator gap, and downstream of the stator, respectively, presumably owing to circumferential variations in fluxes. The left column shows U_z/V_zp, and the right column, U_θ /U_T. For most cases, the blade orientation has very little impact on the passage-averaged velocity profiles. The only exceptions are the flow profiles within the grooves upstream of the rotor at φ = 0.28 (Figs. 19(a) and 19(b)).

Upstream of the rotor (Figs. 19(a) and 19(b)) and at pre-stall, the axial velocity with grooves is higher than that of the untreated casing over the entire span owing to the reverse flow in the grooves (the overall mean is the same). In contrast, differences in the circumferential velocity are confined to the top 20% of the passage. Near BEP, the negative axial velocity in the grooves has a lower magnitude compared to pre-stall; hence, the elevated axial velocity in the passage is also lower and differs from that of the untreated casing only in the outer part of the passage. Conversely, the grooves reduce the circumferential velocity throughout the span, indicating that their impact is not confined to the tip region.

In the rotor–stator gap, Figs. 19(c) and 19(d) show that most of the axial velocity profiles nearly collapse except for the untreated casing at pre-stall. The latter has a nearly 50% velocity deficit near the tip and a substantially higher velocity in the inner part. The corresponding circumferential velocity is high in the tip and hub regions, consistent with the trends discussed earlier. With the grooves and at φ = 0.28, the axial velocity in the upper part of the span is slightly higher than that at φ = 0.37, possibly owing to greater ingestion of the TLV into the grooves. The grooves also significantly reduce the circumferential velocity over the entire outer part of the passage, but increase it near the hub, again affecting the entire span. At φ = 0.37, the circumferential velocity profiles for untreated casing and grooves are similar.

The profiles downstream of the stator (Figs. 19(e) and 19(f)) display more uniformity and considerably lower differences among various cases. The axial velocity profiles nearly collapse, although both high flowrate cases have slightly lower velocity near the tip and higher velocity mid-span. The slightly lower axial velocity at mid-span for the untreated casing at φ = 0.28 corresponds to the widened velocity deficit in the blade wake. The corresponding circumferential velocity is also lower in comparison to other cases, owing to the low circumferential flow within the wake (Fig. 13(h)). In spite of the intermittent separation on the stator blade boundary layer at z/c_A,S = 0.71, the stator is an effective flow homogenizer, even without casing grooves. With the grooves, the axial velocity near the hub upstream of the stator is reduced significantly; hence, the deceleration in the stator passage is milder, the suction side boundary layer stabilizes, and the blade wakes are thinner. Clearly, the grooves located upstream of the rotor modify the flow structure downstream of the stator. Near BEP, the flow downstream is highly homogenized and is very similar with and without the grooves.

SPIV measurements in a one-and-half stage axial compressor characterize the evolution of flow across multiple blade rows while covering entire blade passages and spans. Data are acquired with untreated casing and with semicircular ACGs installed around the rotor leading edge at two operating conditions. For an untreated casing at pre-stall (φ = 0.28), secondary flow structures generated in the rotor passage, most prominently BFVs, extend diagonally upstream beyond the leading edge plane of the rotor. The BFVs, TLV, and leakage flow induce negative axial velocity, i.e., blockage, near the casing that extends to z/c_A = −0.24. As observed by Chen et al. [11], with further reduction in flowrate, the size, frequency, and blockage induced by the BFV and TLV increase rapidly, adversely affecting the machine's performance. Although the reverse axial flow events, upstream of the rotor, are observed throughout the passage, they are more likely to occur near the leading edge of the blade with their distribution being skewed towards the pressure side. In the rotor, the blockage region, characterized by a substantial deficit in axial momentum and high circumferential velocity, expands radially inwards from covering the upper 30% of the span at z/c_A = 0.28 to the upper 50% downstream of the rotor. This non-uniformity disappears rapidly in the stator, where the axial velocity distribution becomes homogenized, except for the blade wakes, and the circumferential velocity decreases to a low value (∼0.15U_T) at the exit from the stator.

Prior measurements in the same machine have shown that the semicircular ACGs ingest a substantial portion of the tip leakage vortex and associated secondary structures, inhibiting the formation and extent of the BFVs. Furthermore, the outflow from the grooves also causes periodic variations in flow angle at the blade leading edge [29]. The present measurements have revealed an additional important mechanism, namely that the outflow jetting out from the upstream end of the grooves rolls up into axially aligned vortices. These vortices act as flow homogenizers by entraining the faster mid-span flow into the tip region. The circumferentially periodic flow phenomena, including the axial vortices, entrainment of mid-span flow to the tip region, and variations in flow angle around the blade leading edge are all mechanisms associated with the discrete ACGs. These effects might contribute to their better stall margin improvements compared to the circumferential grooves [14,41]. The present data also indicate that the effect of ACGs is not limited to the rotor tip region but extends to the entire rotor span, as well as to the flow within and downstream of the stator.

Near the best efficiency point (φ = 0.37) for an untreated casing, the region with low axial and high circumferential velocity associated with the tip leakage flow forms deeper in the passage and remains confined to the vicinity of the casing, even downstream of the rotor. The interactions of grooves with the passage flow are more limited (compared to pre-stall), but still have undesirable effects. Included are roll-up of axial vortices at the exit from the grooves, observed in the present measurements, and the previously observed entrainment of secondary structures generated within the groove into the passage by the TLV [30]. These interactions increase the area occupied by the TLV [30] and are likely contributors to the 2.4% efficiency penalty observed with semicircular ACGs at the BEP flowrate. Indeed, a groove design that minimizes the interaction with the passage at high flowrates avoids efficiency reduction, as demonstrated for the S-shaped ACGs [21,42].

This project along with the facilities and instrumentation involved has been funded in part by the Office of Naval Research under grant N00014-18-1-2430 and in part by NASA (Grant No. NNX17AH42A). The authors would like to express their gratitude to Yury Ronzhes who designed all the mechanical components of the test facility and to previous members of our laboratory that contributed to this project, including Rinaldo Miorini, Huixuan Wu, David Tan, Yuanchao Li, and Huang Chen.

There are no conflicts of interest.

The data sets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.

c =

rotor blade tip chord

L =

nominal distance from hub to casing

U_T =

rotor blade tip speed

V_z =

area-averaged axial velocity

c_A =

rotor blade tip axial chord

c_A,S =

stator blade tip axial chord

r^* =

normalized radial coordinate

r, θ, z =

radial, circumferential, and axial coordinates

u_r, u_θ, u_z =

instantaneous velocities

x, y, z =

cartesian coordinates

U_r, U_θ, U_z =

ensemble-averaged velocities

φ =

flow coefficient

ψ_SS =

static-to-static pressure rise coefficient across entire machine

Θ =

rotor blade orientation

SS =

suction side

PS =

pressure side

LE =

leading edge

TE =

trailing edge

TLV =

tip leakage vortex

BFV =

backflow vortex

 ACG =

axial casing grooves

IGV =

inlet guide vanes

BEP =

best efficiency point