Journal of Thermal Science Vol.19, No.1 (2010) 34−41

DOI: 10.1007/s11630-010-0034-4

Article ID: 1003-2169(2010)01-0034-08

Unsteady Pressure Measurements in a High-Speed Centrifugal Compressor N. Bulot, X. Ottavy, and I. Trebinjac Laboratoire de Mécanique des Fluides et d’Acoustique (LMFA), UMR CNRS 5509 Ecole Centrale de Lyon, UCBLyon I, INSA 36 av. Guy de Collongue, 69134 Ecully cedex, France Tel : 33-4-72-18-61-83, Fax : 33-4-78-64-71-45, E-mail : [email protected] © Science Press and Institute of Engineering Thermophysics, CAS and Springer-Verlag Berlin Heidelberg 2010

This paper presents the unsteady data acquisition system used to measure the pressure field in high speed compressors. Details and electronic sketches are given for the conditioners developed in-house that have been used to amplify and to filter the pressure signal with the aim of acquiring data up to 150 kHz. A discussion of the experimental results carried out in a centrifugal compressor is proposed. Through different processing of the pressure signals and a comparison with URANS simulations, the excitation of the pressure transducers by the pressure waves generated by shock waves that occur between the impeller and the diffuser is highlighted. The levels of pressure fluctuations measured when entering into surge are also presented and reveal very repetitive behaviour of the flow instabilities.

Keywords: unsteady pressure measurements, conditioner, high frequency, high speed, centrifugal compressor.

Introduction Next generation engine design tends towards compact, highly efficient and large operability configurations. The operating range of high speed compressors is limited by choking at high mass flow rates and by the onset of instabilities at low mass flow rates. A safety margin, known as the ‘surge margin’, prevents the compressor from operating close to the stability limit, precisely where the pressure ratio reaches its highest level. Therefore, in order to increase the operating range, there is great interest in predicting the condition at which instability will occur in a compressor. That requires reaching a comprehensive understanding of the physical phenomena which trigger rotating stall and/or surge. Over the last few years the development of numerical

tools such as Computational Fluid Dynamics and high performance parallel computers has led to significant progress, enhancing the understanding of flows in turbomachines. Numerical methods and various models have also been proposed to reduce the cost and to permit unsteady calculations. But in most of the cases, strong assumptions concerning the spatial and time scales limit the validity domain around design conditions, and the results obtained at non-stable operating points have to be validated by experimental results. Experimental validation is all the more necessary as the case under investigation is a realistic high-speed machine with high complexity induced by technological effects. The Turbomachinery Team of the Fluid Mechanics Acoustic Laboratory (LMFA) at Ecole Centrale de Lyon (ECL) has two high-speed compressor test rigs dedicated

Received: November 2009 Xavier OTTAVY: Dr. Permanent Researcher www.springerlink.com

N. Bulot et al. Unsteady Pressure Measurements in a High-Speed Centrifugal Compressor

to studies of the flow in a 3½-stage axial compressor named CREATE designed and built by Snecma, and in a transonic centrifugal compressor stage TM designed and built by Turbomeca. Several works [1][2][3][4] have been published on detailed measurements acquired in these machines using pressure and temperature probes and a laser anemometer. Ensemble averages of the data acquired with the laser anemometry technique lead to an interesting description of the flow if only the phenomena phased with the rotation speed of the machine are under investigation. When looking at the onset of instabilities, such measurement techniques are not suitable. Instantaneous time-dependent information can be obtained using fast-response pressure sensors. A detailed review which concentrates on the silicon piezoresistive sensor is proposed by Ainsworth et al. [5]. It is nowadays common to use pressure measurements to study the unsteady flow phenomena in compressors, especially in the presence of rotating stall, rotating instability, or clearance effects, see for example [6][7][8]. Nevertheless, pressure measurements in high speed machines, with high temporal resolution (above 100 kHz) are still very rare. Let us mention the works carried out at Technische Universität at Darmstadt [9] in a high speed compressor with a sampling rate of 125 kHz, the works of Justen et al. [10] concerning unsteady measurements of impeller-diffuser interactions with a sampling rate of 10 MHz but with a natural frequency of their pressure sensor of only 200 kHz, and the works of Higashimori et al. [11] who have carried out a detailed investigation in a high transonic impeller using pressure transducers with a sampling rate of 1 MHz (but averaging the data every two points, which is equivalent to 500 kHz) and a natural frequency of 600 kHz. The need for time-dependent information led the LMFA to develop skills in acquiring unsteady pressure data at different locations in the compressors with a high temporal resolution. The data signals, using up to 48 channels, have to be amplified, filtered and acquired simultaneously in synchronization with the rotation of the machine. The commercial solutions dedicated to this kind of application exist, but they are really too expensive, and details of the data processing are most of the time not given by the manufacturers. This paper presents the in-house development of a new data acquisition and data reduction system together with the methodology used for unsteady pressure measurement in high-speed compressors.

Unsteady data acquisition system Due to the expected time scales within a high-speed compressor, the acquisition data rate has to be high enough to reach a sufficient time resolution of the pres-

35

sure signal at any given stable operating point, but also high enough in order to record the disturbances which may lead to the onset of instabilities when moving towards surge. Taking this into account, our aim was to be able to measure unsteady pressure with a high level of frequency, typically reaching 150 kHz. A specific unsteady pressure acquisition system has been developed and is presented in Fig. 2. This system is split into three parts: the pressure transducer, the signal conditioner and the digitizing and storing system. The pressure transducer The first part concerns the pressure transducers whose main features and requirements may be found in the review of Ainsworth et al.[4]. Not only the time resolution but also its resistance to high temperatures has to be checked. Temperature can be increased by around 350 K in this machine. High frequency response transducers from Kulite (Fig. 1) were chosen, without a protective grid. Their natural frequency calculated by the manufacturer is 300 kHz, but judging from the results obtained in a shock tube and in our compressor, this natural frequency is rather around 440 kHz. This normally permits measurements to be made up to one third of the natural frequency, i.e. in the range [0;145 kHz]. Each Kulite sensor is excited by an accurate DC power supply using batteries and an electronic device developed in-house. The excitation is set to 12 V (instead of the 10 V recommended by the manufacturer) and causes an increase of the sensitivity of 20%, around 35.1 mV/bar for our sensors.

Fig. 1 Kulite XTEH-7L-190M-50PSIA without protective grid

The signal conditioner The second part of the unsteady data acquisition system is related to the signal conditioning. Each Kulite pressure transducer is connected with a differential input to a conditioner developed in-house, (see the sketch on the left of Fig. 3). This conditioner is a precision DC bridge amplifier with a built-in constant excitation voltage/current supply (5V) and a low-pass filter with a cutoff frequency of 250 kHz. The core of the filter is the LTC1569-7 component from Linear Technology. This is a linear phase, DC accurate and tunable 10th order low-pass filter. Using various resistor values and divider settings, the cutoff frequency can be programmed over a range of seven octaves. The mean signal voltage reaching the filter must, how-

36

J. Therm. Sci., Vol.19, No.1, 2010

Fig. 2 Unsteady data acquisition system

between input and output signals is neglected because it is lower than 0.5% of the blade passing frequency which is a bit less than 10 kHz in our case.

Fig. 3 Sketch of the conditioner developed in-house

ever, be around 500 mV to obtain optimum results, so a DC bridge amplifier was inserted before the filter. Different components for the amplifier have been tested. As inputs of the low pass filter, it should be noted that some of them could introduce large fluctuations of the signal amplitude when the frequency changes. We finally chose the component LT1102 from Linear Technology which is a high speed, precision, JFET input instrumentation amplifier with a fixed gain equal either to 10 or 100. With our setting the response of the conditioner is not perfect. Below 250 kHz, the output signal amplitude is distorted compared with the input signal amplitude. For each of the 48 filters a calibration was achieved, i.e. the transfer functions were measured and stored. An example of the response of a channel of the conditioner is plotted versus the frequency in Fig.4 and compared with the theoretical dashed line. The Fourier transform of the filtered pressure signal is then restored by dividing the Fourier transform of the distorted output signal by the transfer function. It is worth noticing that the phase shift

Fig. 4 Transfer function associated to a channel of the conditioner developed in-house

The digitizing and storing system The acquisition system was built around the National Instruments PXI Modular Instrumentation System. The PXI-1062Q system was selected because it offers rugged, shielded construction that provides a low-noise environment for data acquisition and signal conditioning. The core of the system is a set of 8 PXI-6123 data acquisition boards with a sampling frequency of 500 kHz. Each board accepts 8 differential analog inputs, for 8 Kulite connections. The boards have 16-bit resolution and communicate with the data acquisition computer (Dell T7400) via 2 communication boards (NI PXIe-8370 in the PXI and NI PCIe-8371 in the station). The files are stored in binary format. As an example, the size of the file containing 20 seconds of the signals coming from 48 pressure transducers reaches 1 Go. Specific software has been developed to acquire, store, read and post-treat the data. To perform the voltage/pressure conversion, classical calibrations and zero drift corrections (before and after the tests) were applied.

Test rig Unsteady pressure measurements are currently carried out in the high-speed centrifugal compressor. This compressor is composed of a backswept splittered unshrouded impeller and a vaned diffuser. At the design operating point its pressure ratio reaches 9.0 and its rotation speed is greater than 50,000 rpm. In this machine, the surge is thought to initiate in the diffuser passages. Indeed it has been shown [3] that the diffuser was the most sensitive component to unsteadiness which came from two sources: the jet-wake structure (emanating from the impeller flow) and the interaction between the diffuser vane bow shock wave and the impeller blade pressure surface leading to pressure waves which propagate in the diffuser passages. Thus, high-frequency pressure transducers were located in the vaneless space (4 transducers), semi-vaneless space and diffuser throat (11 transducers). Fig. 5 shows a

N. Bulot et al. Unsteady Pressure Measurements in a High-Speed Centrifugal Compressor

sketch of the centrifugal compressor and the location of the 4 sensors in the vaneless space. The 4 sensors were actually spread over the circumference of the machine. Their position relative to the vaned diffuser inter blade channel is plotted in this figure.

Fig. 5

3D sketch of the centrifugal compressor (left) and location of the 15 sensors in the vaneless space (right)

Pressure measurements Fig. 6 presents the experimental pressure ratio as a function of the mass flow rate of the compressor, for a rotation speed set to 0.927 of its design speed. The white circles indicate the operating points where unsteady pressure data have been acquired and the diamonds are for the URANS simulations. In this section the pressure measurements are presented in three parts. First, the processing to remove the non-physical part of the measured data is presented. Secondly, the measurements at peak efficiency, i.e. at a stable operating point where ensemble averages can be processed, will be discussed through comparison with the CFD results. Lastly, pressure signals acquired during the onset of the instabilities are presented (see the dashed line in Fig. 6).

First, let us mention that a low voltage unsteady signal is very sensitive to contamination due to electric ground loops and radio frequencies. Close to the high power electrical engine (500 kW) that drives the compressor, all precautions were taken to properly shield all signal-carrying wires. Nevertheless some signals exhibited contamination and spurious spikes that could not be associated with flow field pressure changes. These spurious spikes can interfere when treating the raw data signals, but they appear with a very low frequency compared to the blade passing frequency and are then negligible in an ensemble average. Data were acquired in synchronization with the rotation of the machine (note that the signal contaminations mentioned above can easily be found in the sinus-like shape of the synchronization signal). During acquisition the sampling rate is constant, but the rotation speed of the machine may fluctuate by 0.5%. The acquired data then needed to be treated in order to synchronize the data with each rotor turn. This was carried out using the trigger signal saved with the pressure signals, which accurately define the time origin of each rotor turn. The results presented in Fig. 7 and Fig. 8 show the benefit to the spectrum obtained when applying this processing. Not only the sharpness of the frequency spikes but also their amplitude (from 0.0088 to 0.030) are corrected. The obvious thing to do is to apply this processing when trying to quantify the rotor/stator interaction modes using ensemble averages. Fig. 9 presents a comparison of the time fluctuations of the pressure signals between the CFD solution and the

Fig. 7 Spectra without time processing of the data

Fig. 6 Compressor characteristics for rotation speed equal to 0.927% of the design speed

Processing of the raw data This processing is presented through the example of the raw data acquired with the sensor labeled “B10” in the inter-grid region (see Fig. 5).

37

Fig. 8 Spectra with time processing of the data

38

J. Therm. Sci., Vol.19, No.1, 2010

(a) pressure signal

(b) Standard deviation of the pressure signal.

Fig. 9 Pressure signals with the B10 sensor (a) and its standard deviation (b)

experimental results obtained in three cases: the raw signal, the signal filtered by our conditioner (with fcut = 250 kHz) and the latter signal filtered again, using software, at one third of the natural frequency of the sensor (i.e. f1/3 = 145 kHz). All these signals are obtained from ensemble averages and are interpolated with a zero padding technique during their reconstruction. The x-axis is the normalized time and represents the blade passage period. The y-axis is the pressure which is divided by its temporal averaged value. The raw signal shows fluctuations with a high frequency, which still remain after an ensemble average triggered by the blade passage. Here the pressure sensor is excited by the two massive pressure wave fronts (see the next section “Measurements at peak efficiency”) and it behaves almost as if it were in a shock tube configuration. High frequencies in a range up to the sensor’s natural frequency are amplified and aliasing problems appear. The ensemble average obviously just keeps the frequencies which are multiples of the blade passing frequency. This results in a high non-physical level of the standard deviation. Using the conditioner (signal with black dots) permits frequencies higher than 250 kHz (half the sampling frequency) in the original signal to be filtered out before it is sampled. The Nyquist–Shannon sampling theorem can therefore be verified and an important part of the aliasing problems be removed, thus resulting in a clearly smaller value of the standard deviation of this signal. However the resonance of the sensor pollutes the signal in the [145;250] kHz range. That explains why frequencies higher than one third of the natural frequency are finally removed (see the thick grey curve). Some oscillations in this pressure signal with frequencies higher than 120 kHz exist, but their physical interpretation should be made with caution. Note that the aforementioned data processing - to synchronize the pressure signals accurately with each rotor turn - has obvious effects on the value of the standard

dard deviation calculated in the region of high gradients, as in the case of the pressure wave fronts. The CFD solution resulting from a URANS simulation (see the curve with the square symbols) is finally compared with the last filtered pressure signal distribution. The global shape and the amplitude of the fluctuations are consistent and help to show the reliability of the results. Refer to [3] for more information about the URANS calculation. Measurements at peak efficiency Fig. 10 presents the ensemble averages of the measured pressure of the 4 sensors (B59, B13, B51, B10) located in the inter-row region (cf. Fig. 5), when the compressor is operating at peak efficiency. The x-axis represents a time period of the rotor TR*. The y-axis is for the pressure temporal fluctuations. The y-scale has been adapted and the mean values of the pressure have been modified so as to plot the pressure fluctuations at the spatial location of their sensor. As a consequence the length of the y-axis represents a spatial period of the vaned diffuser Θ*. It is then possible to plot in this “s-t like” diagram the characteristic direction of an event moving with the rotation speed of the impeller (see the grey dashed line labeled Ωrotor). The global diagram of the pressure distribution shows a high level of fluctuations generated by the pressure waves brought about by the interaction between the vane bow shock waves and the impeller main and splitter blades. In order to help understanding Fig. 10, the fluctuations induced by the pressure waves are highlighted in Fig. 11 with CFD results, plotting the pressure gradient times the velocity: the sign of that variable indicates if the pressure gradient has the same direction as the velocity vector, which allows the favorable or unfavorable pressure gradients to be identified. Fig. 11 gives six time steps of the pressure gradient in the impeller-diffuser interaction zone, at 50% blade height. The white curve shows the vane leading edge bow shock wave. The black

N. Bulot et al. Unsteady Pressure Measurements in a High-Speed Centrifugal Compressor

Fig. 10

Measured pressure signals of the 4 sensors located in the inter-row region

Fig. 11 Temporal evolution of the pressure gradient (in the flow direction) at 50% blade height

curves and dotted white curves mark out the α+ and α− pressure waves, respectively. The first time step map (Fig. 11 (a)) shows the shock wave just before it is chopped by the impeller blade. At the next time step (Fig. 11(b)), the trailing edge intersects the strong part of the shock wave which is thus reflected on the blade pressure surface, leading to an α+ wave (labeled α+2), then to an α− wave (visible in the third time step map, Fig. 11(c)) emerging from reflection of the α+ wave on the vane leading edge. A wave labeled α+1 can be observed in Fig.11(b). This wave propagates without any obstacle up to the fourth

39

time step, with a propagation speed, in the circumferential direction, equal to 1.7 times the rotation speed of the impeller. This value has been calculated from the CFD results and is in very good agreement with the propagation speed deduced from the measured data (see black dashed lines labeled Ωα+ in Fig. 10). At the fifth time step (Fig. 11 (e)), the wave passes the shock wave and hits the vane leading edge, leading to the α−1−b wave which rotates backwards. In the last map (Fig. 11 (f)), the α+1 wave has been cut into two branches α+1−a and α+1−b. The α+1−b branch moves quicker than the α+1−a branch because it is in a supersonic flow while α+1−a is in a subsonic flow. Note that within the vaneless space, the α+ pressure waves have a limited lifetime, of about one time period of the rotor. This is the reason why it is possible to reveal them using an ensemble average. Measurements during surge inception Raw pressure measurements when entering into surge are low pass filtered with f1/3 = 145 kHz and are presented in Fig. 12. This pressure signal is obtained with the sensor B10 and is normalized by the mean temporal value obtained during the stabilized operating point just before the surge. The x-axis represents the number of impeller revolutions, which is limited from a zero value arbitrarily chosen as 1400 in order to see the first surge cycle with its duration of about 850 revolutions. In this figure, as in the following ones, the white curves are for the averages over one rotor revolution. During the first rotor revolutions in the surge, the level of the pressure spikes can reach about four times the mean value before entering into surge. Note that at this operating point the pressure ratio is 6.7, as plotted in Fig. 6. These measurements highlight the damaging effect that the surge could lead regarding the instrumentation and the machine. When getting into surge, the pressure signals measured by sensors 1, 2 ,3 and 4 in the vaned diffuser (see Fig. 5) are plotted in Fig. 13 as a function of time (i.e. rotor revolutions). The vertical dashed line at revolution “13” is for what may be the onset of instabilities. The horizontal black lines represent the mean pressure measured with the sensor at position 1 (pos 1) just before surge. It is then easy to see the mean level of the pressure (white curves) increasing from pos-1 to pos-4, i.e. when going from the entrance towards the exit of the diffuser. After the first signs of instabilities, the mean pressure levels at these four locations (as well as at positions 5 to 9, not plotted here) reach the same maximum value Pmax, which correspond to the pressure downstream from the vane bow shock wave emanating from the diffuser blade leading edge. The upper branch of the shock has moved upstream pos-1 and leads to a massive separation on the suction side of the diffuser blade. The topology of the

40

J. Therm. Sci., Vol.19, No.1, 2010

shock and pressure wave system is completely changed and the mass flow passes only through a small part of the diffuser channel, in the region of the pressure side, close to pos-10 and pos-11, where the pressure is still increasing toward the exit of the diffuser. Then, after revolution “20”, the total mass flow decreases and the strength of the shock diminishes until revolution “100” where the average pressure level reaches the value at the exit of the impeller.

Fig. 12

pressure fluctuations without additional energy input is possible through the interference with other pressure waves, such as the α+ waves mentioned above and the reflections at the side-walls. In Fig. 14, the link between the massive pressure fluctuations and the unsteady impeller-diffuser interactions is highlighted. Here the pressure signals have been recorded by the B10 sensor during 6 different entrances into surge and superimposed on the same figure with the same time reference. The three plots have an abscissa which represents the time expressed as a number of rotor blade (main or splitter) pitches. The y-axis is the same for the 3 plots and is the non-dimensional pressure, as defined previously. In the x-axis-zoom-plot at the bottom of the figure, one can note the extremely good repetition of the pressure signals, entirely triggered by the rotor blade passages. These results show how suddenly (in less than 20 main blade passages) and with the same process the spike-type surge occurs.

First surge cycle measured with the B10 sensor

Fig. 13 Comparison of the pressure signal from sensors 1, 2, 3 and 4 when entering into surge

The onset of the instabilities happens in a very short time, less than 3 ms. No particular shape in the pressure signals could be identified as being responsible for this spike type onset, but the level of the fluctuations after rotation “20” which are very high in the inter-row region close to pos-1, and unchanged at pos-4, suggests that this unsteady interaction between the impeller and the diffuser is critical with regard to the triggering of the centrifugal compressor instability, as already observed by Spakovszky [6] and Justen et al. [10]. Between the impeller exit and the diffuser throat, the amplification of the

Fig. 14 Onset of instabilities – pressure signals of the B10 sensor, superimposed for 6 different entrances into surge

Conclusion Unsteady pressure measurements up to 145 kHz have been carried out in a high-speed centrifugal compressor, using an unsteady data acquisition system with conditioners developed in-house. The data acquisition system has been developed so that up to 48 signals can be simultaneously measured in synchronization with a sampling rate of 500 kHz, as it is currently being fully tested for the measurements within the 3.5-stage high-speed axial compressor CREATE. Particular care has been taken in order to reduce the cost of the conditioners and

N. Bulot et al. Unsteady Pressure Measurements in a High-Speed Centrifugal Compressor

lead to a price per channel which is below 100 euros. The measurements within the centrifugal compressor with 15 sensors have shown reliable results, not only when acquiring data at a stable operating point, but also during the onset of instabilities. The pressure flow field is currently being investigated, but the measurements presented in this paper have already permitted the validation of the unsteady shock and pressure waves system predicted by the CFD at peak efficiency and which degenerate when entering into surge. No evidence of the spike type precursors has been detected, but the excellent repeatability of the raw measurements demonstrates a behavior which is fully triggered by the impeller-diffuser interactions.

Acknowledgements The authors wish to thank Sebastien Goguey, Benoit Paoletti, Gilbert Halter and Roger Michelet for their fine work in this acquisition system development. The authors also wish to acknowledge the company Turbomeca which supports the centrifugal compressor research program.

References [1] Arnaud D., Ottavy X., et Vouillarmet A., 2004, “Experimental Investigation of the Rotor-Stator Interactions Within a High Speed, Multi-Stage, Axial Compressor: Part 1 - Experimental Facilities and Results,” ASME, Three Park Avenue New York, NY 10016-5990, USA. [2] Gourdain N., Ottavy X., et Vouillarmet A., 2009, “Experimental and numerical investigation of unsteady flows in a high speed three stages compressor,” European Turbomachinery Conference, Von Karman Institute for Fluid Dynamics Turbomachinery & Propulsion Department 72, Chaussée de Waterloo B-1640 Rhode Saint

41

Genèse – Belgium, pp. 247−266. [3] Trébinjac I., Kulisa P., Bulot N., et Rochuon N., 2009, “Effect of Unsteadiness on the Performance of a Transonic Centrifugal Compressor Stage,” J. Turbomach., 131(4), p. 041011. [4] Bulot N., et Trébinjac I., 2009, “Effect of the Unsteadiness on the Diffuser Flow in a Transonic Centrifugal Compressor Stage,” Inter. J. of Rotating Machinery, 2009, p. 11. [5] Ainsworth R. W., Miller R. J., Moss R. W., et Thorpe S. J., 2000, “Unsteady pressure measurement,” Meas. Sci. Technol., 11(7), pp. 1055−1076. [6] Spakovszky Z. S., 2004, “Backward Traveling Rotating Stall Waves in Centrifugal Compressors,” J. Turbomach., 126(1), p. 1. [7] Mä rz J., Hah C., et Neise W., 2002, “An Experimental and Numerical Investigation into the Mechanisms of Rotating Instability,” J. Turbomach., 124(3), p. 367. [8] Schleer M., Song S. J., et Abhari R. S., 2008, “Clearance Effects on the Onset of Instability in a Centrifugal Compressor,” J. Turbomach., 130(3), pp. 031002-11. [9] Biela C., Muller M. W., Schiffer H., et Zscherp C., 2008, “Unsteady Pressure Measurement in a Single Stage Axial Transonic Compressor Near the Stability Limit,” ASME Conf. Proc., Berlin, 2008(43161), pp. 157−165. [10] Justen F., Ziegler K., et Gallus H., 1999, “Experimental Investigation of Unsteady Flow Phenomena in a Centrifugal Compressor Vaned Diffuser of Variable Geometry,” J. Turbomach., 121(4), pp. 763−771. [11] Higashimori H., Hasagawa K., Sumida K., et Suita T., 2004, “Detailed Flow Study of Mach Number 1.6 High Transonic Flow With a Shock Wave in a Pressure Ratio 11 Centrifugal Compressor Impeller,” J.Turbomach., 126(4), pp. 473−481.