Exp Fluids (2007) 43:371–384 DOI 10.1007/s00348-007-0279-1
R E S E A R C H A RT I C L E
Application of particle image velocimetry to a transonic centrifugal compressor Melanie Voges Æ Manfred Beversdorff Æ Chris Willert Æ Hartmut Krain
Received: 21 September 2006 / Revised: 1 February 2007 / Accepted: 4 February 2007 / Published online: 24 February 2007 Springer-Verlag 2007
Abstract As part of an ongoing research project the performance and internal flow field of a high-pressure ratio centrifugal compressor is being investigated. Based on previous, primarily, point-wise laser-optical measurements the compressor was redesigned and resulted in an improved impeller and diffuser with a single-stage pressure ratio of 6:1 at 50,000 rpm. Current research activities involve the use of particle image velocimetry (PIV) to analyze and further improve the understanding of the complex flow phenomena inside the vaned diffuser given the capability of PIV of capturing spatial structures. The study includes phase-resolved measurements of the flow inside a diffuser vane passage with respect to the impeller blade position. Both, instantaneous and phase-averaged velocity fields are presented. The flow field results obtained by PIV are to be used for future validation of the related CFD calculations, which in turn are expected to lead to further improvements in compressor performance. In addition, the potential of stereo PIV for this type of turbomachinery application could be successfully demonstrated. Keywords PIV Turbomachinery Centrifugal compressor Transonic Vaned diffuser 1 Introduction The demands on modern jet engines concerning performance, reliability, compactness, emissions and
M. Voges (&) M. Beversdorff C. Willert H. Krain German Aerospace Center (DLR), Institute of Propulsion Technology, 51170 Cologne, Germany e-mail:
[email protected]
operational cost continuously increased during the last decades. A prerequisite for a further improvement of today’s highly developed engines is a better understanding of the complex internal flow phenomena of these machines. One way of obtaining reliable information is through flow visualization, provided sufficient optical access is granted to the area of interest inside the machine. A variety of measurement techniques have been applied to investigate the flow phenomena present in advanced turbomachinery components. While single-point laser-optical measurements have been the method of choice to obtain velocity information, the evolving planar techniques such as particle image velocimetry (PIV) have the potential of acquiring similar data more efficiently (i.e., multi-point technique) and may offer additional insights into the flow under investigation by capturing snap-shots of the flow field (i.e., unsteady measurement). In this specific application PIV was chosen to efficiently capture the instantaneous and averaged velocity fields at a high spatial resolution in a relatively short time. PIV is also capable of detecting smallscaled turbulent structures and high velocity gradients. As reported in the past, the application of PIV to centrifugal compressors provides results of good spatial resolution even in transonic flow conditions (Wernet 1999, 2000). Hayami et al. (2004) used PIV to investigate shock waves and rotor–stator interaction in a high-pressure ratio centrifugal compressor. More recently, Ibaraki et al. (2006) performed detailed investigation of the unsteady flow field in a vaned diffuser using phase-averaged PIV measurements. The literature also reports stereoscopic PIV (SPIV) applications to axial compressor flows (Wernet et al. 2005; Liu et al. 2006), but has been focused on
123
372
large-scale test rigs and low-speed operating conditions. Woisetschla¨ger et al. (2003) showed that in combining PIV with other non-invasive techniques (e.g., laser vibrometry) it is possible to investigate pressure fluctuations in turbine wake flow together with the corresponding phase-resolved velocity fields. Latest developments in PIV hardware, such as lasers and cameras, as well as software improvements (e.g., Willert 2004; Scarano et al. 2005; Wernet 2005) support the progress to make this measurement technique available for the increasing demand in high speed applications. In the framework of an ongoing research program at DLR various flow diagnostic and calculation tools have been combined to investigate the performance and the flow field of a high-pressure ratio centrifugal compressor stage. Based on previous results obtained the compressor stage was re-designed resulting in an improved impeller as well as an advanced diffuser geometry. The impeller was designed with splitter blades. In the inlet section, the meridional contour and the blade shape were modified to improve the homogeneity of the impeller exit flow. The diffuser was shaped to match the new impeller exit flow conditions, which implied a conical shape of the diffuser section. In the following investigations on the test rig the improved stage efficiency and characteristic was experimentally verified. The improvement of the flow field inside the impeller, which was predicted by CFD calculations to be more homogeneous, was successfully verified by means of laser-2-focus (L2F) measurements (Fo¨rster et al. 2000; Krain et al. 2001; Krain 2003). PIV was chosen to analyze and further improve the understanding of the complex flow phenomena inside the new vaned diffuser, as this technique is capable of detecting unsteady flow structures and to resolve even high velocity gradients. In addition this measurement technique provides all attributes needed to support and validate the CFD investigations of the compressor flow, which are performed in parallel to the experimental part of the research program. In the presented application measurements were carried out at a rotational speed up to 50,000 rpm.
2 Experimental setup 2.1 Centrifugal compressor stage The centrifugal compressor stage was designed for a pressure ratio of 6:1 at a mass flow rate of 2.6 kg/s and a rotational speed of 50,000 rpm. The corresponding impeller inlet Mach number is M = 1.4, the related
123
Exp Fluids (2007) 43:371–384
impeller exit speed is 586 m/s. This results in transonic flow conditions at the inlet of the diffuser passage; with static temperatures of 83C and a local speed of sound of 378 m/s. Depending on the operating conditions the compressor reaches different temperature levels. In the relevant operating range during PIV investigation the mean temperatures of casing and flow inside the stage varied from 110 to 255C at the diffuser outlet area. The impeller consists of 13 main blades and 13 splitter blades in a backsweep design. Due to the advanced impeller geometry the diffuser section has a conical shape. The impeller exit radius is 112 mm, while the whole stage has a diameter of 600 mm. The diffuser is equipped with 23 profiled guide vanes, which is in contrast to the wedge-vaned diffusers of conventionally designed centrifugal compressors. The leading edge is positioned at a radius ratio of 1.15 related to the impeller exit. The height of the diffuser passage is 8.1 mm and remains constant for the whole passage. 2.2 PIV setup and facility preparation The compressor casing can be equipped with a variety of different diagnostic tools, such as static pressure and temperature probes, small quartz glass windows for point-based laser diagnostics in the impeller region (L2F) and microphone arrays for acoustic investigations (Fig. 1a). To provide sufficient optical access for planar PIV measurements a large quartz window was needed in the diffuser casing. The prepared access port provided a camera observation area of one complete diffuser vane passage, including the impeller exit region (Fig. 1b). A quartz window and a metal window supporting brace were manufactured with considerable effort to precisely match the inner contour of the diffuser casing, thereby minimizing disturbances of the near-wall flow. While the outer surface of the glass was flat, the inner surface was milled and subsequently surface polished by hand. A bulky design of the window was chosen in order to withstand the high temperature and pressure strains during compressor operation and also reduces the likelihood of glass fracture. Compressive stresses caused by mechanical contact of the glass to the diffuser vanes were considered to be the most likely cause for any damage to the glass. To reduce this strain and to provide a reliable seal between the vane passages a silicone sealing was applied to the contact surface of the vanes in the window area. The window itself was set back in the metal support with a clearance of 0.5 mm. The small recess in the casing was considered to have only a minor influence, as the near-wall flow field was known to be subsonic and dominated by wake structures
Exp Fluids (2007) 43:371–384
373
Fig. 1 Compressor facility and PIV set up; a 3D-view of the improved centrifugal compressor stage with casing and instrumentation for flow field and pressure diagnostics; b photograph of optical access and LSP insertion point in the diffuser casing
originating in the impeller exit. In the observed flow area the height of the diffuser passage increased to 8.6 mm over the entire chord length. This increased passage height was limited to the window area covering one diffuser vane passage and did not affect the impeller tip clearance. For flow observation a thermo-electrically cooled, double-shutter CCD camera with a spatial resolution of 1,600 · 1,200 pixels2 at a frame rate of 15 Hz was used. Compared to previously available PIV-cameras, the roughly fivefold increase in frame rate directly matches the laser frequency and significantly reduces overall measurement time, thus reducing operating costs of the test facility. In addition facility seeding is only required for a reduced time period, which reduces window contamination issue (as will be seen later, this was not an issue here). The camera itself was mounted on a Scheimpflug adapter to optimize alignment of the camera optics with the laser light sheet (LS). The focal length of the chosen camera lens is f = 105 mm. A precise calibration target is used to align the LS plane with the PIV camera object plane. The target was made of an aluminum plate of 1.9 mm thickness with a precise 2.0 · 2.0 mm2 dot grid applied on the surface. It could be positioned reproducibly in the diffuser vane passage. Adjustment of the grid in the vane passage was achieved with three setscrews. Due to the conical shape of the diffuser geometry (Fig. 2) as well as high pressure and temperature, it was not possible to use standard, off-the-shelf light sheet probes (LSP) with a common 90 deflection of the light sheet at the exit. Therefore a periscope LSP was specifically designed for this application. The periscope probe shown in Fig. 3 allowed for adjustment of the LS in rotational and axial position and angle relative to the chord of the vane profile. Together with the sturdy probe support in the diffuser casing it was possible to adjust the LS to the vane span
Fig. 2 Schematic of the compressor stage including the PIV setup, indicating the conical shape of the diffuser
Fig. 3 Setup of periscope LSP with beam path schematic
123
374
locations chosen for flow investigation. The probe support close to the diffuser outlet was not perpendicular to the outer machine casing, but inclined to adapt to the conical diffuser area. The outer diameter of the periscope probe was 12 mm; the free beam path inside the probe had a diameter of 6 mm. A pair of cylindrical lenses inside the probe formed the LS with a thickness of 1 mm and a divergence angle of about 6. At the outlet of the periscope probe a mirror deflected the LS with an angle of 97, thus to support the adjustment in the diffuser vane passage. Before entering the LSP the beam diameter of the PIV laser has to be reduced to pass through the periscope probe without streaking the metal surface of the inner tube. Here a pair of spherical lenses was used in a telescopic set up. The complete optical path of the laser beam through the periscope and the related lens set up is illustrated in Fig. 3. The laser beam was delivered via an articulated guide arm to the interface of the LSP. The periscope was also continuously purged with compressed, dry air to prevent deposition of seeding particles as well as to cool the probe with its optical components. In contrast to most commercially available LSP, the delivery end of the probe was not sealed by a glass window. For synchronization of the laser pulses and camera exposure a programmable sequencer unit is used. In combination with a phase shifter the 1/rev-trigger of the compressor can be used to perform phase-resolved PIV measurements with different phase angle relations between impeller and diffuser vanes. Seeding particles were introduced considerably upstream of a contraction leading into the impeller. Here a circumferential traverse supported four seeding probes with different radial positions, which allowed for a nearly uniform seeding distribution across a given sector of the pipe flow. Droplet-based seeding was produced by a battery of three Laskin-nozzle generators filled with paraffin oil with an evaporation temperature range above 200C. It should be noted that the evaporation temperature increases with increased pressures, which in part explains the fact the seeding particles remained visible in the diffuser at temperature above 250C. The use of solid particle seeding as reported by Wernet (2000) was also considered but deemed too risky without additional investigations. The PIV equipment, including laser head, camera and Scheimpflug adapter as well as a light source for camera calibration, was mounted on a massive support decoupled from the test rig. This was necessary to minimize the transmission of machine vibrations to the PIV hardware during operation. The only hardware interface between test rig and PIV set up was the
123
Exp Fluids (2007) 43:371–384
articulated laser guide arm, which was connected to the beam exit of the laser head and to the LSP. 2.3 Parameter studies at low flow conditions Prior to PIV measurements at the relevant operating conditions, parameter studies were carried out for optimization of the PIV set up. The camera was equipped with an objective of 55 mm focal length, allowing for flow observation over the entire chord length of the upper vane (Fig. 4a, top) but at a rather moderate image resolution. The compressor was operated at low speed conditions of 35,000 rpm. At this operating point the seeding probe position was optimized using only one of the four probes. This resulted in a stream tube seeding of the flow. As shown in the middle of Fig. 4a, the seeding was not fully homogeneous in the light sheet section. This was caused by an improper circumferential alignment of the probe to the flow conditions during the feasibility studies. The visible structures quite possibly may be caused by the wake of the impeller blades or mixing phenomena, such as tip clearance flow in the impeller section, where unseeded flow can mix with the seeded streamlines. Despite of the poor seeding quality the trigger chain of laser and camera was verified concerning a correct phase relation between impeller blades and diffuser vanes. Processing of the raw data at the early stages of the feasibility studies showed surprisingly good results. In the bottom part of Fig. 4a, an averaged, phase-resolved velocity field is shown, processed with an interrogation window of 48 · 48 pixels2 at 50% sample overlap. The corresponding grid size was 1.5 · 1.5 mm2 with sample size of 3.0 · 3.0 mm2. Due to the low spatial resolution the expected small-scaled turbulent structures could not be detected, but the mean velocity in the light sheet section matches the predicted values of the CFD calculations. In a next step the camera’s spatial resolution as well as the seeding quality was improved using a f = 105 mm camera lens operating at f/4.0 along with all four seeding probes during parameter studies. The first camera position was chosen to investigate the flow close to the impeller exit; the second camera position followed the flow into the diffuser vane passage. The PIV measurements for both observation areas were performed subsequently by traversing the camera. The resulting combined observation areas of the two camera positions with improved seeding quality are shown in Fig. 4b. With the higher resolution optics the flow could be observed over three quarters of the upper vane’s chord length in two measurement steps allowing
Exp Fluids (2007) 43:371–384
375
Fig. 4 Improvement of seeding quality, camera and LLS position during parameter studies: a first general camera view with low image resolution; b final PIV set up with two camera viewing positions and optimized seeding quality; part b not drawn to scale related to part a
for detailed investigation of small-scaled structures. The use of all four seeding probes resulted in a more homogeneous particle distribution in the illuminated light sheet area. Additionally, the size of the particles could be optimized by switching the impactor of the Laskinnozzle-generators. The impactor is used as a kind of obstacle in the particle flow inside of the generator. Particles of a larger size, that are not able to follow the flow precisely, will be collected by the impactor and removed from the seeding so that only particles of a size 0.3–0.8 lm will leave the particle generator. Taking into account the high temperatures during compressor operation, the smallest particles might evaporate while larger particles survive longer in the flow field, although they have reduced their size when reaching the investigated flow area. In switching off the impactor the particle size distribution in the seeded flow can be increased to 0.8–1.2 lm. This had a significant positive effect on the PIV signal during measurements at design conditions of the compressor stage. Evaluation of one instantaneous PIV image pair applying an interrogation window of 32 · 32 pixels2 (1.0 · 1.0 mm2) with 50% overlap and a corresponding grid size of 0.5 · 0.5 mm2 showed results of very good
quality (Fig. 4b, bottom part). This final PIV setup was chosen for the applied test sequence.
3 PIV measurement sequence 3.1 Compressor operation For flow investigation with PIV the same four operating points as for previous L2F measurements in the impeller region were chosen in order to obtain comparable results with the different measurement techniques (see Fig. 5). The operating conditions, such as corrected rotational speed, corrected mass flow as well as the temperature levels reached in the flow field and at the diffuser casing, are summarized in Table 1. The mass flow is determined with the help of a Venturi nozzle installed in the inlet pipe considering the daily environmental conditions of ambient temperature and barometric pressure. Additionally the tip clearance of the impeller blades was monitored by capacitive sensors. The critical minimum of 0.3 mm was observed during load alternation at a shaft speed of 30,000 rpm, at higher rotational speed the tip clearance increased up to 0.6 mm.
123
376
Exp Fluids (2007) 43:371–384
Fig. 5 Operational characteristics of the improved compressor stage and the operating conditions chosen for laser measurements (marked with red squares)
Table 1 Operating points chosen for PIV investigation of the compressor flow Rotational speed nred (rpm) 35,000
44,000 50,000
50,000
_ (kg/s) Mass flow m Pressure ratio Mean temperature (C) PIV pulse separation (ls)
2.15 4.0:1 175 2.0
2.83 5.3:1 230–235 1.5
1.4 2.5:1 110 2.5
2.6 5.6:1 245–255 1.5
3.2 Data acquisition and post-processing For the PIV data acquisition phase-resolved measurements were carried out using eight equally spaced phase angles per main-splitter passage. As the impeller exit flow was expected not to be symmetric between main-splitter and splitter-main blade passages, the number of phase angles was doubled by treating each half-passage separately. The resulting 16 phase angle relations (0–720) allow for detailed flow investigation related to one complete main-splitter-main passage. Per phase angle a total of 188 PIV image pairs (limited by the 1 GB internal memory of the camera) were recorded at 15 Hz as a continuous PIV sequence. While this number of images is considered to be sufficient for the calculation of phase-average velocities, it certainly is insufficient to reach convergence on the estimation of statistical quantities such as RMS values or Reynolds stresses. Here an estimated 1,000 images per phase angle may have been more adequate. For this application a compromise was made between the detailed investigation of the flow phenomena occurring in the advanced compressor stage on the one hand and the precise analysis of the various parameters characterizing the diffuser flow on the other hand. Thus the
123
limited operation time interval on the compressor rig was used to perform detailed investigation of the diffuser flow field with respect to the various operating conditions. In addition to the measurement procedure covering all phase angles an additional PIV sequence without seeding was recorded to be able to correct for possible laser flare caused by surface reflections using image post processing. As window contamination due to seeding deposits did not occur, it was possible to acquire all phase-angle related measurements for both camera-viewing positions in one session without interruption. Three light sheet planes were selected for each operating point of the centrifugal compressor: one close to the hub at about 19%, one at mid span level (50%) and one close to the casing at about 74% passage height (given values denote the center of the light sheet plane). The upper and lower limitation to the LS position was constrained by the amount of stray light scattered from the hub or the casing surface. The use of the adjustable calibration target, shown in Fig. 6, for the alignment of camera object plane and LS plane assured reproducible measurement conditions for the different operating conditions of the compressor stage. The double-pulse PIV laser was operated with a pulse separation of 1.5–2.5 ls depending on the rotational speed of the impeller (see Table 1) corresponding to a bulk fluid displacement of about 0.3–0.5 mm. Flow analysis in the diffuser vane passage could be performed in an illuminated area of about 60 · 16 mm2 per camera view. Evaluation of the PIV image data was performed after pre-processing with high pass filter, subtraction of background image and masking image areas without velocity information (e.g., diffuser casing or window
Fig. 6 Calibration target with precise 2.0 mm dot grid fixed in the diffuser passage; a adjustment of measurement plane via three micro-screws (only one is visible); b thread for application tool
Exp Fluids (2007) 43:371–384
377
Fig. 7 The effect of image pre-processing from (a) raw data image with scattered light from impeller blades in the lower left corner to (b) pre-processed image with reduced background
light, high pass filter and mask, shown for the first camera position (both images are shown with inverted scale)
support). An example for the image pre-processing is illustrated in Fig. 7. Transformation from the CCD sensors coordinates to physical space was performed using the calibration grid. Fortunately distortion of the particle images (blurring) as well as geometrical distortion (lensing effect) by the curved inner contour of the window was insignificant due to the proximity of the light sheet plane to the window surface. Therefore a dewarping procedure of the images was not necessary. The PIV processing was based on an adaptive, grid refining cross-correlation scheme with continuous image deformation. At the final resolution the algorithm used interrogation window of 32 · 32 pixels at 50% overlap and sub-pixel peak fitter (Whittaker reconstruction). Outlier detection was based on normalized median filtering (Westerweel 2005), followed by linear interpolation of rejected vectors. A correlation plane signal-to-noise ratio of 50 or better could be achieved; the number of spurious vectors was below 3% (mean value upper camera position 2.7%; lower camera position 2.0%). For the calculation of averaged velocity fields all 188 images of a PIV sequence were considered. The re-combination of the obtained velocity fields for both camera views could be easily performed during post-processing of the PIV data with the help of the common calibration grid, as both camera views overlap in one area.
domain). Given a size of 0.5 · 0.5 mm2 for the interrogation area results in structure passing frequencies between 600 kHz and 1.4 MHz in the measured velocity range of 300–700 m/s. Here the size of the particles has an important influence on the obtained velocity data. As the response time of particles about 1 lm in size is on the order of 10 ls (Raffel et al. 1998), the particles behave like a low pass filter with a cut off frequency of 100 kHz applied to the flow. Given a blade passing frequency around 20 kHz suggests that only large-scale structures are faithfully captured, while smaller scales are damped out. Here the use of sub-micron particles could be considered, but this would have the effect of a significant decrease in light scattering efficiency of the particles (Rayleigh scattering regime). In this context it should be noted that PIV is only capable of capturing only a certain portion of the spatial energy spectrum, limited by the wave numbers corresponding to the largest scales (given by the field size) and the smallest scales, respectively (interrogation window size). A detailed analysis on the effect of the velocity spectrum captured with PIV on the measurement accuracy is given in Foucaut et al. (2004), but a corresponding investigation of the measurements presented herein goes beyond the scope of this paper. Another experimental error source for this specific PIV measurement was introduced with the precision of the mechanical set up. A small uncertainty in the adjustment of the two camera positions could not be eliminated. The error made during traversing of camera was 0.5 mm. This has no relevance for each single camera view. For the re-combined velocity maps of both camera views an image shifting tool was used correct for the slight offset. The measurement of the average velocity could be confirmed to be within 1–2% deviation from separate L2F measurements performed at the same operating conditions.
3.3 Error analysis To quantify the order of magnitude of possible errors made during PIV data recording and processing, different error sources should be taken into account (Raffel et al. 1998). Following the assessment suggested by Westerweel (2000) a measurement error of 0.1 pixels can be assumed. With a mean pixel shift varying from 13 pixels at 35,000 rpm up to 20 pixels at 50,000 rpm, this corresponds to a relative measurement error of 0.5–0.8% (2.7–3.5 m/s in the absolute
123
378
Exp Fluids (2007) 43:371–384
4 Results and discussion In the following a selection of the results obtained in the PIV measurement campaign is presented. To give an overview of the different operating conditions, the absolute Mach number distributions at the three measured LS positions are shown in Fig. 8. The Mach number calculation was based on the respective averaged velocity data sets and the mean temperatures given in Table 1. Evaluation of the local static temperatures for calculation of local speed of sound was included in the calculation procedure. It is obvious that the mean passage velocity increases with the impeller rotational speed. An additional velocity variation is visible across the three light sheet positions within the diffuser passage. Here the highest velocity can be found at 50% span level in the vane passage, while the velocity close to the hub at 19% span level is reduced by about 50 m/s, which can be identified as the onset of hub surface influences. A similar reduction of the velocity can be observed in the upper LS plane at 74% passage height. This phenomenon near the casing wall was expected and can be explained with the presence of a wake flow close to the shroud at the impeller exit, which was found by 3D-calculations and additionally during investigation of the impeller exit flow with the L2F method (Krain et al. 2001). Close to the hub the stream traces, included in the contour plots of Fig. 8, nicely follow the deflection imposed by the diffuser vane, with the passage core flow stream lines are passing straight through. In the upper LS plane at 74% passage height the streamlines show evidence of the tip clearance flow as the stream traces turn in the opposite direction compared to the hub flow. The instantaneous velocity fields of the phase-resolved PIV measurements are dominated by small structures and strong gradients, indicating the unsteady and highly turbulent character of the diffuser flow. In contrast to that the flow direction is strongly aligned with the chord line of the diffuser vanes. As shown in Fig. 9, the only variation of flow direction can be observed for the interaction with impeller blade wake flow or with the leading edge of the diffuser vane. Following the development of the transient diffuser passage flow obtained from different phase angles, the flow structures of the passing impeller can be observed. Due to the proximity to the impeller exit the wake appears stronger in the lower left corner of the images than downstream in the diffuser inlet area, where the fast mixing process of blade passage flow and the wake smoothes out the turbulent structures. At a rotational speed of 50,000 rpm the impeller has a blade passing frequency of 21 kHz. Based on the circumferential tip
123
Fig. 8 Mach number distribution calculated from the ensembleaveraged velocity fields at 19, 50 and 74% diffuser passage height for different operating conditions of the compressor: a 35,000 rpm, b 44,000 rpm, c 50,000 rpm (mass flow = 2.6 kg/s at 50,000 rpm)
exit speed of 586 m/s the flow patterns and wake structures followed each other at a time interval of 46 ls, resulting in a spatial distance of 26.7 mm. In the phase angle relations given in Fig. 9 such structures can be identified. Based on former stage designs an asymmetric behavior of the main-splitter and splitter-main passage flow could be expected. The tip clearance flow is represented by low momentum fluid structures in the relative frame of the impeller passage. In the absolute
Exp Fluids (2007) 43:371–384
379
Fig. 9 Instantaneous Mach number distribution obtained at two opponent phase angle relations, characterizing the main passage flow and the splitter passage flow, respectively, at 50% passage height, 50,000 rpm (number of shown vectors reduced for better overview)
frame of reference (diffuser area), the tip clearance flow appears as high momentum fluid patterns (Wernet 2000). Investigating the phase-averaged Mach number distribution, provided as a sequence in Fig. 10, there is no obvious effect of the two different blade types observed downstream of the impeller exit. The typical occurrence of higher momentum core flow in the diffusers absolute frame of reference can be observed, but the difference between main and splitter passage flow is not as pronounced as initially anticipated. The only remarkable aspect is that the high momentum fluid pattern related to the main passage (phase angle 270
in Fig. 10) survives longer than similar patterns originating from the splitter passage (phase angle 540 in Fig. 10). The main passage flow appears as a discrete pattern of higher velocity far downstream in the diffuser throat. This is in contrast to the observations made by Wernet (2000) in a more conventional centrifugal compressor with a 90 exit plane. A closer discussion concerning the impeller main and splitter flow field is given in Krain et al. (2007), including a comparison between L2F measurements and CFD calculations. As a conclusion it can be stated that there still is a minor effect of the impeller tip clearance flow. But due to the advanced geometry of
123
380
Fig. 10 Mach number distribution for a phase-averaged sequence acquired at mid span position and design conditions (50,000 rpm, 2.6 kg/s); phase angles 0–270 showing the higher
123
Exp Fluids (2007) 43:371–384
momentum fluid pattern of the main passage traveling into the diffuser throat; in the corresponding splitter sequence (phase angles 360–630) such structure can not be found
Exp Fluids (2007) 43:371–384
the compressor stage the influence on the diffuser flow character is significantly reduced. In addition two significant effects were identified comparing the instantaneous to the phase-averaged PIV data sets: on the one hand the phase-averaged velocity field appears smooth; the turbulent character, as introduced in the instantaneous flow field, is not evident. The most conspicuous effect is that the expected diffuser passage shock is neither obvious nor located at a defined position; instead the region where the shock position was expected appears to be smeared over a wide area. On the other hand the blade passage as well as wake flow structures are clearly visible in the instantaneous velocity fields. The allocation of the related phase angle can be performed. Even the defined passage shock position can be observed, but the shock position appears to be unstable within one phase relation. Due to this oscillation effect it is not possible to detect a discrete shock position in the phase-averaged velocity field shown in the upper part of Fig. 11. Looking at the detailed instantaneous Mach number distribution shown in Fig. 11a and b, which are chosen from the same PIV sequence at a phase angle relation of 360, it is obvious that the shock position is fluctuating within a characteristic region for the shown phase relation. Additionally, the discrete shape of the shock wave is disturbed by unsteady flow structures generated by the
381
passing impeller blades. Here the diffuser shock wave is affected by the turbulent small-scale flow patterns emerging from the interaction between blade wake flow and the passage shock waves. The vortex streets of the blade wakes periodically trigger the downstream flow, resulting in an oscillating behavior of the shock position. It should be noted that the size of the chosen particles seems to be sufficient to resolve the strong gradient in the shock region. Furthermore, the character of the impeller inlet flow can have a significant influence on the diffuser flow conditions. As reported by Hayami et al. (2004) the impeller inlet flow is dominated by an oscillating behavior of the leading edge shock wave, which can also be a trigger for fluctuating structures downstream in the compressor flow. Here slightly varying blade-toblade tolerance levels are not an issue since phase triggering was performed on a single passage only. Such behavior of the inter-stage shock or similar phenomena were reported in various publications relevant to axial turbomachinery and are well documented in various CFD simulations, e.g., Davis et al. (2002), Miller et al. (2002a, b), Yao and Carson (2006). For centrifugal compressors there is still a need to perform detailed investigations focusing on the interstage flow (as presented by Ziegler et al. 2002a,b), thus to analyze and understand the many interacting flow phenomena.
Fig. 11 Mach number distribution acquired at the phase relation 360 for the phase-averaged flow field (upper part); flow details a and b of two different instantaneous time steps of the same PIV sequence showing the oscillating shock position (Mabs = 1 labeled for better overview)
123
382
5 Outlook on stereo-PIV Subsequent to the PIV test program the setup was extended with an additional camera and Scheimpflug adapter to arrive at a stereoscopic (SPIV) setup in an effort to recover the full three-component (3-C) velocity field. As shown in Fig. 12 the cameras were arranged symmetrically with one camera observing the light sheet in forward scatter while the other operated in back scatter mode. Calibration of the stereoscopic setup was performed on single images of the planar calibration target, as it was not possible to calibrate the field of view within the vane passage using a translated or multilevel target. This camera model based calibration, described in detail in Willert (2006), recovered the camera positions and hence respective viewing angles of ±23 with respect to the light sheet. A disparity correction based on the actual particle image data accounted for a remaining misalignment of 10–20 pixels (0.4–0.8 mm) of the camera views with respect to each other. Following the calibration an
Fig. 12 Setup for the SPIV measurements at the centrifugal compressor; camera 1 is arranged in forward scatter with respect to the light sheet
Fig. 13 Flow details of the SPIV measurement at 50% span level and 50,000 rpm for both camera views, showing vector plots, particle images and the resulting out-of-plane velocity
123
Exp Fluids (2007) 43:371–384
additional test run with SPIV was carried out by acquiring phase resolved measurements at a rotational speed of 35,000 rpm. The convincing results encouraged to further extend the measurement program by running a complete phase-resolved SPIV sequence at 50,000 rpm. A detail from the flow field showing an accelerated blade passage structure at the diffuser inlet is given in Fig. 13. To emphasize the turbulent behavior of the flow a mean value of |U| = 470 m/s has been removed from the images and from the velocity data. For both individual camera-viewing directions the same flow structures can be identified for the in-plane velocity components. In the particle images shown in the left part of Fig. 13 both camera frames are overlaid to visualize that the dominating flow phenomena are already visible and can be recovered in both views. Assuming a displacement measurement error of 0.1 pixels on the PIV data for each camera, the measurement uncertainty of the recovered 3-C velocity data can be estimated to be 2 and 1.5 m/s for the
Exp Fluids (2007) 43:371–384
in-plane velocity components and about 3 m/s for the out-of-plane velocity. The 3-C velocity data presented in Fig. 13 (right part) contains only those velocity vectors for which the residuals in the 3-C reconstruction from the four measured displacements are less than 1 pixel (20 m/s). This shows the necessity of proper reconstruction prior to interpreting any of the acquired results. In general the reconstruction of the recorded SPIV data showed very good results, provided that the seeding quality was homogeneous and background stray light could be reduced to a minimum.
6 Summary In this study the planar PIV technique was successfully implemented to the vaned diffuser passage of a new high-pressure ratio transonic centrifugal compressor. Prior to the experimental part of the research program specific test equipment was designed, such as a periscope LSP and a quartz window for optical access. Measurements were performed at different operating conditions of the centrifugal compressor. For flow investigation in the diffuser vane passage three light sheet positions at different span levels were chosen. To achieve a high spatial resolution the camera was traversed, allowing for a detailed observation of the impeller exit flow and the development of the flow into the diffuser passage. To support and validate the CFD calculations in a parallel part of the research program, phase-resolved measurements were performed up to a rotational speed of 50,000 rpm. The results obtained are of high quality, as evaluation and data analysis have confirmed. An overview on the velocity distribution in the diffuser vane passage was given, as well as a brief discussion of the phase-resolved results. Both averaged and instantaneous velocity fields were presented. The PIV database established in this program can be taken as basis for further detailed evaluations of the flow phenomena occurring in the centrifugal compressor stage. Based on the combined results of PIV measurements and CFD calculations a better understanding of the complex flow characteristics is expected for the future. In extension to the main part of the measurement program a SPIV setup was realized at the test rig with the successful demonstration of this advanced measurement technique in the enclosed environment of the diffuser passage. This additional data set may give an important contribution to the understanding of turbulent flow development in the impeller-diffuser interaction.
383 Acknowledgments The project was sponsored by the German Ministry of Economy via AIF and FVV (BMWi/AIF-no. 13228/ N1, FVV-no. 067980). The authors would like to thank these organizations as well as the industrial partners for the permission to publish the results presented in this paper. We also like to thank the technical personnel of the compressor facility for their support during the test campaign. The suggestions made by the reviewers are gratefully acknowledged.
References Davis RL, Yao J, Clark JP, Stetson G, Alonso JJ, Jameson A, Haldemann CW, Dunn MG (2002) Unsteady interaction between a transonic turbine stage and downstream components. GT2002-30364. In: Proceedings of ASME turbo expo 2002, Amsterdam, June 7–13 Fo¨rster W, Karpinski G, Krain H, Ro¨hle I., Schodl R (2000) 3 Component, Doppler laser-2-focus velocimetry applied to a transonic centrifugal compressor. 10th international symposium on applications of laser techniques to fluid mechanics, Lisbon, July 10–13 Foucault JM, Carlier J, Stanislas M (2004) PIV optimization for the study of turbulent flow using spectral analysis. Meas Sci Technol 15:1046–1058 Hayami H, Hojo M, Hirata N, Aramaki S (2004) Flow with shock waves in a transonic centrifugal compressor with lowsolidity cascade diffuser using PIV. GT2004–53268. In: Proceedings of ASME turbo expo 2004, Vienna, June 14–17 Ibaraki S, Matsuo T, Yokohama T (2006) Investigation of unsteady flow field in a vaned diffuser of a transonic centrifugal compressor. GT2006–90268. In: Proceedings of ASME turbo expo 2006, Barcelona, May 8–11 Krain H (2003) Review of centrifugal compressor’s application and development. GT2003–38971. In: Proceedings of ASME turbo expo 2003, Atlanta, June 16–19 Krain H, Karpinski G, Beversdorff M (2001) Flow analysis in a transonic centrifugal compressor rotor using 3-Component laser velocimetry. 2001-GT-0315. In: Proceedings of ASME turbo expo 2001, New Orleans, June 4–8 Krain H, Hoffmann B, Rohne K-H, Eisenlohr G, Richter F-A (2007) Improved high pressure ratio centrifugal compressor. GT2007–27100 to appear. In: Proceedings of ASME turbo expo 2007, Montreal, May 14–17 Liu B, Yu X, Liu H, Jiang H, Yuan H, Xu Y (2006) Application of SPIV in turbomachinery. Exp Fluids 40:621–642 Miller RJ, Moss RW, Ainsworth RW, Harvey NW (2002a) Wake, shock and potential field interaction in a 1.5 stage turbine: part I: vane-rotor and rotor–vane interaction. 2002GT-30435. In: Proceedings of ASME turbo expo 2002, Amsterdam, June 3–6 Miller RJ, Moss RW, Ainsworth RW, Harvey NW (2002b) Wake, shock and potential field interaction in a 1.5 stage turbine: part II: vane–vane interaction and discussion of results.2002-GT-30436. In: Proceedings of ASME turbo expo 2002, Amsterdam, June 3–6 Raffel M, Willert C, Kompenhans J (1998) Particle image velocimetry, a practical guide. Springer, Berlin Scarano F, David L, Bsibsi M, Calluaud D (2005) S-PIV comparative assessment: image dewarping and misalignment correction and pinhole and geometric back projection. Exp Fluids 39:257–266 Wernet MP (1999) Application of digital particle image velocimetry to turbomachinery. Lecture series on planar optical measurement methods for gas turbine components.
123
384 RTO-EN-6 AC/323(AVT)TP/20, Cranfield, September 16–17, and Cleveland, September 21–22 Wernet MP (2000) Application of DPIV to study both steady state and transient turbomachinery flows. Opt Laser Technol 32:497–525 Wernet MP (2005) Application of planar velocimetry in high speed flows: state-of-the-art and perspectives. PIVnet II international workshop on the application of PIV to compressible flows, Delft, June 6–8 Wernet MP, van Zante D, Strazisar TJ, John WT, Prahst PS (2005) Characterization of tip clearance flow in an axial compressor using 3-D digital PIV. Exp Fluids 39:743–754 Westerweel J (2000) Theoretical analysis of the measurement precision in particle image velocimetry. Exp Fluids 29(Suppl):S3–S12 Westerweel J, Scarano F (2005) Universal outlier detection for PIV data. Exp Fluids 39:1096–1100 Willert C (2004) Application potential of advanced PIV algorithms for industrial applications. PIVnet/ERCOFTAC workshop on particle image Velocimetry, Lisbon, July 9–10
123
Exp Fluids (2007) 43:371–384 Willert C (2006) Assessment of camera models for use in planar velocity calibration. Exp Fluids 41:135–143 Woisetschla¨ger J, Mayrhofer B, Hampel B, Lang H, Sanz W (2003) Laser-optical investigation of turbine wake flow. Exp Fluids 34:371–378 Yao J, Carson S (2006) HPT/LPT interaction and flow management in the inter-turbine space of a modern axial flow turbine. GT2006–90636. In: Proceedings of ASME turbo expo 2006, Barcelona, May 8–11 Ziegler KU, Gallus HE, Niehuis, R (2002a) A study on impeller– diffuser interaction: part I—influence on the performance. GT-2002–30381. In: Proceedings of ASME turbo expo 2002, Amsterdam, June 3–6 Ziegler KU, Gallus HE, Niehuis, R (2002b) A study on impeller– diffuser interaction: part II—detailed flow analysis. GT2002–30382. In: Proceedings of ASME turbo expo 2002, Amsterdam, June 3–6