International Journal of Heat and Fluid Flow 52 (2015) 129–139

Contents lists available at ScienceDirect

International Journal of Heat and Fluid Flow

journal homepage: www.elsevier .com/ locate/ i jhf f

Influence of tip clearance on flow behavior and noise generationof centrifugal compressors in near-surge conditions

http://dx.doi.org/10.1016/j.ijheatfluidflow.2014.12.0040142-727X/� 2014 Elsevier Inc. All rights reserved.

⇑ Corresponding author. Tel.: +34 963 877 650.E-mail addresses: galindo@mot.upv.es (J. Galindo), anti1@mot.upv.es (A. Tiseira),

ronagar1@mot.upv.es (R. Navarro), milpehi@mot.upv.es (M.A. López).

J. Galindo, A. Tiseira, R. Navarro ⇑, M.A. LópezCMT – Motores Térmicos, Universitat Politècnica de València, Camino de Vera, 46022 Valencia, Spain

a r t i c l e i n f o

Article history:Received 3 May 2014Received in revised form 14 November 2014Accepted 17 December 2014

Keywords:CFD simulationTurbochargerShaft motionAeroacousticsDESURANS

a b s t r a c t

CFD has become an essential tool for researchers to analyze centrifugal compressors. Tip leakage flow isusually considered one of the main mechanisms that dictate compressor flow field and stability. How-ever, it is a common practice to rely on CAD tip clearance, even though the gap between blades andshroud changes when compressor is running. In this paper, sensitivity of centrifugal compressor flowfield and noise prediction to tip clearance ratio is investigated. 3D CFD simulations are performed withthree different tip clearance ratios in accordance to expected operating values, extracted from shaftmotion measurements and FEM predictions of temperature and rotational deformation. Near-surge oper-ating conditions are simulated with URANS and DES. DES shows superior performance for acoustic pre-dictions. Cases with reduced tip clearance present higher pressure ratio and isentropic efficiency, but nosignificant changes in compressor acoustic signature are found when varying clearance. In this workingpoint, tip clearance is immersed in a region of strongly swirling backflow. Therefore, tip leakage cannotestablish any coherent noise source mechanism.

� 2014 Elsevier Inc. All rights reserved.

1. Introduction

Turbochargers are present in almost all diesel engines and itsuse in gasoline engines has been steadily increasing in the lastyears. Turbocharging allows reductions of engine displacementand weight without reducing the power and torque produced, ina trend known as downsizing. This approach reduces fuel con-sumption and emissions, which is a must for successfully fulfillingthe current regulations.

Yet strongly downsizing of engines and increase of low speedtorque raises an issue with turbocharger airborne noise (Evansand Ward, 2005), since compressor operating points are shiftedtowards surge region (Teng and Homco, 2009).

Leakage flow from blade PS to SS across tip clearance is oftenconsidered as a key point in the stability and noise production ofcentrifugal compressors. Raitor and Neise (2008) conducted anexperimental study to analyze the sound generation mechanismsof centrifugal compressors. Tip clearance noise (TCN) predomi-nates over BPF tone for subsonic flow conditions at low compressorspeeds. TCN is a narrow-band noise observed at frequencies abouthalf the blade passing frequency (BPF), which increase with speed.

Raitor and Neise considered that secondary flow through blade tipclearance is the source of TCN.

In this frame, CFD is a tool used by researchers to investigatecentrifugal compressor flow field and particularly tip leakage flow.

Mendonça et al. (2012) analyzed flow-induced aeroacoustics ofan automotive radial compressor using DES. A narrow band noiseat a frequency about 70% of rotational speed was detected. Rotat-ing stall was found to be the source of the narrow band noise.Tip leakage allows the stalled passages to recover by pushing thelow momentum region to the rotation-trailing passage, accordingto Mendonça et al.

Tomita et al. (2013) studied two compressors with similar mapexcept in surge vicinity. Unsteady pressure measurements per-formed just upstream the impeller leading edge at low mass flowrate revealed differences between compressors: the one with thenarrowest operating range showed large pressure fluctuations atsubsynchronous speeds moving circumferentially at about50–80% of compressor speed, whereas compressor with wide oper-ating range did not exhibit this large amplitude fluctuations. 3DCFD numerical simulations were also performed in order to ana-lyze fluid phenomena at this low mass flow rate conditions.Compressor with less surge margin presented a tornado-typevortex between blades, blocking the incoming flow. This vortex isconvected downstream and a new one is created in the adjacentblade, in a rotating stall pattern. Conversely, spiral-type tip leakage

Nomenclature

List of symbolsa speed of sound (m s�1)cp specific heat capacity at constant pressure (J kg�1 -

K�1)_m mass flow rate (kg s�1)

N compressor rotational speed (rpm)p pressure (Pa)t time (s)T temperature (K)u axial velocity (m s�1)_W compressor absorbed power (kg m2 s�3)

Wu compressor specific work (m2 s�2)g efficiency (%)� relative difference (%)c ratio of specific heats (–)/ generic variablePt;t total-to-total pressure ratio (–)s compressor torque (kg m2 s�2)

Sub- and superscripts� corrected variable0 stagnation variable1/3, 2/3, 3/3 related to one third, two thirds or three thirds of ori-

ginal clearance, respectivelyback backward traveling waveCFD related to simulation

exp related to experimental measurementin inlet ductforw forward traveling waveout outlet ducts isentropicref reference value

List of abbreviationsBPF blade passing frequencyCAD computer-aided designCFD computational fluid dynamicsDDES delayed DESDES Detached Eddy SimulationFEM finite element methodIDDES improved DDESLES Large Eddy SimulationLIC Line Integral ConvolutionMoC Method of CharacteristicsPS (blade) pressure sidePSD power spectral densityRANS Reynolds-averaged Navier–StokesTCN tip clearance noiseTCR tip clearance ratioSS (blade) suction sideURANS Unsteady RANSWMLES wall-modeled LES

130 J. Galindo et al. / International Journal of Heat and Fluid Flow 52 (2015) 129–139

vortex breakdown occurred at all blades for compressor withhigher surge margin. Some rotating instability was observed in thiscompressor, but not as severe as the rotating stall experienced as inthe other compressor. Tomita et al. (2013) concluded that tip leak-age vortex breakdown could stabilize the flow structure and thusincrease surge margin.

Increased computational resources have made possible toinclude tip clearance in most of the current CFD simulations of cen-trifugal compressor (Mendonça et al., 2012; Tomita et al., 2013).However, compressor geometry is often obtained from reverseengineering or manufacturer’s CAD, thus modeling a cold tip clear-ance that is not representative of actual working conditions. Defor-mations due to temperature, rotation or pressure are usuallyneglected, and shaft motion is not considered. Including theseeffects will led to coupled fluid–structure simulations, which arevery expensive. Nevertheless, the effect of these factors should bestudied to assess the validity of simulations using CAD clearance.

Many researchers have performed studies of tip clearance sen-sitivity on global variables and local flow features.

Danish et al. (2006) conducted a numerical study in which over-all performance and flow field were investigated for nine tip clear-ance levels from no gap to 16% tip clearance ratio (TCR). Numericalsimulations were performed with frozen rotor technique. Two dif-ferent clearance increase approaches were used: reduction of bladeheight while keeping same housing dimensions and increase ofshroud diameter with same blade height. For operating conditionsclose to best efficiency point, maximum increase of TCR bydecrease of blade height caused a loss of 15% in pressure ratioand 10% in efficiency. If shroud size is increased instead, pressureratio is reduced by 13% and efficiency by 8.5%.

Jung et al. (2012) performed steady simulations using the sameimpeller with six different tip clearance profiles. Reductions in tipclearance caused an increase in pressure ratio and efficiency. Par-ticularly, a decrease in tip clearance at the trailing edge improvedthe compressor performance more than implementing the same

reduction in leading edge tip clearance. A reduced tip leakage flowwas found to significantly improve the diffusion process down-stream the impeller.

Measurements performed by Wang et al. (2011) indicated thatvariations of tip clearance caused small influence on stall incep-tion. However, the reasons could not be investigated because theirmixing-plane, steady simulations failed to converge at near surgeconditions.

However, in these sensitivity studies, particular TCRs are notusually justified by expected values of compressor actual workingconditions and effect on noise generation is not assessed.

According to the literature review, tip leakage flow may play arole in noise generation and stability of centrifugal compressors.CFD can provide a deeper insight into the issue, but clearance sen-sitivity to noise generation must be first assessed to decide if clear-ance variations from baseline CAD conditions can be safelyneglected. In such a sensitivity study, TCRs should be set in accor-dance to expected values while compressor is operating.

In this paper, the effect of tip clearance size on fluid flow andnoise spectra of a 49 mm exducer diameter turbocharger compres-sor is studied. Section 2 is devoted to the estimation of tip clear-ance reduction during actual compressor operation. In Section 3,the numerical set-up is described and the approach to performthe clearance sensitivity analysis is explained. Tip clearance sizeimpact on global variables and noise generation is evaluated inSection 4 by means of PSD comparison. Flow field is investigatedin Section 5 in order to explain the spectra obtained by differentTCRs. Finally, Section 6 includes the main findings of the paper.

2. Estimation of actual tip clearance

In Section 1, some effects usually not considered when simulat-ing turbocharger compressors were introduced. For the sake ofaccuracy, a two-way coupled fluid–structure simulation should be

J. Galindo et al. / International Journal of Heat and Fluid Flow 52 (2015) 129–139 131

performed to take into account potential aeroelastic effects, alongwith impeller and housing deformations caused by temperature,rotation and pressure differences. However, this approach wouldlead to unaffordable computational times.

Wang et al. (2011) used a two-step calculation procedure totake into account impeller deformation when simulating centrif-ugal compressor flow, avoiding a two-way coupled fluid–structure simulation. First, the deformation of the compressorblades were simulated with finite element method (FEM). Then,the flow field was recalculated with the deformed impellerestimated with FEM. In this way, tip clearance changed from auniform clearance ratio of 7.3% into a non-uniform profile thatpresents greater reduction with increasing meridional coordinate(TCR is 50% larger at the leading edge than at the trailing edge).Moreover, incident and exit blade metal angles were alsodeformed. Particularly, tip outlet angle at trailing edge increased3.5�. Leakage flowrate is thus affected by impeller deformation,over all at trailing edge.

In this paper, sensitivity of TCR to centrifugal compressor aero-acoustics will be first investigated. Aeroelasticity is neglectedbecause centrifugal compressor blades are not as long and slenderas in their axial counterparts, which do suffer from phenomenasuch as flutter. Deformation due to thermal and pressure loadsand centrifugal forces can be taken into account with a one-wayfluid-mechanical simulation. If noise happens to be highly depen-dent on TCR, the flow field should be recalculated with the actualtip clearance geometry, following the approach of Wang et al.(2011).

Besides, dynamic mesh approach should be employed to con-sider the complex shaft motion pattern described (see for examplethe explanation of Nelson (2007)). This would imply a greatincrease in computational effort compared to traditional slidingmesh technique, in which rotor region is meshed only once andit is rotated across sliding interfaces, in accordance with compres-sor speed. The possible effect of shaft precession on noise genera-tion is thus neglected in this paper. Instead, eccentricitymeasurements are performed in Section 2.2 to assess maximumclearance reduction due to shaft motion, so as to be consideredin the subsequent sensitivity analysis.

2.1. Thermal and rotational deformation

To estimate decrease of tip clearance due to impeller deforma-tion, mean wheel temperature (120 �C) predicted by CFD and rota-tional speed (160 krpm) were used as an input to a structuralsimulation with FEM, which computed the nodal displacement.The impeller is meshed with more than 12,000 quad/tri elements.The equations are then solved using the preconditioned conjugategradient solved implemented in ANSYS Structural.

Fig. 1. Impeller deformation due to rotational and thermal loads.

Pressure deformation was not considered because its contribu-tion to impeller deformation is marginal (Wang et al., 2011). Selec-tion of mean value instead of temperature distribution is justifiedby high thermal diffusivity of aluminum, which is the material ofmost compressor wheels (Sotome and Sakoda, 2007). The simula-tion was done with characteristics of the Aluminum Alloy(density = 2770 kg/m3, Young’s Modulus = 71 GPa).

Fig. 1 shows the total deformation of the compressor wheel.Maximum deformation of the inducer blades is 11% of CAD tipclearance.

2.2. Eccentricity due to shaft motion

In order to estimate the tip clearance variation due to shaftmotion, the technique developed by Pastor et al. (2012) is applied.It consists in an image recording methodology with a processingalgorithm to estimate the shaft motion. Pastor et al. (2012) showedthat this technique requires a less intrusive apparatus than infra-red sensor measurements, also providing direct observation ofshaft motion phenomenon. Galindo et al. (2013) used this method-ology to improve the knowledge of the behavior of the turbocharg-ers in critical conditions of lubrication, because, as López (2014)indicated, image recording is better suited than standard inductivesensors for shaft motion measurements in these lubrication condi-tions. For more details, please refer to the work of López (2014).

Fig. 2 depicts the layout of the image recording apparatus. Thecamera is placed in front of the turbocharger on the compressorside, positioning it as coaxial as possible to the turbocharger shaftand focused to the shaft tip. A 90� elbow is attached at the inlet ofthe compressor housing with a glass window to allow the properdisplay of the shaft through the camera. Images are taken with aPCO Pixelfly 12-bit CCD Camera with spatial resolution of1280 � 1024 pixels. The distance between the compressor andthe camera is approximately 300 mm. The resolution obtainedwith this layout is approximately 5 lm per pixel. This resolutionallows the observation of shaft movements.

The processing algorithm consists on differentiating specificzones of the image in order to obtain their coordinates. Two screwsare placed at the impeller’s eye plane (see Fig. 3). They are used asreference points to calculate the relative position of the shaft,avoiding errors due to structural vibrations. Fig. 3 shows a photo-graph of the shaft motion. The impeller’s eye can be seen alongwith the reference points (left and right screws). The processingalgorithm identifies the shaft tip and its center (in blue in Fig. 3),whose coordinates are calculated using the reference points forthe subsequent determination of the shaft motion.

The maximum eccentricity of the turbocharger is obtained bymanually achieving the physical limit of shaft motion. Fig. 4 showsthe maximum eccentricity (ellipse with solid black line) and theshaft motion at 160 krpm (blue dot markers). The shaft motiondescribes a zone with a radius between 100 lm and 150 lmregarding the center of the impeller’s plane eye. This motion repre-sents a 22–34% of reduction of the original clearance. Therefore,the maximum reduction of the original tip clearance is between33% and 45%.

3. Numerical model

3.1. Domain and numerical set-up

The impeller, vaneless diffuser and volute of the compressorwere digitalized. A structured-light 3D scanner provided a pointcloud of the bodies, which were processed by reverse engineeringsoftware to reconstruct smooth surfaces. In order to avoid influ-ence of manufacturing variability, only one main blade and one

Fig. 2. Camera layout for shaft motion observation.

Fig. 3. Photograph of shaft motion, showing the estimated position of the shaftcenter.

Fig. 4. Shaft motion at 160 krpm.

Fig. 5. CFD domain.

132 J. Galindo et al. / International Journal of Heat and Fluid Flow 52 (2015) 129–139

splitter blade were digitalized, creating the rest by revolution. Theimpeller and the diffuser are thus 360� modeled. Backplate cavityalong with (cold) tip clearance are also provided by the digitaliza-tion process. In addition, inlet and outlet cross sections wereextruded 5 diameters long to obtain the CFD domain shown inFig. 5. The full domain (inlet pipe, impeller, hub cavity, vaneless

diffuser, discharge volute and outlet duct) is meshed with 9.5 mil-lion polyhedral cells. Navarro (2014) performed a mesh indepen-dence analysis, showing that a mesh twice as fine does notprovide significant differences in terms of global variables andnoise predictions.

Rigid body motion approach was employed for the computa-tions, in which heat transfer with the surroundings was neglected.Smooth walls were considered. Fixed outlet pressure, constantcompressor speed of about 160 krpm and mass flow rate corre-sponding to 1.4 times surge mass flow at this compressor speedwere set to represent a near-stall operating condition. Second-order accurate transient solver was used with a time-step size sothat the impeller mesh turns 1� per time step, since Navarro(2014) proved that this time-step is sufficient to accurately resolvepressure spectra within human hearing range.

Turbulence modeling plays an important role in flow field andnoise prediction of a centrifugal compressor, particularly at near-surge conditions. In this paper, Direct Numerical Simulations(DNS) are rejected due to their infeasibility in terms of computa-tional cost. In order to predict compressor aeroacoustics, threeapproaches to model turbulence could be thus considered:

Fig. 6. Rotor region mesh: cross-section (left) and impeller (right), including aclose-up of tip clearance for the cases studied and isomeridional surfaces.

J. Galindo et al. / International Journal of Heat and Fluid Flow 52 (2015) 129–139 133

Unsteady Reynolds Averaged Navier Stokes Simulations (URANS),Detached Eddy Simulation (DES) or Large Eddy Simulation(LES).

LES are often used in computational aeroacoustics (CAA)(Wagner et al., 2007), due to their potential to capture broadbandnoise turbulent sources. Dufour et al. (2009) reviewed some worksusing large eddy simulations for predicting turbomachinery flow.By looking at their survey, it can be concluded that LES is muchmore extended in axial turbomachinery than in their radial coun-terparts. To the authors’ knowledge, LES in turbocharger compres-sors have only been covered by Karim et al. (2013) and the set ofworks performed by researchers from KTH and University of Cin-cinnati. Hellström et al. (2010, 2012), Jyothishkumar et al. (2013)and Semlitsch et al. (2014) reported good agreement between theirlarge eddy simulations and experimental measurements, showingthe potential of LES to predict centrifugal compressor flow behav-ior, particularly at near-surge conditions. However, LES is not con-sidered in this paper due to the computational cost that wouldimply using a LES-suited grid.

In this work, URANS simulations are performed with k�x SSTmodel (Menter, 1994) closure model. This model is very suited forcompressors, for its accuracy in predicting flow with adverse pres-sure gradients. Menter et al. (2004) applied the SST model for dif-ferent turbomachinery cases, for which a good agreement againstexperimental measurements was found. The model proved to begood at capturing usual turbomachinery flow features such as flowseparation, swirl or strong flow mixing. Robinson et al. (2012) alsoused the aforementioned turbulence model to analyze the influ-ence of the spacing between impeller and vaned diffuser in a cen-trifugal compressor. Smirnov et al. (2007) applied the k�x SSTmodel for the calculation of flow in a one-stage centrifugal com-pressor with a vaned diffuser. Good agreement of the computedvelocity field with the measurements data at the impeller exit isobtained by Smirnov et al. Downstream of the diffuser vane, pre-diction quality depends on the operating point. A comparisonbetween turbulence models was done at one operating point. TheSST model provided better agreement with experimental data thank� � and k�x models.

Finally, DES model is an hybrid RANS-LES approach in which theboundary layer is modeled using URANS and LES is employed forthe free stream detached flow. With such a blending between tur-bulence approaches, DES is intended to save computational costwhen compared with LES but retaining some of the eddy-resolvingcapabilities, with the corresponding improvement in broadbandnoise predictions regarding URANS. For this reason, DES model isalso used in this paper, although it is still not very popular in tur-bomachinery simulations. Tucker (2011) listed 40 turbomachinerycases with eddy-resolving turbulence models, only 3 of them beinghybrid RANS-LES. In this work, DES will be performed in combina-tion with a SST k�x turbulence model, ‘‘which functions as a sub-grid-scale model in regions where the grid density is fine enoughfor a large-eddy simulation, and as a Reynolds-averaged model inregions where it is not’’ (Travin et al., 2000). Low-Reynolds numbernear-wall resolution is used, since Mendonça et al. (2012) showedthat this approach improved the prediction of the compressor mapcompared to experimental measurements when using steady-stateRANS k�x SST simulations. Particularly, IDDES (Travin et al.,2006; Shur et al., 2008) is used, which combines WMLES and DDEShybrid RANS-LES approaches. Mockett (2009) studied several DESmodels, concluding that IDDES extended the range of suitableapplications of DDES, by reducing the issue with log-layer mis-match (LLM). The mesh used in this paper may not be fine enoughto take advantage of the WMLES branch in IDDES, but at least thisturbulence model should perform as DDES. van Rennings et al.(2012) used DDES for simulations of an axial compressor near stall.Imposed geometrical periodicity was found to affect large scale

turbulent structures, concluding that the full compressor shouldbe modeled for the sake of accuracy.

Simulations using URANS and DES turbulence approaches wereperformed using the segregated solver of Star-CCM+ (CD-Adapco,2013). DES results using the described model has been already val-idated against experimental measurements for aeroacoustic pur-poses by Broatch et al. (2014). Navarro (2014) showed that LESmode of detached eddy simulations is present in the core of therotor passages and the recirculating flows upstream the impeller.A comparison between predicting capabilities of URANS and DESwill be performed in Section 4.

3.2. Tip clearance reduction approach

In Section 2, a maximum clearance reduction of 45% is esti-mated. In order to assess compressor performance indicators, flowfeatures and noise production sensitivity to tip clearance, threelevels are considered: original (CAD) clearance, 2/3 of originalclearance and 1/3 of original clearance. In this way, expected clear-ance reduction range is comprised in the sensitivity analysis.

Tip clearance reduction can be done with two opposedapproaches, either increasing impeller size or decreasing shrouddiameter. Impeller increase would mimic temperature and rota-tional deformation, but tip clearance reduction due to shaft motionrepresents an approximation regardless of the method used. How-ever, unrealistic deformation of impeller geometry may impactflow behavior more than subsequent reduction of tip clearance.In this paper it is thus preferred to keep impeller size and reducethe shroud to obtain the desired TCRs. In any case, Danish et al.(2006) proved that these two approaches to change tip clearanceyield similar results.

The reduction of shroud that causes the three different TCRsstudied in this paper can be observed in the left-hand side ofFig. 6. The three meshes produced with this approach present asimilar amount of elements, being about 9.5 million polyhedralcells.

4. Comparison of overall performance and acoustic signatures

4.1. Experimental setup

Aside from the numerical comparison between turbulencemodels and different TCRs, experimental spectra are used as a ref-erence. The experimental measurements shown in this paper wereobtained installing a turbocharger test rig in an anechoic chamber.A detailed description of the gas stand is given by Galindo et al.(2006).

50

50

42

1008 32

60.5

539

32

32

180 363 1495

150

210

1070

1 2 3

3 2 1

CFD domain

Sensor array

Sensor array

Fig. 7. Compressor piping layout, highlighting the CFD domain and the sensor arrays (dimensions in mm).

134 J. Galindo et al. / International Journal of Heat and Fluid Flow 52 (2015) 129–139

Long, straight pipes were placed at the inlet and outlet of thecompressor, as depicted in Fig. 7. Position of piezoelectric pressuresensor arrays are also included in the sketch. These probes wereused to sample pressure signals at 100 kHz during 1 s at thedesired operating condition, using a Yokogawa digital oscilloscope.

Inlet and outlet linear arrays of three piezoelectric sensors eachwere designed in accordance with Piñero et al. (2000), to takeadvantage of their beamforming-based signal analysis method todecompose pressure signals into a forward-traveling wave pforwðtÞand a backward-traveling wave pbackðtÞ.

A full description of the experimental apparatus used to obtainthe acoustical measurements is presented by Broatch et al. (2014).

4.2. Compressor performance variables

In first place, compressor global variables are compared. Total-to-total pressure ratio, specific work and isentropic efficiency,

Pt;t ¼pout;0

pin;0

Wu ¼_W_m¼

2pNðrpmÞ60 s

_m¼ cpðTout;0 � Tin;0Þ

gs ¼_Ws

_W¼

Tin;0 Pc�1c �1

t;t

� �Tout;0 � Tin;0

;

ð1Þ

are either experimentally measured or numerically calculated.Table 1 includes these global variables along with relative differ-ence against experimental measurements and relative differenceagainst CAD clearance case with the corresponding turbulencemodel, defined as:

�expð%Þ ¼/CFD � /exp

/exp� 100 �3=3ð%Þ ¼

/CFD � /CFD;3=3

/CFD;3=3� 100: ð2Þ

For both turbulence approaches, cases with original clearanceprovide compressor performance values closer to experimental

Table 1Compressor global variables measured in the experimental test rig and predicted by the nexperimental measurements and the relative difference against CAD clearance case with t

Case Pt;t [-] (//�exp/�3=3)

Exp. 2.22/–/–

URANS 3/3 2.21/�0.5/–2/3 2.26/1.7/2.21/3 2.31/4.1/4.7

DES 3/3 2.23/0.4/–2/3 2.27/2.1/1.61/3 2.32/4.2/3.8

results than cases with reduced clearance. Decreasing tip gapimproves compressor isentropic efficiency and pressure ratio. InURANS simulations, a reduction of TCR slightly increases specificwork, but in DES there is not a monotonous trend.

URANS case with CAD clearance predicts compressor perfor-mance parameters showing better agreement with experimentalmeasurements than DES, particularly in terms of isentropic effi-ciency. However, a better estimation of global variables doesnot necessarily imply that URANS is more accurate than DES inpredicting compressor noise generation, as the next section willprove.

4.3. Compressor acoustic spectra

In order to assess the effect of tip clearance on centrifugal com-pressor noise, the methodology developed by Broatch et al. (2014)is followed. Experimental signals were sampled during 1 s,whereas numerical monitors were stored at each case for morethan 50 ms (corresponding to 140 impeller revolutions) afterreaching a steady state in terms of global variables. Welch’s over-lapped segmented average (Welch, 1967) is used to estimate pres-sure PSD, which are the tool to evaluate and compare compressornoise generation. Hamming windowing is used at blocks with 50%overlap. The number of blocks is selected so as to obtain the fre-quency resolution closest to 150 Hz.

The acoustic analysis is performed dividing the spectra in tworegions: planar wave range and high frequency range.

Planar wave range starts at lowest frequency attainable and fin-ishes when the first asymmetric mode is cut on. In this range,numerical simulations require pressure decomposition to avoidthe nodes of standing waves. Experimental signals are decomposedused the beamforming-based technique described by Piñero et al.(2000), although it imposes a constraint on maximum frequency(Broatch et al., 2014). Pressure components from the simulationsare obtained by means of Eq. (3):

umerical model. In each box, the value is followed by the relative difference againsthe corresponding turbulence model.

Wu [kJ � kg�1] (//�exp/�3=3) gs [%] (//�exp/�3=3)

121/–/– 62.2/–/–

122/0.7/– 62.3/0.1/–123/1.2/0.5 63.9/2.7/2.6124/2.2/1.5 65.3/5.0/4.9

122/0.6/– 63.2/1.6/–122/0.1/�0.5 65.0/4.4/2.8122/0.7/0.1 66.5/6.8/5.1

100

105

110

115

120

125

Inle

t PS

D (

dB)

Exp. pback

J. Galindo et al. / International Journal of Heat and Fluid Flow 52 (2015) 129–139 135

pforw ¼ pref12

1þ ppref

!c�12c

1þ c� 12

ua

� �0@

1A

24

35

2cc�1

pback ¼ pref12

1þ ppref

!c�12c

1� c� 12

ua

� �0@

1A

24

35

2cc�1

ð3Þ

500 1000 1500 2000 2500 300085

90

95

Frequency [Hz]

CFD pback

3/3 (DES)

CFD pback

2/3 (DES)

CFD pback

1/3 (DES)

500 1000 1500 2000 2500 3000 3500 4000 4500 5000 550095

100

105

110

115

120

125

130

Frequency [Hz]O

utle

t PS

D (

dB)

Exp. pforw

CFD pforw

3/3 (DES)

CFD pforw

2/3 (DES)

CFD pforw

1/3 (DES)

Fig. 9. Low frequency PSD of experimental and DES pressure components at inlet(top) and outlet (bottom) ducts.

85

90

95

100

105

110

115

120

Inle

t PS

D (

dB)

Exp. pCFD p

back 3/3 (RANS)

CFD pback

2/3 (RANS)

which are inspired by the Method of Characteristics (MoC), asdescribed by Torregrosa et al. (2012). Flow field variables are aver-aged at cross-sections.

High frequency range starts with the onset of first asymmetricmode and is extended until 20 kHz. Point monitors located at wallducts should be used in this frequency range so as to capturehigher order acoustic modes (Broatch et al., 2014). Eq. (3) is againused to obtain the numerical spectra of pressure components inorder to reduce the impact of standing waves, whereas beamform-ing cannot be applied to experimental pressure traces and thus asingle probe of each array is used to obtain the raw PSD.

In this way, Figs. 8–11 are obtained. URANS and detached-eddysimulations were performed using the configuration described inSection 3.1.

Figs. 8 and 9 show that, at the inlet, experimental spectrum atlow frequency range decays monotonously. Inlet numerical spectrais predicted by both turbulence approaches alike. Simulations alsoprovide decreasing spectra, but amplitude is overpredicted. More-over, TCR does not have an impact on PSD.

At the outlet duct, experimental spectrum decreases until3.5 kHz, presenting a narrow band noise and a constant level after-wards. URANS PSD depicted in Fig. 8 fails to predict this change inbehavior, showing a constant decay for all TCRs. Conversely, DESspectra (Fig. 9) does present a narrow band centered at a frequencyfrom 3.5 to 4 kHz, depending on tip clearance. Then, PSD are con-stant. The only difference between DES spectrum and experimentalone is the broad band that simulations predict from 1.3 kHz to2.5 kHz. It corresponds to the whoosh noise phenomenon detectedat experiments for higher mass flows (see Broatch et al. (2014)).

500 1000 1500 2000 2500 300085

90

95

100

105

110

115

120

125

Frequency [Hz]

Inle

t PS

D (

dB)

Exp. pback

CFD pback

3/3 (RANS)

CFD pback

2/3 (RANS)

CFD pback

1/3 (RANS)

500 1000 1500 2000 2500 3000 3500 4000 4500 5000 550095

100

105

110

115

120

125

130

Frequency [Hz]

Out

let P

SD

(dB

)

Exp. pforw

CFD pforw

3/3 (RANS)

CFD pforw

2/3 (RANS)

CFD pforw

1/3 (RANS)

Fig. 8. Low frequency PSD of experimental and URANS pressure components atinlet (top) and outlet (bottom) ducts.

75

80

Frequency [kHz]

1 2

6 8 10 12 14 16 18 20

CFD pback

1/3 (RANS)

90

100

110

120

130

140

Frequency [kHz]

Out

let P

SD

(dB

)

1 2

8 10 12 14 16 18 20

Exp. pCFD p

forw 3/3 (RANS)

CFD pforw

2/3 (RANS)

CFD pforw

1/3 (RANS)

Fig. 10. High frequency PSD of experimental and URANS pressure at inlet (top) andoutlet (bottom) ducts.

The differences in spectra prediction between turbulenceapproaches increase when frequencies above planar-wave rangeare considered. Fig. 10 presents spectra provided by URANS simu-lation against experimental ones. At the inlet, the broadband eleva-tion from 5 to 12 kHz is clearly underpredicted, showing a betteragreement in the subsequent decay. PSD at the outlet duct isunderpredicted, particularly for frequencies below 10 kHz.

75

80

85

90

95

100

105

110

115

120

Frequency [kHz]

Inle

t PS

D (

dB)

1 2

6 8 10 12 14 16 18 20

Exp. pCFD p

back 3/3 (DES)

CFD pback

2/3 (DES)

CFD pback

1/3 (DES)

90

100

110

120

130

140

Frequency [kHz]

Out

let P

SD

(dB

)

1 2

8 10 12 14 16 18 20

Exp. pCFD p

forw 3/3 (DES)

CFD pforw

2/3 (DES)

CFD pforw

1/3 (DES)

Fig. 11. High frequency PSD of experimental and DES pressure at inlet (top) andoutlet (bottom) ducts.

Fig. 12. LIC of relative velocity vectors colored by pressure at isomeridionalsurfaces for DES cases with different clearance. Zero meridional velocity isrepresented as a solid black line.

136 J. Galindo et al. / International Journal of Heat and Fluid Flow 52 (2015) 129–139

DES (Fig. 11) provides an inlet spectrum with a closer estima-tion of the mean amplitude during the broadband elevation,although it fails to predict PSD decrease above 12 kHz. The narrowband between 4.7 and 6 kHz is not registered by the experimentalmeasurement. At the outlet duct, PSD is accurately reproducedusing the DES, save for a broadband elevation from 14 to 16 kHzthat the experiments include.

Even though planar wave spectra predictions for both turbu-lence approaches are alike, DES shows superior performance athigh frequencies. Only inlet spectrum above 12 kHz is overpre-dicted, and it does not represent a major issue since inlet PSD issignificantly lower than outlet one (average of 10 dB), and humanhearing sensitivity from 12 kHz onwards is limited.

5. Flow field investigation

In this section, flow field is investigated to explain why TCR isnot relevant to compressor noise generation in these near-surgeoperating conditions (mass flow rate is 1.4 times surge mass flowat the speed line of 160 krpm). Only detached-eddy simulationsare analyzed, since Section 4.3 has confirmed their (slightly) supe-rior performance over URANS in terms of noise prediction.

Fig. 12 shows the mean flow field for the cases with different tipclearance. The different fluid variables have been time-averagedduring more than 10 impeller rotations. Line Integral Convolution(LIC) (Cabral and Leedom, 1993) is used to represent relative veloc-ity vectors as an oil flow over isomeridional surfaces depicted inFig. 6. These surfaces of revolution have been flattened by project-ing them on a normalized-meridional vs. circumferential plane(Drela and Youngren, 2008). In Fig. 12, LIC is blended with pressurecontours.

The impeller rotation is transformed into a horizontal transla-tion from right to left in Fig. 12. Therefore, the left side of theblades corresponds to the PS whereas the right side is the SS, asconfirmed by pressure contours. The solid black line shown inFig. 12 represents a change of sign of meridional component ofvelocity. The majority of the impeller isomeridional section

presents incoming flow, save for the vicinity of the shroud, inwhich backflow appear. Since tip clearance is included in this recir-culating region, tip leakage flow is not only driven by favorablepressure gradient across the blade tip.

In order to shed some light on tip leakage flow, Fig. 13 showsthe flow field at isomeridional surface number 3 (see Fig. 6), forthe three cases with different tip clearance. Time-averaged flowfield is considered in the left side pictures, whereas the pictureson the right side represent a snapshot of the instantaneous flow.In addition to oil flow colored by pressure and a solid black lineindicating zero meridional velocity, as in Fig. 12, relative velocityvectors are depicted. It can be seen that some of these vectors indi-cate flow going to the SS of the passages, i.e., there are cells withrelative velocity oriented in the rotational direction. The regionin which relative tangential velocity points to the left is highlightedand delimited by a solid violet line in Fig. 13.

Left side of Fig. 13 shows that the time-averaged flow goes inthe streamwise direction and moves from SS to PS at the inner coreof the passages (close to the hub). Conversely, flow is going back-wards near the shroud as already noted, and a fraction of this recir-culating flow goes from passage PS to SS. In the shroud boundarylayer, absolute velocity is close to zero, so relative tangential veloc-ity points to the right. These changes in circumferential flow direc-tion results in the creation of secondary flows, such as the vorticesobserved in Fig. 13. Vortex cores are usually regions with localminimum of pressure (see for instance the instantaneous flow forthe case with 1/3 of the original clearance).

The air flowing in two opposite directions in the tip clearancethat can be observed in the right side of Fig. 13 creates a strongshear stress in this region, thus disabling the inviscid character oftip leakage flow (Storer and Cumpsty, 1991), at least for near-stalloperating conditions. If tip leakage mechanism were mainly invis-cid, tip clearance could be resolved with few spanwise nodes (lessthan 10) (Eum et al., 2004; Shum et al., 2000). In this paper, about20 cells were used to mesh the spanwise direction of tip clearance(see Fig. 6). Van Zante et al. (2000) already noted that tip grid caninfluence prediction of stall inception.

Fig. 13. LIC of relative velocity vectors colored by pressure and relative velocity vectors glyph at isomeridional surface number 3 for DES cases with different clearance. Zeromeridional velocity is represented as a solid black line whereas zero relative circumferential velocity is represented as a solid violet line. (For interpretation of the referencesto color in this figure legend, the reader is referred to the web version of this article.)

J. Galindo et al. / International Journal of Heat and Fluid Flow 52 (2015) 129–139 137

Left side of Fig. 13 indicates that mean tip leakage flow movesfrom the right to the left, and this is justified by the pressure gra-dient that exists at the different sides of the blade tip. However, ifthe snapshots at right side of Fig. 13 are considered, it can benoticed that instantaneous net tip flow can be oriented in the rota-tional direction. The reason of the unexpected behavior of tip leak-age flow can be explained by Fig. 14, which depicts the mean flowin a meridional view at midpitch distance between main blade SSand splitter blade PS. The solid black line representing change inmeridional velocity direction shows that recirculating flow nearthe shroud almost extends to the trailing edge of the impeller forthese operating conditions, although this conclusion could also

Fig. 14. LIC of relative velocity vectors colored by tangential velocity and relativevelocity vectors glyph at meridional surface for baseline clearance DES. Zeromeridional velocity is represented as a solid black line whereas zero relativecircumferential velocity is represented as a solid violet line. (For interpretation ofthe references to color in this figure legend, the reader is referred to the web versionof this article.)

be drawn from Fig. 12. Highest tangential velocities are obtainedat the blade trailing edge, since it is the place where blade tipvelocity is greatest. The backflow transports the angular momen-tum that the impeller transmitted to the flow when it was goingin the streamwise direction. When the flow is reversed, it will losesome angular momentum because of the shear stress with theincoming flow, but also it will also gain tangential velocity, asradius is reduced.

Fig. 14 shows that mean tangential velocity is in the order ofblade tip velocity near the shroud, which explains the existenceof regions in which relative velocity is oriented towards the pas-sage SS (marked with a solid violet line in Fig. 14). Flow leakingfrom blade PS to SS through the tip is therefore diminished dueto angular momentum of backflows. This fact may explain thereduction of clearance flow rate detected by Liu et al. (2013) asoperating condition is shifted towards surge.

Some researchers have studied tip clearance noise (TCN), whichis attributed to jet-like tip leakage flow, the roll-up of this streaminto a vortex and the interaction of these phenomena with adja-cent blades (Jacob et al., 2010). TCN has been thoroughly investi-gated in axial turbomachinery (Fukano and Jang, 2004; Kameierand Neise, 1997), but it is not so well know in the case of radial tur-bomachinery. Raitor and Neise (2008) reported a narrow bandnoise in a centrifugal compressor that was considered as TCNbecause it presented similarities to the case of axial turbomachin-ery. Bousquet et al. (2014) found a periodic vortex shedding due totip leakage flow in their simulation of a centrifugal compressor atnear stall conditions, which may produce a narrow band noise.However, no recirculating flow appeared near the compressorshroud.

In order to produce a narrow ban noise, tip vortex sheddingshould occur in a freestream flow without any other strong fluctu-ating phenomena. For the operating conditions studied in thispaper, the whole tip is immersed in the recirculating region, whichwould mask any noise that may produce tip leakage or even dis-able the noise source mechanisms. In the simulations, it couldnot be found any periodic tip vortex shedding nor acoustic fluctu-ations that could be related to tip leakage flow. Only rotating stallwas found, but it is a subsynchronous phenomenon that produces

138 J. Galindo et al. / International Journal of Heat and Fluid Flow 52 (2015) 129–139

the so-called whoosh noise (see the work of Broatch et al. (2015)),and according to Figs. 8–11 TCR does not affect whoosh noise nei-ther. The modification of tip leakage behavior due to the existenceof a strong swirling recirculating flow near the shroud is the pro-posed mechanism to explain that noise generated by the centrifu-gal compressor is not affected by TCR, at least for the near-stalloperating conditions studied in this paper.

6. Concluding remarks

In this paper, tip clearance sensitivity to flow field and noisegeneration of a 49 mm exducer diameter centrifugal compressorhas been investigated. Measurements of turbocharger shaft motionshow that, for a compressor speed of 160 krpm, tip clearance canbe modified by a factor of 22–34%. FEM predicts that temperatureand rotation cause an additional tip clearance reduction of 11%.

3D CFD simulations with three different tip clearance levelsshow that clearance reductions cause increases in pressure ratioand isentropic efficiency, but no significant changes in compressoracoustic signature. Two different turbulence approaches (URANSand DES) are studied, and DES is found to provide spectra with bet-ter agreement against experimental measurements, particularly athigh frequencies (above onset of higher order acoustic modes).

Insensitivity of spectra to TCR is explained by flow field config-uration. Since the working point is close to surge, the flow in thevicinity of the shroud is reversed and tip leakage flow is not onlydriven by pressure gradient at blade tip. A complicated flow pat-tern is detected in this region, but the recirculating flow comprisestip clearance. A coherent noise source mechanism due to tip leak-age flow cannot be established, so variations of TCR do not have animpact on noise generation. Conversely, compressor performanceindicators are affected by TCR, because tighter tip clearances leadto smaller backflow regions and reduced mixing losses.

These conclusions are valid for the studied operating condi-tions. Compressors working with higher mass flow rates shouldpresent less backflows close to the shroud and tip sensitivity tonoise generation may be different. In any case, the present papershows that, for operating conditions close to surge, researcherscan rely on CAD tip clearance profile when analyzing acousticbehavior of centrifugal compressors.

Acknowledgements

The equipment used in this work has been partially supported byFEDER project funds ‘‘Dotación de infraestructuras científico técni-cas para el Centro Integral de Mejora Energética y Medioambientalde Sistemas de Transporte (CiMeT), (FEDER-ICTS-2012-06)’’, framedin the operational program of unique scientific and technical infra-structure of the Ministry of Science and Innovation of Spain. Theauthors wish to thank Mr. Pau Raga for his worthy assistance duringthe meshing process.

References

Bousquet, Y., Carbonneau, X., Dufour, G., Binder, N., Trebinjac, I., 2014. Analysis ofthe unsteady flow field in a centrifugal compressor from peak efficiency to nearstall with full-annulus simulations. Int. J. Rotating Mach. 2014, 11. http://dx.doi.org/10.1155/2014/729629.

Broatch, A., Galindo, J., Navarro, R., García-Tíscar, J., 2014. Methodology forexperimental validation of a CFD model for predicting noise generation incentrifugal compressors. Int. J. Heat Fluid Flow. http://dx.doi.org/10.1016/j.ijheatfluidflow.2014.06.006.

Broatch, A., Galindo, J., Navarro, R., García-Tíscar, J., Daglish, A., Sharma, R.K., 2015.Simulations and measurements of automotive turbocharger compressorwhoosh noise. Eng. Appl. Comput. Fluid Mech. (accepted for publication)

Cabral, B., Leedom, L.C., 1993. Imaging vector fields using line integral convolution.In: Proceedings of the 20th Annual Conference on Computer Graphics andInteractive Techniques. ACM, pp. 263–270.

CD-Adapco. STAR-CCM+, Release Version 8.04.007 Edition (June 2013). <http://www.cd-adapco.com>.

Danish, S.N., Chaochen, M., Ce, Y., Wei, L., 2006. Comparison of two methods toincrease the tip clearance and its effect on performance of a turbochargercentrifugal compressor stage. In: International Conference on Energy andEnvironment, pp. 7.

Drela, M., Youngren, H., 2008. A User’s Guide to MISES 2.63, MIT AerospaceComputational Design Laboratory.

Dufour, G., Gourdain, N., Duchaine, F., Vermorel, O., Gicquel, L., Boussuge, J.F.,Poinsot, T., 2009. Large eddy simulation applications. In: NumericalInvestigations in Turbomachinery: A State of the Art, VKI Lecture Series.

Eum, H.-J., Kang, Y.-S., Kang, S.-H., 2004. Tip clearance effect on through-flow andperformance of a centrifugal compressor. KSME Int. J. 18 (6), 979–989. http://dx.doi.org/10.1007/BF02990870.

Evans, D., Ward, A., 2005. Minimizing Turbocharger Whoosh Noise for DieselPowertrains. SAE Technical Paper 2005-01-2485. http://dx.doi.org/10.4271/2005-01-2485.

Fukano, T., Jang, C.-M., 2004. Tip clearance noise of axial flow fans operating atdesign and off-design condition. J. Sound Vib. 275 (3), 1027–1050. http://dx.doi.org/10.1016/S0022-460X(03)00815-0.

Galindo, J., Serrano, J.R., Guardiola, C., Cervelló, C., 2006. Surge limit definition in aspecific test bench for the characterization of automotive turbochargers. Exp.Therm. Fluid Sci. 30 (5), 449–462. http://dx.doi.org/10.1016/j.expthermflusci.2005.06.002.

Galindo, J., Serrano, J.R., Dolz, V., López, M.A., 2013. Behavior of an IC engineturbocharger in critical condition of lubrication. SAE Int. J. Engines 6. http://dx.doi.org/10.4271/2013-01-0921.

Hellström, F., Guillou, E., Gancedo, M., DiMicco, R., Mohamed, A., Gutmark, E., Fuchs,L., 2010. Stall Development in a Ported Shroud Compressor using PIVMeasurements and Large Eddy Simulation. Tech. Rep., SAE Technical Paper2010-01-0184. http://dx.doi.org/10.4271/2010-01-0184.

Hellstrom, F., Gutmark, E., Fuchs, L., 2012. Large eddy simulation of the unsteadyflow in a radial compressor operating near surge. J. Turbomach. 134 (5). http://dx.doi.org/10.1115/1.4003816.

Jacob, M.C., Grilliat, J., Camussi, R., Gennaro, G.C., 2010. Aeroacoustic investigationof a single airfoil tip leakage flow. Int. J. Aeroacoust. 9 (3), 253–272.

Jung, Y., Choi, M., Oh, S., Baek, J., 2012. Effects of a nonuniform tip clearance profileon the performance and flow field in a centrifugal compressor. Int. J. RotatingMach. http://dx.doi.org/10.1155/2012/340439.

Jyothishkumar, V., Mihaescu, M., Semlitsch, B., Fuchs, L., 2013. Numerical flowanalysis in centrifugal compressor near surge condition. In: Fluid Dynamics andCo-Located Conferences. American Institute of Aeronautics and Astronautics, p.13. http://dx.doi.org/10.2514/6.2013-2730.

Kameier, F., Neise, W., 1997. Experimental study of tip clearance losses and noise inaxial turbomachines and their reduction. J. Turbomach. 119 (3), 460–471.

Karim, A., Miazgowicz, K., Lizotte, B., Zouani, A., 2013. Computational Aero-Acoustics Simulation of Compressor Whoosh Noise in AutomotiveTurbochargers. SAE Technical Paper 2013-01-1880. http://dx.doi.org/10.4271/2013-01-1880.

Liu, Z., Ping, Y., Zangeneh, M., 2013. On the nature of tip clearance flow in subsoniccentrifugal impellers. Sci. China Technol. Sci. 56 (9), 2170–2177. http://dx.doi.org/10.1007/s11431-013-5313-3.

López, M.A., 2014. Estudio teórico-experimental de la dinámica rotacional deturbocompresores de MCIA. Aplicación al diagnóstico de fallos., Ph.D. Thesis.Universitat Politècnica de València.

Mendonça, F., Baris, O., Capon, G., 2012. Simulation of radial compressor aeroacousticsusing CFD. In: Proceedings of ASME Turbo Expo 2012, No. GT2012-70028. ASME,pp. 1823–1832. http://dx.doi.org/10.1115/GT2012-70028.

Menter, F.R., 1994. Two-equation eddy-viscosity turbulence models for engineeringapplications. AIAA J. 32 (8), 1598–1605.

Menter, F.R., Langtry, R., Hansen, T., 2004. CFD simulation of turbomachinery flows-verification, validation and modeling. In: European Congress on ComputationalMethods in Applied Sciences and Engineering, ECCOMAS.

Mockett, C., 2009. A Comprehensive Study of Detached-Eddy Simulation. Ph.D.Thesis. Technische Universität Berlin.

Navarro, R., 2014. A Numerical Approach for Predicting Flow-Induced Acoustics atNear-Stall Conditions in an Automotive Turbocharger Compressor. Ph.D. Thesis.Universitat Politècnica de València.

Nelson, F., 2007. Rotordynamics without equations. Int. J. COMADEM 10 (3), 2–10.Pastor, J.V., Serrano, J.R., Dolz, V., López, M.A., Bouffaud, F., 2012. Study of

turbocharger shaft motion by means of non-invasive optical techniques:application to the behaviour analysis in turbocharger lubrication failures.Mech. Syst. Signal Process. 32, 292–305. http://dx.doi.org/10.1016/j.ymssp.2012.04.020.

Pastor, J.V., Serrano, J.R., Dolz, V., López, M.A., 2012. A non-invasive opticaltechnique to observe turbocharger shaft whirl. In: XXI Biennial Symposiumon Measuring Techniques in Turbomachinery. ISBN: 978-84-8363-966-5.

Piñero, G., Vergara, L., Desantes, J., Broatch, A., 2000. Estimation of velocityfluctuation in internal combustion engine exhaust systems throughbeamforming techniques. Meas. Sci. Technol. 11 (11), 1585–1595. http://dx.doi.org/10.1088/0957-0233/11/11/307.

Raitor, T., Neise, W., 2008. Sound generation in centrifugal compressors. J. SoundVib. 314, 738–756. http://dx.doi.org/10.1016/j.jsv.2008.01.034.

Robinson, C., Casey, M., Hutchinson, B., Steed, R., 2012. Impeller-diffuser interactionin centrifugal compressors. In: Proceedings of ASME Turbo Expo 2012, No.GT2012-69151. ASME.

J. Galindo et al. / International Journal of Heat and Fluid Flow 52 (2015) 129–139 139

Semlitsch, B., JyothishKumar, V., Mihaescu, M., Fuchs, L., Gutmark, E., Gancedo, M.,2014. Numerical Flow Analysis of a Centrifugal Compressor with Ported andwithout Ported Shroud. SAE Technical Paper 2014-01-1655. http://dx.doi.org/10.4271/2014-01-1655.

Shum, Yu Kwong Patrick, Cumpsty, N.A., Tan, C.S., 2000. Impeller–diffuserinteraction in a centrifugal compressor. J. Turbomach. 122 (4), 777–786.http://dx.doi.org/10.1115/1.1308570.

Shur, M.L., Spalart, P.R., Strelets, M.K., Travin, A.K., 2008. A hybrid RANS-LESapproach with delayed-DES and wall-modelled LES capabilities. Int. J. HeatFluid Flow 29 (6), 1638–1649. http://dx.doi.org/10.1016/j.ijheatfluidflow.2008.07.001.

Smirnov, P.E., Hansen, T., Menter, F.R., 2007. Numerical simulation of turbulentflows in centrifugal compressor stages with different radial gaps. In:Proceedings of GT2007, No. GT2007-27376. ASME. http://dx.doi.org/10.1115/GT2007-27376.

Sotome, T., Sakoda, S., 2007. Development of manufacturing technology forprecision compressor wheel castings for turbochargers. Furukawa Rev. 32,56–59, <https://www.furukawa.co.jp/review/fr032.htm>.

Storer, J.A., Cumpsty, N.A., 1991. Tip leakage flow in axial compressors. J.Turbomach. 113 (2), 252–259. http://dx.doi.org/10.1115/1.2929095.

Teng, C., Homco, S., 2009. Investigation of compressor whoosh noise in automotiveturbochargers. SAE Int. J. Passeng. Cars-Mech. Syst. 2 (1), 1345–1351. http://dx.doi.org/10.4271/2009-01-2053.

Tomita, I., Ibaraki, S., Furukawa, M., Yamada, K., 2013. The effect of tip leakagevortex for operating range enhancement of centrifugal compressor. J.Turbomach. 135 (5), 8. http://dx.doi.org/10.1115/1.4007894.

Torregrosa, A.J., Fajardo, P., Gil, A., Navarro, R., 2012. Development of a non-reflecting boundary condition for application in 3D computational fluiddynamic codes. Eng. Appl. Comput. Fluid Mech. 6 (3), 447–460.

Travin, A.K., Shur, M.L., Strelets, M.K., Spalart, P.R., 2000. Detached-eddy simulationspast a circular cylinder. Flow Turbul. Combust. 63 (1-4), 293–313.

Travin, A.K., Shur, M.L., Spalart, P.R., Strelets, M.K., 2006. Improvement of delayeddetached-eddy simulation for les with wall modelling. In: European Conferenceon Computational Fluid Dynamics (ECCOMAS CFD 2006).

Tucker, P.G., 2011. Computation of unsteady turbomachinery flows: Part 2–LES andhybrids. Prog. Aerosp. Sci. 47 (7), 546–569. http://dx.doi.org/10.1016/j.paerosci.2011.07.002.

van Rennings, R., Shi, K., Fu, S., Thiele, F., 2012. Delayed-detached-eddy simulationof near-stall axial compressor flow with varying passage numbers. In: Fu, S.,Haase, W., Peng, S.-H., Schwamborn, D. (Eds.), Progress in Hybrid RANS-LESModelling, Notes on Numerical Fluid Mechanics and Multidisciplinary Design,vol. 117. Springer, Berlin, Heidelberg, pp. 439–448. http://dx.doi.org/10.1007/978-3-642-31818-4_38.

Van Zante, D.E., Hathaway, M.D., Okiishi, T.H., Strazisar, A.J., Wood, J.R., 2000.Recommendations for achieving accurate numerical simulation of tip clearanceflows in transonic compressor rotors. J. Turbomach. 122 (4), 733–742. http://dx.doi.org/10.1115/1.1314609.

Wagner, C.A., Hüttl, T., Sagaut, P. (Eds.), 2007. Large-Eddy Simulation for Acoustics.Cambridge University Press.

Wang, H.-L., Xi, G., Li, J.-Y., Yuan, M.-J., 2011. Effect of the tip clearance variation onthe performance of a centrifugal compressor with considering impellerdeformation. Proc. Inst. Mech. Eng., Part A: J. Power Energy 225 (8), 1143–1155. http://dx.doi.org/10.1177/0957650911416914.

Welch, P., 1967. The use of fast fourier transform for the estimation of powerspectra: a method based on time averaging over short, modified periodograms.IEEE Trans. Audio Electroacoust. 15 (2), 70–73.