CEAS Aeronaut J DOI 10.1007/s13272-014-0098-z
ORIGINAL PAPER
Numerical and experimental investigation of an impeller tip clearance variation in an aero-engine centrifugal compressor with close-coupled pipe-diffuser Benjamin Wilkosz • Robert Kunte • Philipp Schwarz • Peter Jeschke • Caitlin Smythe
Received: 23 November 2012 / Revised: 5 January 2014 / Accepted: 21 January 2014 Ó Deutsches Zentrum fu¨r Luft- und Raumfahrt e.V. 2014
Abstract The subject of this paper is the experimental and numerical investigation of the influence of impeller tip clearance on the aerodynamics of a high-pressure transonic centrifugal compressor used in an aero-engine application. The overall change in aerodynamic performance of the stage, the isolated impeller and the isolated diffuser is analyzed. Local flow phenomena, responsible for the change in performance, are examined in closer detail. Experimental data from a state-of-the-art test rig, containing detailed 2D particle-image-velocimetry measurements, are used. 3D Reynolds-averaged Navier–Stokes simulations are conducted with the CFD solver turbo-machinery research aerodynamics computational environment to get a detailed insight into the flow field. This study contributes towards a better understanding of the principal flow phenomena of a centrifugal compressor with a close-coupled pipe-diffuser and the additional losses introduced in the individual compressor components by an increased impeller tip clearance.
This paper is based on a presentation at the German Aerospace Congress, September 10–12, 2012, Berlin, Germany. B. Wilkosz (&) R. Kunte P. Schwarz P. Jeschke Institute of Jet Propulsion and Turbomachinery, RWTH Aachen, Templergraben 55, 52062 Aachen, Germany e-mail:
[email protected] C. Smythe GE Aviation, 1000 Western Ave MD47410, Lynn, MA 01910, USA
Keywords Centrifugal compressor Tip clearance variation Close-coupled pipe-diffuser Aero-engine List of symbols b (mm) Blade height—hub to tip cp Static pressure recovery MFC (kg/s) Corrected mass flow NML Normalized meridional length NOM Nominal tip clearance OP Operation point PIV Particle-image-velocimetry PS Pressure side RANS Reynolds-averaged Navier–Stokes s [J/(kg K)] Entropy SS Suction side t (mm) Tip clearance TC Maximum tip clearance TTR Total temperature ratio TPR Total pressure ratio RPM (1/min) Revolutions per minute v (m/s) Absolute velocity w (m/s) Relative velocity Greek a (°) g k2 n x X
symbols Absolute flow angle Efficiency Relative tip clearance = t/b Diffuser center line coordinate Total pressure loss Impeller rotation
Subscripts is Isentropic m Meridional
123
B. Wilkosz et al.
norm h
Normalized value Circumferential
1 Introduction The centrifugal compressor is often used as the last compressor stage in small jet engines. The centrifugal compressor yields a high stage total pressure ratio (TPR) and therefore enables a compact engine design with a short shaft. This advantage is offset by a lower efficiency, essentially due to the long flow path and the high amount of secondary flow within the centrifugal compressor. The impeller tip clearance of an unshrouded impeller is an important design parameter. The impeller tip clearance has a major impact on the aerodynamic performance due to the extended chord length and small blade height of the centrifugal impeller. Furthermore, the impeller exit flow has a significant influence on the performance of the downstream diffusion system, gaining impact for a closecoupled impeller-diffuser configuration. The clearance between the impeller blade tip and casing is responsible for one of the major loss mechanisms within the impeller: the tip clearance flow. Empirical correlations for the decrease in impeller efficiency can be found widely in the literature [1–4] and vary from dg/dk2 = -0.1 up to -1.0. In an attempt to understand the physics and to predict the impact of the tip clearance flow, many models have been developed over the years. Brasz [1] defines a linear dependence between tip clearance and total pressure loss. The loss model of Brasz considers the losses due to the flow within the tip-gap, as well as the sudden expansion on the pressure side (PS) of the blade. However, not all experimental data show a linear dependency. Senoo [5] also suggested a linear dependence, but extended his model to include a correlation for losses generated by the mixing of the tip clearance flow with the core flow. According to the development of the secondary flow within a centrifugal impeller, as described in detail by Majidi [6], the tip clearance flow mixes with other low-momentum fluid and forms an enclosed region, often referred to as a ‘‘wake’’. Weiss [7] showed that the stream-wise distribution of the tip clearance height has an impact on the formation of the secondary flow and thereby the location of the wake. The ‘‘jet-wake’’ structure which develops affects the inlet flow profile of the downstream diffusion system. In the past, a great deal of research was conducted focusing on the diffusing system. Experimental investigations conducted by Runstadler [8] emphasized the significance of the inlet profile and blockage on the aerodynamic performance of a linear diffuser. Ziegler [9, 10] showed that the impellerdiffuser coupling plays an essential role in the stage performance and the development of the diffuser flow. Detailed experimental and numerical work, conducted by
123
Zachau [11] and Grates [12], underlined the complexity of the flow field at the impeller-diffuser interface for the compressor investigated and its importance on the aerodynamic performance of the diffusing system. Zachau presented experimental data regarding tip clearance effects and concluded that further numerical investigations are necessary to better understand the root cause of the decrease in performance. This paper shows the impact of the impeller tip clearance on the aerodynamic performance of an aero-engine centrifugal compressor with a unique type of diffuser. The centrifugal compressor investigated contains an unshrouded impeller, which is close-coupled with a pipe-diffuser. Understanding the influence of tip clearance on both the aerodynamics of the impeller, as well as the close-coupled diffuser, plays an important role in the design of this type of compressor and data are not widely available in the open literature. The experimental data presented in this study were gathered using a state-of-the-art centrifugal compressor test rig at the Institute of Jet Propulsion and Turbomachinery (RWTH Aachen University). In the past, attempts with both the steady-state [13, 14] and the unsteady [12, 15] CFD approach were made to investigate the compressor investigated. For the investigation presented, the 3D Reynolds-averaged Navier–Stokes (RANS) solver turbo-machinery research aerodynamics computational environment (TRACE) was used. Since the unsteady approach merely results in an off-set in performance [15], a steady-state RANS approach with a conservative mixing plane between the impeller and diffuser was chosen to reduce the computational time. The present investigation yields an improved understanding of the flow phenomena in a centrifugal compressor and provides new insights for the design of future centrifugal compressors.
2 Geometry Figure 1 shows a cross-section of the centrifugal compressor under investigation, which consists of three components: impeller, diffuser and deswirler. The unshrouded impeller contains 23 full and 23 splitter blades with an exit back sweep angle of 24.6°. The diffusion system consists of a diffuser and a downstream deswirler which turns the flow in the axial direction. The passage-type diffuser has features in common with the pipe-diffuser [16]. The 30 passage drillings, manufactured from a solid ring, intersect and form an elliptical leading edge. These ridges function as vortex generators which introduce two counter-rotating vortices close to the PS of the diffuser passage. The impeller and diffuser are closely coupled, with a small
Numerical and experimental investigation of an impeller tip clearance variation
Fig. 1 Schematic blade-to-blade view of the diffuser (left) and meridional view of the centrifugal stage (right)
radial clearance of less than 3.6 %. The deswirler has 90 prismatic vanes. The tip clearance variation investigated is shown to the right in Fig. 1. The impeller tip clearance is varied by shifting the impeller in axial direction relative to the casing and the diffuser. The nominal axial impeller exducer tip clearance is 0.25 mm. This condition will be referred to by the abbreviation nominal tip clearance (NOM). The tip clearance is increased in this study up to ?0.4 mm. The condition with a maximum tip clearance (TC) of 0.65 mm is referred to by the abbreviation TC.
3 Test rig The centrifugal compressor test rig was built at the Institute of Jet Propulsion and Turbomachinery in cooperation with GE Aviation. Zachau [11] describes in detail the design and construction of the test rig, as well as the measurement technique implemented. It is operated in a closed cycle to set the inlet conditions. The inlet total pressure of 108,000 Pa and the inlet total temperature of 288 K are regulated in the upstream settling chamber. A cross-section of the test rig is shown in Fig. 2. The measurement section is installed downstream of a honeycomb and an inlet duct . A spinner ` emulates the flow channel corresponding to the axial compressor in the real engine. The inlet pre-swirl flow angle of 19° is set by an inlet guide vane (IGV ´). The flow through the impeller ˆ, diffuser ˜ and deswirler Þ is collected in the exit plenum þ. The exit pressure is set by a downstream throttle. An active magnetic bearing (AMB ¼) allows the impeller to move in the axial direction during operation. By utilizing the AMB, the impeller
Fig. 2 Centrifugal compressor test rig: cross-section
exit tip clearance is set and varied for the purpose of this investigation. In order to calculate the compressor’s aerodynamic performance, the total pressure (PSI9016 Modul) and temperature (type K thermocouple) are measured by three Pitot rakes, each with three span-wise measurement locations at the outlet plane 7M (Fig. 1). The compressor inlet condition at plane 1M is calculated from the static temperature and static pressure in the upstream settling chamber using an empirical correlation based on detailed measurements. The relative accuracies amount to ±0.1 % for the TPR and ±0.2 % for the total-to-total isentropic adiabatic efficiency gis. The absolute values can be measured to within ±0.3 % for the TPR and ±1 % for the total-to-total isentropic adiabatic efficiency. The mass flow is measured to within an accuracy of ±0.2 % using a measuring orifice. The test facility offers a large number of detailed measurement techniques at a wide range of
123
B. Wilkosz et al.
operating points (OPs). For the presented study, Pitot measurements at the plane 4M (Fig. 1) are used to isolate the impeller and diffuser performance. 2D time-averaged particle-image-velocimetry (PIV) (LaVision system) data are used as a detailed validation of the flow field within the pipe-diffuser.
4 Numerical method The steady-state aerodynamic simulations in this work are conducted with the flow solver TRACE, developed by the German Aerospace Center (DLR) in Cologne. TRACE is a cell-centered, finite-volume, approximate Riemann-solver, based on the steady and unsteady RANS equations. The parallelized TRACE code, based on open source Message Passing Interface, allows the computation of large domains with multi-block meshes. For the spatial discretization, the convective fluxes are computed using an upwind scheme from Roe in combination with a Van Leer’s MUSCL extrapolation. The gradients that are needed to approximate the viscous fluxes on the cell surfaces are computed from a central scheme. The spatial discretization used is of second-order accuracy. For the pseudo time-integration an implicit Predictor–CorrectorScheme is used. In order to couple the rotor and stator for the steady-state simulation, a conservative mixing plane model [17] and non-reflecting boundary conditions are used [18]. For turbulence modeling, a two-equation model is used, based on Wilcox’s k–x-model. The turbulence model implemented in TRACE includes special
Fig. 3 3D view of the computational domain
123
extensions for rotating compressible flows and accounts for streamline curvature effects [19]. A single passage (Fig. 3) of the centrifugal compressor is discretized with a structured multi-block grid. A high resolution mesh with approx. 9 9 106 cells and a nondimensional wall distance y? below 2 is used. At the stage inlet (Fig. 1 1M), a measured span-wise profile for the total pressure, total temperature, and flow angles is applied. Hotwire measurements are used to define the turbulent kinetic energy at the inlet of the stage. Air is defined as an ideal gas. The solid boundaries of the integration domain are modeled as a hydraulically smooth wall. A non-slip and adiabatic boundary condition is applied. In between the impeller and the diffuser, the extraction of the bleed flow is accounted for in the simulation.
5 Results The impact of the relative tip clearance height on the compressor efficiency is discussed in this section. In the ANALYSIS section which follows, the impeller and diffuser are examined separately and a detailed analysis of the changes within the flow field is given. The compressor speed lines (100 % RPM) in Fig. 4 show the normalized TPRnorm and isentropic compressor efficiency gis,norm versus the normalized corrected stage inlet mass flow MFCnorm. Each graph presents both experimental (EXP—dash-dotted line) and numerical data (NUM—solid line). The red speed lines represent the nominal conditions (NOM), whereas the blue speed lines represent the conditions with the largest tip clearance (TC) of 0.65 mm. Between these, two additional speed lines are shown with 0.375 and 0.50 mm tip clearance for the experiment. The numerical results for the NOM and TC condition consist of seven OPs each, from choke to near surge. For the discussion on the detailed flow phenomena, the OPs referred to are marked in Fig. 4. In order to ensure a fair comparison, OPs with an identical impeller inlet mass flow for the NOM and TC simulations are chosen for the detailed numerical analysis. In contrast, the PIV data for both tip clearances are gathered at a fixed throttling position at the stage exit. Only the corresponding parameter TC has been changed. Therefore, the experimental operation point (OP) used for the detailed flow analysis at the TC conditions is slightly shifted to a smaller mass flow. The effect of the increased tip clearance on the experimental result can be summarized in three points: 1. The TPR decreases by -1.2 % (Fig. 4a). 2. The isentropic efficiency decreases by -0.9 %-points (Fig. 4b).
Numerical and experimental investigation of an impeller tip clearance variation
Fig. 4 Compressor performance for different tip clearances at 100 % RPM
3.
The choke mass flow decreases by -1 %. The shape of the speed line remains unchanged. The choke limit is lower due to the decrease in compression and increased aerodynamic blockage at the diffuser inlet. The critical Mach-number within the diffuser throat is therefore reached at a smaller compressor inlet mass flow.
For the numerical simulation the following results can be stated: 1.
In general, CFD is not able to capture the curvature of the speed line due to the almost linear relation between the TPR and the corrected mass flow (MFC). Furthermore, the predicted choke mass flow is overestimated by 2 %. This tendency has been reported previously by other authors in reference to centrifugal compressors [12, 14]. This is caused by the underprediction of the aerodynamic blockage within the diffuser’s throat [15]. 2. CFD predicts a constant decrease of -2 % in TPRnorm. In comparison to the experiment this means an over-estimation of the TPR drop of 0.8 %-points. 3. The total reduction in isentropic efficiency amounts to -1.1 %-points. This reduction meets the experimental delta of -0.9 %-points to within the measurement accuracy of ±0.2 %. 4. The reduction of the choke mass flow is in good agreement with the experimental data. From the flow physics, it can be expected that the averaged TTR should decrease with a larger tip clearance, caused by the lower work at equivalent MFC [5]. The good agreement of the change in stage efficiency, despite the discrepancy in the TPR change, is the result of a TTR decrease of -0.25 % predicted by CFD. However, the
Fig. 5 dgmax/dk2 plot for different speed lines
experimental data does not show this decrease in TTR. It should be noted that the TTR decrease predicted by the CFD simulation is very close to the measurement accuracy and may well be over-predicted. Figure 5 shows the sensitivity of the maximum stage efficiency to changes in the relative tip clearance. The vertical axis shows the normalized maximum isentropic efficiency gis,norm and the horizontal axis shows the relative impeller tip clearance k2. The experimental data points for 90, 95 and 100 % RPM are marked by empty symbols. For each speed line, four points representing the four different tip clearances are shown. The two numerical results for the 100 % RPM speed line at NOM and TC conditions are presented as solid squares. From the experimental data two results can be stated:
123
B. Wilkosz et al.
1 2
The data points for all three speed lines show an almost linear behavior between 0.028 \ k2 \ 0.095. The data points for the three speed lines are congruent. The change in efficiency in relation to the relative tip clearance is therefore independent of the rotational speed and decoupled from the aerodynamic loading within the investigated range.
Using a best fit, an incline of dg/dk2 = -0.22 is found for the experimental 100 % RPM speed line. The incline predicted by CFD is in good agreement with the experimental data. The CFD predicts an incline of dg/dk2 = -0.25 for the 100 % speed line. In the following section of this paper, the impeller and the diffuser—containing the pipe-diffuser and the deswirler—are analyzed separately. The split-up of the additional losses over the impeller and diffuser caused by the larger tip clearance is validated using experimental data. CFD results are used for a detailed analysis of the flow field.
6 Analysis 6.1 Impeller aerodynamics 6.1.1 Impeller performance Figure 6 shows the isentropic impeller efficiency for the experimental and CFD simulation versus the MFC for the 100 % RPM speed line. The experiment shows two OPs, whereas the numerical result contains five OPs. The OPs close to the design point are marked by the larger symbols. In order to compare the experimental results with the numerical results, the Pitot measurement plane (4M Fig. 1) is used to calculate the TPR over the impeller. A detailed validation of the Pitot measurement plane has been given by Kunte et al. [13]. The total pressure is area-averaged. Since the static pressure and temperature are not available in the measurement plane 4M, the absolute flow velocity cannot be determined. Other averaging techniques therefore cannot be used. In order to guarantee a fair comparison, the data from the CFD simulation are interpolated on the discrete measurement positions of the experiment and averaged in the same manner. For the impeller efficiency, the stage’s total temperature ratio (TTR) is used, assuming an adiabatic flow between impeller outlet and stage outlet, and neglecting the change in TTR due to the bleed extraction. The experiments shows a TPRnorm reduction of -1.5 %. CFD predicts an almost constant drop in total pressure of -1.4 %. The decrease in TPRnorm predicted by CFD is in good agreement with the experimental data. The experimental data show an efficiency decrease of -2 %-points at
123
Fig. 6 Normalized TPR and isentropic efficiency of the impeller
the design point. Toward higher mass flow the efficiency delta decreases to -1 %-point. CFD shows a constant reduction in efficiency of -0.7 %-points. Both the experimental and the CFD show the highest efficiency close to the design point. For the analysis of impeller performance, the location of the balance plane has to be considered. It should be kept in mind that only the core of the flow is captured [13] and the Pitot plane is located near the diffusers throat. This location implies that losses generated in the semi-vaneless space are partially included. According to CFD, the losses in this region increase with larger TC (see Fig. 10). This results in an over-estimation of the TPRnorm decrease for the impeller. The maximum efficiency close to the compressor’s design point is in agreement with the impeller design philosophy and supports the impeller performance being
Numerical and experimental investigation of an impeller tip clearance variation
determined using the Pitot plane. The deviation of the experimentally measured change in impeller efficiency can be traced back to differences in the stage’s TTR. Within the experiment a different decrease in TTR for both OPs is calculated. However, this difference in the TTR change is within the measuring accuracy. This shows the difficulty in capturing the impeller efficiency using the afore-mentioned method. 6.1.2 1D impeller outlet flow In order to get a detailed insight into the change in impeller and diffuser aerodynamics, the flow field between these two components is analyzed. Circumferentially flux-averaged 1D-profiles between hub and casing at the impeller exit (Fig. 1 Plane 2) are shown in Fig. 7. Figure 7a–d show the normalized velocity vnorm in the stationary frame, the flow angle a, the TPRnorm, the TTRnorm and the entropy. The span is calculated using the axial hub and shroud position from the TC conditions. The absolute velocity vnorm shows a maximum at 80 % span [Fig. 7a (1)], close to the shroud. This area is dominated by the wake flow. The impeller back sweep results for the wake flow—with a low relative velocity—in a high absolute flow velocity. The reverse applies in the case of
the jet, forced by the meridional curvature towards the hub. When comparing TC to NOM, a decrease of up to -2 % in absolute velocity can be identified in the wake region [Fig. 7a (1)]. For the TC conditions, the flow angle a shows an increase of ?0.5° near the hub and a decrease of -1° near the shroud [Fig. 7b (2)]. This change in the flow angle a profile is caused by an increased blockage in the shroud region. The mass flow density is oppressed by the higher blockage, resulting in an increased meridional velocity near the hub. This causes the flow angle a to increase. The increased distortion of the inlet profile affects the diffuser performance, as shown in the diffuser analysis. As a result of the larger tip clearance, the TTRnorm decreases by -1.5 % in the shroud region [Fig. 7c (3)], reducing the overall work input. The TPRnorm in this region drops by -5 %. The relaxation of the tip clearance flow causes a small decrease in entropy production in the vicinity of the shroud (Span [80 %). Nevertheless, this effect is over-compensated by a significant increase in entropy production in the jet-wake mixing region [Fig. 7d (4)]. 6.1.3 2D impeller outlet flow Figure 8 shows a comparison of the 2D impeller exit flow fields for the NOM (left) and the TC (right)
Fig. 7 Circumferentially fluxaveraged radial profiles at the impeller outlet for the NOM and TC conditions
123
B. Wilkosz et al.
Fig. 8 Normalized relative velocity, entropy and normalized relative velocity of the secondary flow at the impeller exit plane for the NOM (left) and the TC (right) conditions
conditions. At the top (Fig. 8a, b) the normalized relative velocity is shown. In the center (Fig. 8c, d) the entropy is shown. The secondary flow velocity is shown at the bottom of the graph. The latter quantity is calculated as the difference between the main velocity (flux-averaged) and the local velocity vector, projected on the exit plane. The relative velocity (Fig. 8a) shows three characteristic areas of the impeller flow field. Near the casing, a layer of high-velocity fluid can be seen, generated by the passing of the casing in the relative frame of reference. Near the hub two areas with a high relative velocity are observed. These areas represent the jet flow in the impeller. In between a large area of low energy fluid (wake) develops. The split jet [Fig. 8a (1)] is remarkable. Up to 60 % along the meridional length of the impeller, an explicit single-jet singlewake develops within both impeller channels. This classical break-up of the flow field is caused by the radial curvature combined with the higher Coriolis forces within the faster moving jet flow, pushing it towards the hub PS [9]. Due to the contraction of the meridional flow path within the exducer, the jet is divided into two parts by the growing wake area. The entropy distribution (Fig. 8c) provides an insight into the loss generation within the impeller. Near the shroud and within the tip clearance region, the high shear stresses in the flow generate a layer with high entropy [Fig. 8c (2)]. Low-momentum fluid from the tip clearance flow is collected in the wake area. The shear stress between the low-momentum wake and the high-momentum jet flow is responsible for a large part of the entropy production within the channel. The lowest entropy level is found within the jet flow [Fig. 8c (3)]. The low-entropy jet region
123
is separated from the solid wall by the boundary layer with increased entropy. The secondary flow (Fig. 8e) shows four characteristic flow patterns, described in detail by Ziegler [9]. At the shroud, a thick layer of fluid motion towards the PS can be seen (A), caused by the passing shroud. The area dominated by the jet flow (C) shows a fluid motion towards the PS evoked by the high Coriolis force within the jet flow. In between these secondary flows, a fluid motion towards the suction side (SS) can be seen (B). This transport is a direct result of the fluid transport mechanisms (A and C) mentioned previously. It serves to balance the two mass flow transports from the PS to the SS. Within the low-momentum hub boundary layer, a secondary flow is observed (D), induced by the hydrostatic pressure difference between the PS and the SS. The following effects of the TC conditions on the impeller exit flow field can be seen: 1
As has also been described by Weiss [7], the wake region increases and is pushed towards the PS for the larger exducer TC [Fig. 8b (4)]. The core velocity of the jet increases. This increase in the jet velocity underlines the result of the flux-averaged profile at the impeller exit. 2 The entropy production in the tip clearance mixing region decreased due to the lower pressure gradient over the impeller blade [Fig. 8d (5)]. The entropy production in the jet and jet-wake mixing region increases [Fig. 8d (6)] due to the increased flow velocity of the jet and increased difference in flow velocity between the jet and the wake [20].
Numerical and experimental investigation of an impeller tip clearance variation
3
The cross-transport of fluid decreases due to the lower pressure difference between the PS and SS of the impeller [Fig. 8f (7)].
6.2 Diffuser aerodynamics The correct simulation of the diffuser inlet region is essential for predicting the performance of the centrifugal compressor. This region is decisive for the impeller-diffuser coupling and responsible for a large part of the losses being generated. In reality, the flow in this region is highly unsteady due to the jet-wake structure discharged from the impeller [15]. The use of a mixing plane at the impellerdiffuser interface for the steady CFD simulation has an impact on the aerodynamics within this region. Nevertheless, steady CFD is able to predict the shape of the total pressure field at the Pitot plane, in good agreement with experimental data. Furthermore, the values for the total pressure are predicted to within 1.5 % accuracy. A detailed validation of the Pitot measurement plane was shown by Kunte et al. [13]. 6.2.1 Diffuser performance In order to judge the impact of the impeller tip clearance on the diffuser performance, the normalized static pressure recovery coefficient cp,norm and the total pressure loss xnorm for the diffuser at 100 % RPM are shown in Fig. 9. As with the impeller performance analysis (Fig. 6), the Pitot measurement plane 4M (Fig. 1) is used to calculate the total pressure at the diffuser inlet. The static pressure at the impeller trailing edge is used to determine the dynamic head at the diffuser inlet. Figure 9a shows the diffuser’s static pressure recovery cp.norm. It is found that the pressure recovery of the diffuser improves as the compressor is throttled. Simultaneously, the total pressure loss xnorm (Fig. 9b) decreases. This trend in pressure loss correlates with the stage performance behavior shown in Fig. 4. The best diffuser performance is found for the nominal conditions (NOM). At the design point, CFD over-predicts cp,norm by 0.01 for the nominal configuration and underestimates the total pressure losses by -0.005. The characteristics predicted by CFD of both cp,norm and xnorm are in qualitative agreement with the experimental data. For the TC conditions, the experimental data show a decrease in cp,norm and an increase in xnorm. cp,norm is changed by -0.0075, whereas xnorm increases by 0.008. The same tendencies are represented by the CFD simulation. The decrease in cp,norm for TC is predicted as being twice as high (-0.015). For xnorm CFD predicts an
Fig. 9 cp,norm and xnorm for the diffuser
increase of 0.016. The predicted changes are over-estimated, indicating that the CFD method used reacts more sensitive to the changed inlet conditions caused by the larger tip clearance. 6.2.2 Split-up of the additional losses induced by the large tip clearance The split-up of the decrease in efficiency over the impeller and close-coupled diffuser is of special interest in the investigation presented here. When balancing the single impeller domain and the single diffuser domain, CFD shows that the diffuser is responsible for 50 % of the additional losses in the compressor’s isentropic efficiency gis as a result of the large tip clearance. The decrease in diffuser performance is responsible for -0.5 %-points of the total predicted decrease in isentropic efficiency of -
123
B. Wilkosz et al.
1.1 %-points. However, Fig. 9a shows that steady CFD over-predicts the increase in total pressure loss. Nevertheless, the assertion that the decrease in diffuser performance significantly influences the compressor isentropic efficiency remains. When applying the experimental measured change in xnorm of the diffuser, in combination with the efficiency drop predicted by CFD for the impeller, the diffuser is still responsible for 30 % of the additional decrease in isentropic compressor efficiency. This estimation of the influence of the diffuser performance on the deterioration of the compressor’s efficiency is conservative, since the additional losses in the vane less space are not charged upon the diffuser due to the location of the Pitot plane (see Sect. 6.1.1). 6.2.3 1D streamwise diffuser performance In order to identify the location of the additional loss generated within the diffuser, the meridional development of the flow is analyzed. Figure 10 shows the entropy averaged [21] total pressure loss of the diffusing system versus the normalized meridional length (NML). The NML is the meridional length in percentage terms between the diffuser inlet (0 %) and outlet (100 %), as shown in Fig. 1. The left-hand vertical axis shows the normalized total pressure loss xnorm for NOM (green) and TC (red) conditions. The right-hand vertical axis shows the difference in xnorm between NOM and TC (blue). The most important characteristic locations along the flow path are marked within the plot. The bulk of the total pressure loss is generated in the front part of the diffuser. 80 % of the total pressure loss is generated up to the middle of the pipe-diffuser, at 30 %
Fig. 10 Meridional development of the total pressure loss in the diffusing system
123
NML. Furthermore, 50 % of the total pressure loss is generated from the diffuser inlet up until the throat area at 10 % NML of the diffuser. These high losses in the front region of the diffuser are caused by the high Mach-number of the core flow, the flow incidence and high mixing due to the inhomogeneous impeller discharge flow. Within the diffuser, a pronounced behavior can be observed for the TC conditions. The majority of the increased total pressure loss is generated within the first 30 % of the diffusing system. Dxnorm increases rapidly from the diffuser inlet to the diffuser’s leading edge. This is caused by the higher mixing losses and incidence losses due to the more disturbed inlet profile (see Fig. 7b) and increased vortex strength close to the back wall of the diffuser. A more detailed discussion of these phenomena will follow later in the analysis (see Sect. 6.2.5) 6.2.4 Blade-to-blade plane in the diffuser Figure 11 shows the reduced normalized velocity as contours and white streamlines at a plane of constant span (span = 50 %) in the diffuser. On the left-hand side, experimental data from time-averaged PIV measurements are shown, while the corresponding steady-state CFD simulations are shown to the right. In order to ease the comparison, frames of the three measurement windows are shown in the numerical results. The experiment shows a strong separation in the large rear window (right) on the PS of the diffuser. The tendency of the jet following the SS is already visible more upstream in the middle window. The CFD method used captures the diffuser’s flow field, with respect to the magnitude of the velocity as well as the direction of the flow, in good agreement with the experimental data. For the throat and middle window differences of smaller than 5 % in the velocity are found. The rear window, containing the separation and recirculation shows a few locations with a velocity underestimated by 10 %. Nevertheless, these relative differences are considered to be small when taking into account the low velocity magnitude and the complexity of the flow field in the rear window. A shift of the jet towards the diffuser’s SS [Fig. 11d (1)] can be seen for the TC conditions in the middle and rear window. In the middle window near the PS, the experiment shows fluid with lower velocity. In this area, maximum changes of up to -11 % relative to the NOM configuration are detected. Furthermore, the onset of the PS separation moves upstream [Fig. 11d (2)]. The corresponding CFD simulation shows both tendencies. An intensification of the separation, characterized by the faster expansion of the low-momentum region [Fig. 11d (3)] is observed.
Numerical and experimental investigation of an impeller tip clearance variation
Fig. 11 Velocity distribution captured using time-averaged PIV (left) and steady-state CFD (right) within the diffuser Fig. 12 Changes in the pipediffuser’s flow field caused by the larger impeller tip clearance
6.2.5 3D flow in the pipe-diffuser A detailed view of the change in loss mechanisms and flow structures due to the TC conditions within the front part of the pipe-diffuser is shown in Fig. 12. Figure 12 shows two channels of the pipe-diffuser in the downstream direction. The flow field is visualized on planes of a constant n (Fig. 1), aligned orthogonally to the center line of the pipe. Two quantities are shown. The upper figure shows the absolute change in entropy level. The lower figure shows the absolute change in Mach number, giving a detailed insight into the changed deceleration rate within the pipe-
diffuser. Red indicates an increase in the flow quantity, whereas green indicates a decrease. For the change in entropy level, three major effects can be identified. First, starting at the inlet of the diffuser, the core flow shows an increased entropy level, discharged from the impeller core flow [Fig. 12 (1)]. Second, an intensified back wall vortex increases the entropy [Fig. 12 (2)]. Third, an increased entropy level in between the pipes jet flow and recirculation, as a result of the intensified separation, is observed [Fig. 12 (3)]. The latter effect is induced by a change in the two streamwise counter-rotating vortices, generated by the pipe-diffuser inlet ridges. The increased
123
B. Wilkosz et al. Fig. 13 Normalized secondary velocity in the diffuser’s throat
vortex strength for the back wall vortex is a result of the axial movement of the impeller for the larger tip clearance settings, increasing the back-ward facing step (Fig. 1). At the front wall an increased clockwise rotating vortex is generated by the higher distortion of the inlet flow angle. Hence, both the two counter-rotating vortices increase in strength, as illustrated using the normalized secondary flow velocity in the throat in Fig. 13. As shown by the authors in [22], the vortex strength can be linked with the diffusion-rate and magnitude of separation within the pipe. Between the two vortices, the fluid is pulled from the SS towards the PS, destabilizing the PS. At the PS secondary velocity stagnation point, in between the two vortices, the flow separation is initiated. For the TC conditions, the two vortices increase in size and strength, relocating the separation point further upstream [Fig. 12 (4)]. Furthermore, the separation is intensified. This phenomenon was also confirmed by the PIV measurements (Fig. 11). The higher impulse flow from the PS to the SS and the increased blockage, due to the larger separation, push the jet towards the SS [Fig. 12 (5)]. Kunte et al. [13] confirmed this trend using detailed three-hole measurements within the diffuser. As Fig. 10 shows, these phenomena cause an additional loss in the diffuser and underline the impact of the diffuser inlet condition on the compressor performance when using a close-coupled diffuser.
7 Conclusion This paper presents an experimental and numerical analysis of the effect of the relative impeller tip clearance in a highpressure centrifugal compressor with a close-coupled pipediffuser. The study provides a fundamental insight into the split-up of the additional loss induced in the impeller and close-coupled diffuser. Furthermore, the flow phenomena responsible for the deterioration of the compressor’s efficiency are identified.
123
Experimental data show that for the compressor investigated, the efficiency loss dg/dk2 within the range of 90–100 % RMP behaves in a linear way. The inclination measured is dg/dk2 = -0.22. It has been shown that steady CFD predicts this sensitivity with good agreement, overestimating it slightly with dg/dk2 = -0.25. The drop in impeller TPR is constant for all OPs. Within the impeller an increased wake, fed by the tip clearance flow, is observed for the TC conditions. This additional blockage is responsible for the higher distortion of the flow field at the impeller exit. Increased jet-wake mixing losses and higher losses within the jet boundary layer are identified as causes for the lower impeller efficiency. In the diffusion system, an increase in total pressure loss was found experimentally and numerically for the TC conditions. Within the diffuser approximately 80 % of the additional losses generated by the larger tip clearance develop in the front end of the passage-diffuser. These additional losses are traced back to additional mixing losses due to the higher inlet distortion, intensified inlet vortexes, and additional incidence losses. The intensification of the counter-rotating vortices, typical for the diffuser investigated, is identified as the reason for the earlier and more intensified separation within the pipediffuser. In the case of the close-coupled impeller-diffuser configuration investigated, this study shows that the change in both the impeller and the diffuser aerodynamics is responsible in equal measure for the decrease in isentropic efficiency caused by an increased tip clearance. These results emphasize the impact of impeller-diffuser matching for close-coupled components, as they are often found in centrifugal compressors used in aero-engines. Acknowledgments General Electric Aviation (GEA) funds the Centrifugal Compressor Technology Project at the Institute of Jet Propulsion and Turbomachinery. This is gratefully acknowledged. Further the GEA Compressor and Fan Aero group (Lynn) extensively supports this research. Special thanks go to the DLR Cologne for the close cooperation and support regarding TRACE.
Numerical and experimental investigation of an impeller tip clearance variation
References 1. Brasz, J.: Investigation into the effect of tip clearance on centrifugal compressor performance. ASME 88-GT-190 (1988) 2. Eckert, B., Schnell, E.:Axial- und Radialkompressoren. Anwendung, Theorie, Berechnung, Springer (1961) 3. Turunen-Saaresti, T., Jaatinen, A.: Influence of the different design parameters to the centrifugal compressor tip clearance loss. In: Proceedings of ASME Turbo Expo 2011, GT2011-46627 (2011) 4. Bansod, P., Rhie, C.: Computation of flow through a centrifugal impeller with tip leakage. AIAA 90-2021 (1990) 5. Senoo, Y., Ishida, M.: Deterioration of compressor performance due to tip clearance of centrifugal impellers. ASME J. Turbomach. 109, 55–61 (1987) 6. Majidi, K.: Numerische Berechnung der Sekunda¨rstro¨mung in radialen Kreiselpumpen zur Feststofffo¨rderung. PhD thesis, Technische Universita¨t Berlin (1997) 7. Weiss, C., Grates, D.R., Thermann, H., Niehuis, R.: Numerical investigation of the influence of the tip clearance on wake formation inside a radial impeller. GT2003-38279 (2003) 8. Runstadler, P.W., Dolan, F.X.: Diffuser Data Book. Creare Incorporated (1975) 9. Ziegler, U., et al.: A Study on impeller-diffuser interaction-part I: influence on the performance. GT-2002-30381 (2002) 10. Ziegler, U., et al.: A study on impeller-diffuser interaction-part II: detailed flow analysis: influence on the performance. GT-200230382 (2002) 11. Zachau, U., Niehuis, R., Hoenen, H., Wisler, D.C.: Experimental investigation of the flow in the pipe diffuser of a centrifugal
12.
13.
14. 15.
16. 17.
18.
19. 20. 21. 22.
compressor stage under selected parameter variations. GT200959320 (2009) Grates, D.R., et al.: Numercial investigation of the unsteady flow inside a centrifugal compressor stage with pipe diffuser. J. Turbomach. 136, 031012-1-14 (2014) Kunte, R., et al.: Experimental and numerical investigation of tip clearance and bleed effects in a centrifugal compressor with pipe diffuser. J. Turbomach. 135(1), 011005 (2012) Wallis, C., Moussa, Z., Srivastava, B.: A stage calculation in a centrifugal compressor. ICAS2002 CONGRESS (2002) Wilkosz, B., et al.: Numerical investigation of the unsteady interaction within a close-coupled centrifugal compressor used in an aero engine. J. Turbomach. 136(4), 041006 (2013) Bryans, A.: Diffuser for a centrifugal compressor. US Pate Number 4,576,550 (1986) Zachcial, A.: Mischungsebenenmodellierung zur Analyse der raeumlichen Stroemung in mehrstufigen Turbomaschinenkomponenten. ISBN-10: 3-8322-5115-4 (2006) Giles, M.: Non-reflecting boundary conditions for the Euler equations. In: Massachusetts Institute of Technology, Department of Aeronautics & Astronautics—internal report (1988) Kozulovic, D., et al.: Modelling the streamline curvature effects in turbomachinery flows. GT2006-90265 (2006) Denton, J.D.: Loss mechanisms in turbomachines. ASME 93-GT435 (1993) Schimming, P., Starken, H.: Data reduction of two dimensional cascade measurements. AGARD 328 (1975) Wilkosz, B.: Numerical investigation of the steady separation inducing mechanisms in a passage diffuser with application of twoequation turbulence models. In: proceedings CMFF 2012-254 (2012)
123