ICLEAC Instability Control of Low Emission Aero-Engine Combustors

Transcription

1 ICLEAC Instability Control of Low Emission Aero-Engine Combustors CONTRACT N : G4RD-CT PROJECT N : GRD DELIVERABLE NUMBER : D2.5 TITLE : Periodic Results for a Generic Turbomeca LP Burner under Elevated Pressure Ewald Freitag, Johannes Eckstein, Thomas Sattelmayer Date of issue : 15/02/2004 Project funded by the European Community under the Competitive and Sustainable Growth Programme ( )

2 Table of Contents 1 Objectives Test Rig General Features Design High Pressure Siren Measurement Techniques Dynamic Pressure Transducers Photomultiplier High Speed Camera Data Acquisition Boundary Conditions for the Measurements Results and Discussion General Flame Behaviour OH*-Imaging High Speed Camera Acoustics in the Non-Reacting Case Acoustics in the Reacting Case Summary & Conclusions Data Formats Appendix...30 Page 2 of 42

3 Index of Figures Figure 1: Schematic sketch of test rig and high pressure siren as pulsation device... 5 Figure 2: Design of test rig and pulsation device... 6 Figure 3: Window module with mounted Turbomeca burner... 7 Figure 4: Cross sections through the window module. Left: Setup for Light sheet techniques, Right: Setup for PDA and other laser beam techniques Figure 5: High pressure siren aperture geometry: Double sinus projected onto a circle (stator) and ring sector (rotor)... 9 Figure 6: PCB dynamic pressure transducer J106B with water cooling adapter Figure 7: Example images for the two flame modes, observed at OP5. Left: anchored, right: lifted. Intensifier gain setting and contrast enhancement were the same for both images. The central axis of the burner is given Figure 8: Flame position represented by averaged images for several operating points. The flame position does not change with pressure, but strongly with air excess ratio?. OP9 represents a superposition of anchored and lifted flame Figure 9: Plot of the OH* intensity fluctuation of the high speed camera images over image number for different operating points. The sequence of 1024 images corresponds to seconds Figure 10: Comparison of photomultiplier voltage with intensity of HS-Camera images, OP Figure 11: Phase-averaged images of the OH*-intensity oscillation at OP7 (10 Hz siren excitation). The flame mode locks to the excitation Figure 12: Cumulated image intensity plotted over phase for the sequence illustrated in Figure 11. On the abscissa the image number is plotted, from 0 (1) to 330 (12) Figure 13: Phase-averaged images of the OH*-intensity oscillation at OP4 (360 Hz siren excitation). The flame mode does not lock to the excitation. Compared to Figure 11, the intensity scaling is stronger here, resulting in increased intensities for the lifted mode Figure 14: Cumulated image intensity plotted over phase for the sequence illustrated in Figure 13. On the abscissa the image number is plotted, from 0 (1) to 330 (12) Figure 15: Reconstructed flame section of the integral flame data of Figure 11 for OP7 after application of the BASEX-algorithm Figure 16: FFT pressure amplitude spectra for non-reacting conditions at OP Figure 17: FFT pressure amplitude spectra for non-reacting conditions at OP Figure 18: FFT pressure amplitude spectra for non-reacting conditions at OP Figure 19: Typical time plot of photomultiplier and PCB signals for an operating point without siren excitation (OP5). A transition of anchored to lifted flame occurs at about 13 ms, the reverse transition back to anchored at about 87 ms Figure 20: FFT amplitude spectra for reacting conditions at OP1, each averaged over 10 individual spectra.31 Figure 21: FFT amplitude spectra for reacting conditions at OP2, each averaged over 10 individual spectra.32 Figure 22: FFT amplitude spectra for reacting conditions at OP3, each averaged over 10 individual spectra.33 Figure 23: FFT amplitude spectra for reacting conditions at OP4, each averaged over 10 individual spectra.34 Figure 24: FFT amplitude spectra for reacting conditions at OP5, each averaged over 10 individual spectra.35 Figure 25: FFT amplitude spectra for reacting conditions at OP6, each averaged over 10 individual spectra.36 Figure 26: FFT amplitude spectra for reacting conditions at OP7, each averaged over 10 individual spectra.37 Figure 27: FFT amplitude spectra for reacting conditions at OP8, each averaged over 10 individual spectra.38 Figure 28: FFT amplitude spectra for reacting conditions at OP Figure 29: FFT amplitude spectra for reacting conditions at OP10, each averaged over 10 individual spectra.40 Figure 30: FFT amplitude spectra for reacting conditions at OP11, each averaged over 10 individual spectra.41 Figure 31: FFT amplitude spectra for reacting conditions at OP12, each averaged over 10 individual spectra.42 Page 3 of 42

4 1 Objectives Deliverable D2.5 presented in this report is part of Workpackage 2, Subtask1.2, which comprises experimental investigations of periodicities in non-reactive and reactive mixing processes. Subject to investigation is the Turbomeca LP module already examined in Deliverables D2.4, D3.6, D3.7 and D5.3. Main objective of this deliverable is to characterise the behaviour of the module under elevated pressure with the aim of determining the pressure dependency of phenomena identified at atmospheric conditions. Close coherence is thus necessary with the work performed in Workpackages 3 and 5 to complement the findings achieved there. The measurements are focussed on:? The investigation of low amplitude triggers that might start or help maintain the combustion instability process? The verification of correlations between combustor geometries and oscillation frequencies and amplitudes? The comprehension of the contribution of specific geometries to generation, amplification or damping of instabilities? The analysis of unsteady behaviours induced by the injection in the near burner region and global unsteady flame behaviour? The application of advanced flame investigation techniques developed in WP5, Task 3. 2 Test Rig Unlike all other measurements of TDM for ICLEAC, the measurements for Deliverable D2.5 have been performed on a separate test rig to be able to comply with the requirement for elevated pressures. This rig was newly designed and manufactured by TDM, D2.5 represents the first measurement campaign performed on this rig. 2.1 General Features The main purpose of this rig is to be able to perform acoustic and optical measurements on generic and industrial gas turbine burners under elevated and high pressure conditions. Since the focus of operation is similar to the low pressure rig designed earlier in the project, the general design features the same basic components (see Figure 1). The burner is supplied with pre-heated air via a variable length supply tube. A combustion chamber, which can also be varied in length, allows optical access through multiple Page 4 of 42

5 windows. The pressure load is achieved by water-cooled choking valves at the end of the combustor. To comply with the requirement for exhaust gas temperatures not exceeding 250 C, water is injected to cool the exhaust flow downstream of the choking valves. Good sensor access is guaranteed by a large number of universal sensor ports in supply tube, burner module and combustion chamber. To generate acoustic excitation, a high pressure siren was designed, operating with the same principle as the siren in the low pressure rig. This siren can be connected alternatively to the combustion chamber or to the supply tube and can be operated in two modes: 1. Air injection into the combustor or supply tube 2. Air blow-off from the combustor or supply tube (not shown in Figure 1) Figure 1: Schematic sketch of test rig and high pressure siren as pulsation device The main parameters of the rig are as follows:? Operating pressure at the burner of p3 = 40 bar. Nevertheless, laboratory infrastructure supply limitations restrict the operating pressure to 10 bar for the foreseeable future? Air flow of 300 g/s at 10 bar or 1 kg/s at 40 bar? Preheater Power of 128 kw? Maximum air temperature at the burner of 800 K Page 5 of 42

6 2.2 Design Figure 2 shows the design of test rig and the high pressure siren. On the upstream end the air preheater feeds the supply tube via an acoustically nearly choked assembly of perforated plates. The supply tube consists of two modules, one 400 mm, the other one 950 mm long, with the long module being optional. The inner diameter of supply as well as combustion chamber amounts to 150 mm. Three mounting ports for the siren are located on the lower side. Figure 2: Design of test rig and pulsation device On the downstream end, the supply tube connects to the heart of the rig: A window module incorporating both burner module and the first part of the combustion chamber (see Figure 3). This module is symmetric with respect to the vertical axis through the support point on the lower side seen in Figure 2, it can be mounted into the rig in both orientations depending on the optical access required (see also Figure 4). Independently of the orientation, the downstream part is operated as combustion chamber, while the upstream part serves as burner module.? One side of the window module features two large windows sized 150 x 90 mm^2 at the side and two slot windows 150 x 10 mm^2 at the top and bottom for application of light sheet techniques such as PIV, LIF,... This side was used as combustion chamber in the campaign of D2.5.? The other side has two large windows with the same size as in the light sheet side at an angle of 115 for the application of PDA and other measurement techniques where 90 access is unfavourable. This side served as burner module in the present campaign. Page 6 of 42

7 The fuel supply and the access for the mounting of the burner is achieved via a lid replacing the large window. Further lids are available to house several kinds of measurement access ports. The windows allow optical access to the primary zone of the combustion chamber. As seen in the left part of Figure 4, the large windows are oriented with a vertical offset to the burner axis to enable observation of the whole height of the combustion chamber. Since symmetry can be assumed for most flames, even one window suffices for observation. Figure 3: Window module with mounted Turbomeca burner Figure 4: Cross sections through the window module. Left: Setup for Light sheet techniques, Right: Setup for PDA and other laser beam techniques. Page 7 of 42

8 Connected to the window module respectively the first part of the combustion chamber follow two further combustor modules, one of them being optional. The total length of the combustion chamber can be varied this way to 840 or 590 mm. To further improve optical access, an axial window with 90 mm diameter has been implemented at the rear end of the combustion chamber, which allows for a global view of the whole flame instead of seeing only a part of it through the 150 x 90 mm^2 windows at the injector exit. The 11 custom-designed choking valves that enable the pressure build-up in the chamber are thus arranged around this rear window and the flow is directed radially outwards. Downstream of the valves, a ring duct collects the exhaust gases and discharges them into the chimney. Water injection into the flow to reduce the exhaust temperature is applied via the needle of the valves. Two more locations for siren mounting are available at the rear end of the combustor. Window module, combustion chamber and choking valves are water-cooled with the cooling water preheated up to 120 C to prevent condensation of water vapour resulting from the combustion reaction. The choking valves feature a separate non-preheated cooling circuit to get maximum cooling power at the critical cross section where Mach=1 is achieved and the heat transfer becomes maximum. Also, the double-walled ring duct collecting the exhaust gases and the frames for the quartz glass windows are watercooled. Thus, a considerable amount of energy is taken out of the exhaust gases on their way to the choking valves. This results in a temperature decrease that has to be considered in an acoustic modelling of the system. Two methods can be applied to set the correct value of combustor pressure for a given setting of burner air mass flow and fuel flow: 1. Adjustment of the choking valve cross section until the desired pressure is reached. 2. Setting of the valve cross section to a value larger than required and then using the siren air mass flow injected upstream of the valves to obtain the correct pressure. When applying siren excitation, this method has to be used. Tuning is required, since the siren air mass flow is responsible for both chamber pressure and excitation amplitude. One consequence of this approach is that there is cold air injected into the rear part of the combustor, mixing with exhaust gases and considerably reducing temperature. This is favourable, since the thermal load on the valves is lower, but at the same time the mixing temperature in the rear part is hard to model, leading to problems when applying low order modelling. Since the combustion chamber is long enough and the cold air injection takes place in the rear part, no influence of the cold flow on the primary combustion zone is expected. A total number of 37 universal sensor ports is available within supply tube, window module and combustor. These are used to accommodate sensors e.g. dynamic pressure transducers, pressure taps, temperature measurement, flame sensor as well as the ignition device and allow substantial flexibility in sensor application. If more bulky techniques have to be applied, the lids replacing the windows can be adapted. For the measurements described in this report, the rig was set up as follows:? Optional supply tube module mounted (overall length: 1680 mm)? Optional combustor module mounted overall length: 840 mm) Page 8 of 42

9 ? Siren mounted to the rear end of the combustion chamber at an axial distance of 795 mm from the injector exit, being operated in air injection mode. The air flow through the siren was even in cases without excitation used to trim the combustor pressure for a given burner air mass flow (method 2 described above). The siren air flow amounted to the same value as the air flow through the burner for all cases, with the according consequences for the mixing temperature. An exception are the nonreacting measurements described in Chapter5.3, where the siren was not connected to the rig.? Of the 11 available choking valves only 4 were mounted due to the rather low throughput of the Turbomeca injector, the rest was replaced with dummies. 2.3 High Pressure Siren One of the key components of the rig is the siren to impose harmonic oscillations by modulating the air flow. This was realised in accordance to the low pressure siren by incorporating the principle of a rotating disc, where the waveform is assumed to correlate with the open area of two shapes passing each other at a specified relative velocity. Since our concept consists of a stator plate and a rotor, this relative velocity corresponds to the angular velocity of the shaft on which the rotor is mounted. Concerning the shapes, a rectangle passing a double sine shape was chosen, which ideally yields an acoustic excitation without any unwanted higher harmonics. Since, in this approach, the rectangular and sinusoidal shape originally refer to a movement along straight axes, the geometry of the orifices was modified to the given finite geometrical radii of the stator and the rotor Figure 5: High pressure siren aperture geometry: Double sinus projected onto a circle (stator) and ring sector (rotor). Page 9 of 42

10 As a difference to the low pressure siren, only one large stator opening is implemented instead of 6 small ones in the LP design. This makes it easier to connect the siren to the rig. At the moment, the siren is rated to a maximum pressure of 12 bar. However, the casing can stand more, with the limiting feature being the shaft seal. An upgrade is possible if required. The maximum air mass flow through the siren amounts to 300 g/s and is only limited by the supply. 3 Measurement Techniques 3.1 Dynamic Pressure Transducers Piezo-type dynamic pressure transducers with integrated preamplifier PCB J-106B (see Figure 6) were used to measure the acoustic pressure at several locations. These acceleration-compensated transducers measure dynamic pressure within a range of 0.7 Pa to 0.57 bar on a static head of up to 138 bar. The frequency range stretches from 1 Hz to 5 khz (resonance at 60 khz). Integral electronics eliminate problems with high sensitivity to moisture and drift of high impedance charge mode sensors. Due to the sensitivity of the transducers to temperature, water-cooling was applied. The temperature of the water was held constant at 60 C by a thermostat. A sampling frequency of 10 khz per sensor was applied for all measurements. Sampling time was held constant at 1 second, resulting in individual data points per run. Three sensors were applied within this campaign, their mounting position being as follows: PCB1: Located in the window module upstream of the combustor at the entry into the prevaporising tube of the Turbomeca burner. PCB2: Located in the combustor, 5 mm downstream of the exit of the prevaporisation tube of the burner. PCB3: Located in the combustor at a distance of 277 mm downstream of the burner exit. 3.2 Photomultiplier A photomultiplier with a UG11 bandpass filter was used to capture the flame radiation in the UV-band. The UG11 filter corresponds to the wavelength of the radiation generated by OH*-radicals (308 nm) in the flame front. This OH*-chemiluminescence is related to both heat release and local temperature and can thus be judged as an indirect measure for the heat release of prevaporised flames. Nevertheless, further processing of the data is required to get the reaction rate information. The photomultiplier was mounted at the downstream end of the combustion chamber with optical access to the whole flame through the circular window. This setup ensures that the Page 10 of 42

11 total radiation generated by the flame is captured and not only the part visible through the side windows. Figure 6: PCB dynamic pressure transducer J106B with water cooling adapter 3.3 High Speed Camera Images of the flame OH*-chemiluminescence have been recorded using an intensified Kodak High Speed Motion Analyser EKTAPRO 4540mx. For detecting the OH*- chemiluminescence at a wavelength of 308nm, the same UG11 band-pass filter as applied on the photomultiplier was mounted to the camera. The sampling frequency applied was 4500Hz in all cases, and a sequence of 1024 camera images was saved for each operating point. The amplification of the intensifier unit was held constant over wide operating ranges, aiming at optimum comparability between the individual image sets but partially reducing image quality. Of the two 150 x 90 mm^2 windows available, the lower one was used to observe the flame, with the camera being focused on the central plane of the combustion chamber. Due to the window geometry (see Figure 4), an asymmetric image of the flame is captured, showing only the central and lower part. On the images, the flow direction is from left to right, the burner exit and thus the symmetry axis is located at about 75% of the image height. The grey-scaled images obtained have been further processed with MATLAB. Postprocessing consisted of cropping to the area of interest, contrast enhancement and conversion into false colour images. Page 11 of 42

12 3.4 Data Acquisition All measurement techniques described above have been applied simultaneously to ensure comparability. Each measurement was started by a hardware-generated trigger signal activating both high speed camera and data acquisition. In cases without siren excitation, this signal was generated at an arbitrary instant. For all other cases, the rectangular reference signal generated by an encoder on the shaft of the siren motor was used for triggering. A National Instruments data acquisition board was used to sample the data from dynamic pressure sensors and photomultiplier. Also, the reference signal has been sampled simultaneously. The total sampling frequency applied was 200 khz, the sampling rate per channel was 10 khz. A total of points have been acquired for each channel. 4 Boundary Conditions for the Measurements The measurement points of the burner investigated where derived in orientation to the directives of D5.1, where pressure scaling is achieved by the corresponding reduction of the Air Mass Flows and Fuel Mass Flows while keeping the temperature... m air, scaled? m air, real? p p AFR scaled? AFR real T scaled? T real This approach aims to maintain convective air velocities and AFRs with respect to the original operating point (Mach-equality). Thermal flame output scales down with the pressure. Within this measurement campaign, focus die lie on operating points at moderate preheating temperatures and pressures below 4 bar. The TDM operating points shown in Table 1 have been set to generic values in close accordance with the pressure scaled original conditions, similar to the approach already applied in the preceding deliverables. The operating points chosen feature a variation of pressure, air excess ratio and excitation frequency. For a potential later evaluation of the data the gain factors applied for the intensifier of the high speed camera and the amplifier of the photomultiplier shall be given (linearity of the devices can not be guaranteed): scaled OP 1-4: Gain HS-Camera: 5.71 Gain Photomultiplier: 35 OP 5-12: Gain HS-Camera: 4.40 Gain Photomultiplier: 30 real Page 12 of 42

13 OP Nr. P3 T3? P Burner? P Burner AMF? = 1/? Excitation Frequency [-] [bar] [K] [mbar] [%] [g/s] [-] [Hz] Table 1: Operating points for the Turbomeca injector Following the common gas turbine notation, P3 and T3 correspond to the pressure upstream of the injector. 5 Results and Discussion 5.1 General Flame Behaviour Unexpected from the behaviour of the Turbomeca burner investigated in the preceding Deliverables D2.4, D3.6, D3.7 and D5.3, a new flame mode was found occurring only at elevated pressure conditions. In the lifted mode well known up to now, the flame is located 4-15 cm downstream of the injector exit into the combustor, with the axial distance depending on the pressure drop over the burner and even more on the air excess ratio larger distance for leaner mixture, caused by the flame speed decline with temperature at rising air excess ratios. This mode can also be found here, with the same characteristics of variable flame position. However, within the range of air excess ratios investigated, the lifted flame does never come closer to the injector exit than about 4 cm. This mode is characterised by a comparably dim blue flame with a rather large radial extension. Newly encountered under elevated pressure conditions at and above 2.5 bar absolute pressure at the burner was a second flame mode, where the flame sits directly at the burner exit. This flame in the following referred to as anchored mode is much brighter and more compact and shows a more yellowish colour indicating a higher degree of unmixedness and/or droplet burning. The two flame modes are illustrated in Figure 7 by UG11 filtered images recorded with the High Speed Camera. Page 13 of 42

14 Figure 7: Example images for the two flame modes, observed at OP5. Left: anchored, right: lifted. Intensifier gain setting and contrast enhancement were the same for both images. The central axis of the burner is given. OP1 to OP4 at 1.5 bar pressure show no anchored flame within the investigated region of air excess ratio? >= 1.3. At 2.5 bar and above, the flame starts to shift to the anchored mode at?? 1.6. However, this mode is not stable in the long term and the flame falls back to lifted, with the relative time in anchored mode rising with equivalence ratio. This can be observed in the movie of OP5 and OP9. The anchored flame mode itself is not considered unstable, nevertheless, the rapid transition can be interpreted as a promoter for instabilities. This is demonstrated in Chapter 5.2 with the finding that the shift in flame mode can also be triggered by pressure oscillations applied with the siren. The following mechanisms can be thought of being responsible for the shift lifted to anchored and back: 1. Very low frequency (1 to 5 Hz) pressure oscillations with amplitudes up to 25% of the pressure drop were detected. These could trigger a mode shift if the flame type depends on the pressure drop over the injector and thus on the burner air velocity. However, there is no reason why this should only happen at higher pressures. 2. Heat release intensity rises linearly with pressure, while heat transfer to the walls rises only with Re^n, n<1, Re? p. This makes an ignition by hot gases and free radicals via the weak corner recirculation zones more probable. 3. Studies at TDM on flashback in stationary gas turbine burners with methane/hydrogen fuel recently showed a new flashback phenomenon called Combustion Induced Vortex Breakdown (CVIB). This mechanism produces a flashback along the axis into the premixing tube, depending on the radial velocity profile of the swirled flow. The shift from lifted to anchored flame encountered here could also originate from a breakdown of the swirled flow exiting the injector. It does indeed show strong similarities with the typical images obtained for CIVB. 4. Although the laminar flame speed scales down with pressure, some researchers have reported a small increase in turbulent flame speed with pressure due to the increased turbulent Reynolds number, becoming manifest in a shifting flame gravity centre (for methane fuel). Nevertheless, this increase has not been observed here (see Chapter 5.2), but at least a stable flame position was found. Page 14 of 42

15 5.2 OH*-Imaging High Speed Camera High Speed Camera recordings have been performed for various operating points with and without siren excitation. An investigation of the flame position has been performed for OPs without siren excitation, resulting in Figure 8 1. Averaged images of the OH*-chemiluminescence of the flame have been plotted for varying pressure and air excess ratio. The findings are as follows: 1. A strong dependence of the flame position on the air excess ratio was found. For a small?, the flame sits near the injector exit (OP9,?=1.3), while for a large? (OP12,?=2.05) it moves to a far downstream position almost out of the visible area. This deplacement is in accordance with earlier experiments and the literature, it can be attributed to a rise in laminar flame speed with temperature when varying the air excess ratio towards stoichiometry. 2. No pressure influence on the flame position was detected with air excess ratio held constant, as can be seen in the centre column of images recorded for?=1.7. The flame in the lowest image (OP10) is shifted slightly to the right, this is not because of a pressure influence, but because data were recorded at?=1.76 instead of 1.7 for this point. 3. The image of OP9 is an example for the ambiguity of the flame, it shows a superposition of anchored and lifted flame. From the individual images can be seen that the flame is in the anchored mode for about 30% and in the lifted mode for 70% of the recording time. The superposition has been resolved into averaged images for each flame mode in the lower part of Figure 8. Although observed for a shorter time, the anchored flame dominates the superposition due to its increased brightness. Since the global conditions of pressure, air excess ratio and preheating temperature do not change, this is a hint for reaction taking place at higher temperatures. As already mentioned in the preceding section, the increased brightness is attributed to a higher degree of unmixedness at a location closer to the atomiser. 4. The flame position change with air excess ratio is of course limited to the lifted mode, the anchored flame is fixed to the injector exit. Since the lifted flame does not continually grow into anchored, but shows a rather discontinuous jump, this behaviour can not be sufficiently explained by an increased flame turbulent velocity, which would allow for a continuous mode change. A good impression of the behaviour of the flame concerning heat release at certain operating conditions can be achieved when investigating the average flame luminosity over each image. Figure 9 shows this average flame intensity when being plotted over images or time. Three operating points at? = 1.7 are plotted, comprising a pressure variation from 1.5 to 3.5 bar (OP2 to OP10). Within this variation, the scaling factor has been held constant. Also, the amplification of the image intensifier has been accounted for, making the individual plots comparable. The unit of the intensity is arbitrary. 1 Despite OP12 shown in Figure 8 was recorded with siren excitation, no significant flame movement is detected for this point, so that averaging over the total image number and evaluation together with non-excited data is tolerable. Page 15 of 42

16 Figure 8: Flame position represented by averaged images for several operating points. The flame position does not change with pressure, but strongly with air excess ratio?. OP9 represents a superposition of anchored and lifted flame. For low pressures at OP2, a very strong fluctuation of the OH* signal is encountered that can also be seen on the individual flame images. These intensity changes are reduced drastically when moving to higher pressures. At 2.5 bar (OP6), the fluctuation is strongly reduced, while the mean intensity level has risen. This trend is continued at 3.5 bar. Numeric values for mean intensity and standard deviation of the fluctuation are given in Table 2: The standard deviation compared to the averaged value decreases from 110% (OP2) to 19% (OP9). An explanation for this strong fluctuation can be found in the instantaneous spray images obtained for the respective injector under atmospheric pressure conditions in Deliverable D2.4. In the atmospheric campaign, a strong fluctuation of the droplet number was observed, even without siren excitation. It was already mentioned in D2.4 that the elevated OH*-emissions in the range of Hz as found in Deliverable D5.3 are associated Page 16 of 42

17 with this phenomenon. The same result is seen here: The major fluctuation peaks in Figure 9 do also appear with a frequency of between 50 and 150 Hz. The source for this behaviour is probably the injector itself. A relatively large fuel volume lies between injector nozzle and fuel flow restrictor with low convection velocities within. We believe that the low dynamic pressure of the flow at low power settings makes the large fuel volume susceptible to perturbations, influencing temporal fuel flow and atomisation. The fuel flow increases with pressure due to the higher thermal load and so does the dynamic pressure with the square of the fuel flow ratio, which leads to a more stable fuel supply of the flame. Figure 9: Plot of the OH* intensity fluctuation of the high speed camera images over image number for different operating points. The sequence of 1024 images corresponds to seconds. OP2 OP7 OP12 mean St.dev Dev/mean Table 2: Mean value and standard deviation for the plots of Figure 9 Figure 9 shows a lifted flame. Nevertheless, a similar behaviour is also observed for the anchored flame mode, with the difference that the fluctuations are generally higher in Page 17 of 42

18 anchored mode. This is explained by the reduced distance to the atomiser and thus worsened mixing and pre-vaporisation conditions. Parallel to the high speed camera, a photomultiplier has been mounted to observe the global OH*-radiation. Since the camera does not see the whole flame but only a part of it, a comparison of the two signals is presented in Figure 10 to find out how well the signals correlate. As seen in the figure, the correlation is excellent. The high speed camera signal has been shifted by a constant of 0.4 intensity points to make up for an image intensity offset. Thus, the images obtained with the high speed camera do indeed represent the global OH*-radiation of the flame. Figure 11 shows a sequence of phase-averaged images for an operating point incorporating siren excitation at 2.5 bar pressure, where the anchored mode has been observed (OP7). For this operating point, the air excess ratio amounts to 1.5, meaning that the flame does show an occasional shift to anchored and back even without siren excitation. As seen in the images, the shift from lifted to anchored flame and back locks to the low-frequency excitation. The shift lifted - anchored takes place at the instant where the pressure within the combustor (PCB2) reaches its maximum, the backwards shift anchored lifted happens shortly before the minimum pressure is reached. The pressure amplitude for the investigated operating point amounts to 35 mbar (from the FFT calculation), which is equivalent to 32% of the pressure drop over the burner. This lock to the excitation is primarily associated with an equivalence ratio fluctuation. The air velocity through the burner becomes maximum when the pressure minimum is achieved in the combustor and vice versa. Since a constant fuel mass flow independent of short-scaled fluctuations can be assumed at the low excitation frequency of 10 Hz, the combination of air and fuel flow results in a rising equivalence ratio for rising pressure in the combustor the flame shifts to anchored. Minimum pressure on the other hand causes a rise in air excess ratio and thus a blow-off of the flame into the lifted mode. The drastic increase in OH*-radiation when shifting to anchored is documented in Figure 12, where the cumulated intensity of all pixels of each phase-averaged image is plotted over the phase angle. A factor 4 in overall intensity is encountered between the brightest and the dimmest image. Similar plots are achieved for the other operating points featuring low-frequency excitation. The intensity varies strongly with the phase angle, even if no flame mode shift occurs, and so does the flame position. A different picture comes up when raising the excitation frequency to 360 Hz, for example regarding OP8. Although there are several shifts seen in the image sequence recorded, a phase-locking of the flame mode to the excitation does not occur. This happens despite the fact that the excitation amplitude of 38 mbar is even larger than in OP7. Obviously, the mechanism is too slow to follow the excitation. Generally, the flame response in terms of position and intensity change is very low at this high frequency. This is illustrated in Figure 13 and Figure 14, showing the same information for OP4 corresponding to that of OP7 in Figure 11 and Figure 12. The amplitude of the intensity fluctuation amounts to only 3.8% at a pressure amplitude reaching 30% of the relative pressure drop over the burner, which is very small compared Page 18 of 42

19 to the factor 4 of OP7. Also, the flame position shows almost no change, as can be seen in Figure 13. The low flame answer observed for the 360 Hz excitation is reasonable, since the velocity fluctuation through the burner is low for this high frequency, as it has been shown in the preceding deliverables. Thus, several effects take place to limit the answer of the flame:? The equivalence ratio fluctuation is small from the beginning due to the low air velocity fluctuation? The small fluctuations generated at the injector nozzle are additionally diluted within the convection time necessary to reach the flame? The convection time is large compared to the oscillation period, inducing further dilution. The basic mechanism is thus an over-critical behaviour, based on the high damping of the velocity-fluctuation of the burner for high frequencies. Figure 10: Comparison of photomultiplier voltage with intensity of HS-Camera images, OP6 Page 19 of 42

20 Figure 11: Phase-averaged images of the OH*-intensity oscillation at OP7 (10 Hz siren excitation). The flame mode locks to the excitation. Figure 12: Cumulated image intensity plotted over phase for the sequence illustrated in Figure 11. On the abscissa the image number is plotted, from 0 (1) to 330 (12). Page 20 of 42

21 Figure 13: Phase-averaged images of the OH*-intensity oscillation at OP4 (360 Hz siren excitation). The flame mode does not lock to the excitation. Compared to Figure 11, the intensity scaling is stronger here, resulting in increased intensities for the lifted mode. Figure 14: Cumulated image intensity plotted over phase for the sequence illustrated in Figure 13. On the abscissa the image number is plotted, from 0 (1) to 330 (12). Page 21 of 42

22 Figure 15: Reconstructed flame section of the integral flame data of Figure 11 for OP7 after application of the BASEX-algorithm Figure 15 shows the corresponding flame from the integral projection of Figure 11 after the BASEX image reconstruction algorithm developed in Workpackage 5 has been applied. This algorithm deconvolutes an integral flame image into a central section through this flame, assuming rotational symmetry. Due to the fact that the entire flame image was not available, deconvolution has been based on the lower half-image after localisation of the burner s central axis by digital image processing. The bright lines correspond to the symmetry axis and its immediate vicinity, where reconstruction is mathematically not possible. This algorithm is especially useful to analyse the subtle structure changes of the anchored flame that can be seen more clearly in a cross section than in the integral image. A widespread intensity distribution and deviations from rotational symmetry occurring make the lifted flame regime less suited for a treatment with the BASEX algorithm. Page 22 of 42

23 5.3 Acoustics in the Non-Reacting Case Measurements with dynamic pressure transducers have been performed for the nonreacting case without fuel injection to determine the acoustic behaviour of the burner/combustor system without flame. This information is valuable for a basic validation of low order models in the non-reacting case. The temperature at the burner entry was kept constant while the pressure was varied to match the conditions of the combusting case. A differentiation by air excess ratio does not make sense here and siren excitation was not applied, meaning that 4 nominal operating points reduce to one here. The data are FFT-transformed, and the resulting amplitude spectra are plotted in Figure 16 (OP1) to Figure 18 (OP9) on a frequency scale up to 5 khz. Scales are identical for all plots, and the amplitude is given in mbar = 100 Pa. The amplitude spectra have been averaged over 10 individual FFTs. From these plots, the following information can be deduced: 1. Amplitudes upstream of the burner in the supply tube are typically lower than the downstream amplitudes for the low frequency regime up to 2 khz. Also, the noise floor is lower than in the combustor. The noise level in the combustor declines with rising frequency. 2. Generally, the injector with its rather complicated setup and swirled flow is considered to impose strong damping, especially for higher frequencies. Of course, the drastic area change supply tube injector and injector supply tube does also restrict the propagation of acoustic waves. 3. The amplitudes recorded are generally low, amounting to 20 Pa maximum. An exception are very low frequency oscillations (< 5 Hz, typically 1-3 Hz) with an amplitude maximum of 2 mbar that are recognised for all OPs. 4. Some activity is recognised in the supply tube at high frequencies from 3 to 4.5 khz and low pressure, this is accounted to vortex shedding at characteristic features of the injector. Also possible but much less probable is that these sounds originate from the flow restrictor between supply tube and injector. These high frequency oscillations grow in amplitude with pressure. 5. Two distinct peaks exist in the combustor at low pressure, one large at about 1500 Hz and one small at about 750 Hz. The large peak can also be observed in the supply tube with substantial attenuation. When moving to higher pressures, the 1500 Hz peak goes down and splits up in two peaks at about 1550 Hz and about 1700 Hz, the 750 Hz peak moves to 800 Hz and grows. The shift in frequency can be explained by slightly rising air temperature with higher pressure (see Table 1). 6. Generally, more peaks rise out of the noise floor with rising pressure. Page 23 of 42

24 Figure 16: FFT pressure amplitude spectra for non-reacting conditions at OP1 Figure 17: FFT pressure amplitude spectra for non-reacting conditions at OP5 Page 24 of 42

25 Figure 18: FFT pressure amplitude spectra for non-reacting conditions at OP9 5.4 Acoustics in the Reacting Case The aim of this section is to provide information on the dynamic signals recorded by photomultiplier and dynamic pressure sensors in the reacting case. This information is used to complement and fortify the statements derived before. Also, these data can be utilised to validate results from low order modelling. Figure 19 shows typical time-traces of the photomultiplier and dynamic pressure sensor signals for an operating point without siren excitation (OP5). Two flame mode shifts occur within this trace: one at about 13 ms from anchored to lifted and a shift back to anchored at about 87 ms. Similar to the results with low frequency siren excitation given in Chapter 5.2, the flame mode transition lifted anchored takes place at high values of instantaneous combustor pressure, while the reverse shift anchored - lifted is associated with low pressure. However, for non-excited conditions, this can not be reproducibly related to fixed pressure values, meaning that there must be another effect influencing the phenomenon. A dominant oscillation of about 380 Hz can be seen in all pressure sensor signals, however considerably attenuated for PCB1 in the supply tube. In the flame response represented by the photomultiplier voltage, this signal can not be found at the first glance. One general behaviour indicated by this plot and also found for other examples is that the amplitude of the pressure oscillation is lower in the lifted flame mode. Page 25 of 42

26 Since the flame does not noticeably lock to this oscillation, its nature does with high probability not relate to thermoacoustic self-excitation, but rather to an acoustic amplification of the flame noise. The higher amplitude in anchored mode is explained by the more compact flame with a higher heat release density, the same applies to the rise of amplitude with pressure (see FFT plots in the appendix). The wavelength of the 380 Hz oscillation is associated with a?/2 mode of the combustor. Also, low frequency pressure oscillations occur in the combustor, as already mentioned before. For each operating point, an FFT transformation has been performed with the data of photomultiplier and dynamic pressure sensor, the results can be found as amplitude spectrum plots within Figure 20 to Figure 31 in the appendix. Some general features of these plots shall be discussed in detail: 1. FFT plots were generally averaged over 10 individual plots. An exception is OP9, where only one plot was recorded due to a problem with data acquisition. Amplitude scales for the dynamic pressure sensors are the same within one plot for better comparability. The frequency scale was adapted to the phenomena occurring. Generally, activity above the frequency domain illustrated is very low to negligible. The gain setting of the amplifier for the photomultiplier is not taken into account within these plots, limiting the absolute comparability between individual OPs. 2. The general behaviour of the heat release already measured in the scope of Deliverable D5.3 for atmospheric pressure has been reproduced here: A flat maximum of flame activity at about 100 Hz and amplitudes strongly decreasing above 150 Hz. This maximum is associated with the non-uniform atomisation and fuel supply of the burner, as explained in the individual spray images of Deliverable D2.4 and the context of Figure 9 above. A strong reduction in heat release fluctuation is encountered with rising pressure, as also seen in Figure 9. The decreasing standard deviation to mean intensity ratio can also be found in the FFT plots when comparing the amplitude of the flat maximum around 100 Hz to the amplitude at 0 Hz representing the average intensity. 3. Dynamic pressure noise occurs in the low frequency regime up to 200 Hz without distinct peaks. The 380 Hz mode described above can be seen as medium to large peak depending on the operating conditions. Also, more peaks come up especially at higher pressures, partially associated with higher harmonics of the 360 Hz oscillation. 4. Main source of pressure activity is the combustion chamber. Damping through the burner is high and the amplitudes observed upstream of the burner in the supply tube are typically attenuated by a factor of The amplitude of the dominant 380 Hz mode increases with pressure and equivalence ratio, which is in accordance with the explanation that the peak originates from amplified flame noise. Very little response of the flame can be seen in the photomultiplier signal. However, a very small peak almost vanishing in the background noise is identified. This fits to the findings of Chapter 5.2, where also very little reaction of the flame was recorded for 360 Hz excitation. Page 26 of 42

27 6. The excitation signal imposed by the siren shows a good quality with no major harmonics. The amplitude of the signal does generally amount to about 30% of the pressure drop over the burner, measured at the injector exit into the combustor. 7. At operating points with siren excitation at 10 Hz, the amplitudes of the pressure signal of PCB2 and PCB3 are identical, showing the long wavelength of the mode that can be considered a plug flow. 8. At 360 Hz excitation, PCB3 does generally show a smaller amplitude than PCB2. This is due to a standing wave in the combustor related to a (almost)?/2 mode with a pressure node at the burner exit and the rear end of the combustor and a pressure antinode in between. This does also apply for the 380 Hz resonance mode without siren excitation. 9. The operating point with siren excitation at 360 Hz does almost overlap with the resonance mode at about 380 Hz. This overlapping frequency was not chosen on purpose, but is rather a coincidence that was discovered after the measurement campaign was completed. 10. Flame response to siren excitation as recorded with the photomultiplier is stronger for lower frequencies. It seems that the flame is not able to follow higher frequencies, these are attenuated very effectively. Figure 19: Typical time plot of photomultiplier and PCB signals for an operating point without siren excitation (OP5). A transition of anchored to lifted flame occurs at about 13 ms, the reverse transition back to anchored at about 87 ms. Page 27 of 42

28 6 Summary & Conclusions To extend the envelope of achievements on the Turbomeca LP injector already investigated in the scope of Deliverables D2.4, D3.6, D3.7 and 5.3, measurements were performed under elevated pressure conditions to uncover effects not accessible at atmospheric conditions. The campaign resulting in this report has been conducted on a new combustion test rig featuring a new high pressure siren for acoustic excitation. High speed video, photomultiplier and dynamic pressure sensor data were recorded with and without siren excitation at pressures up to 3.5 bar, leading to the following findings: 1. As already known from previous campaigns, very strong fluctuations in flame activity are encountered in a broad frequency region from 50 to 200 Hz at low pressures. These statistical fluctuations relate to the unsteady atomisation behaviour of the injector. A remarkable reduction of the standard deviation with respect to the average is encountered when applying higher pressures: 110% at 1.5 bar declining to 19% at 3.5 bar, resulting in a much more uniform combustion. 2. The burner shows considerable damping, especially for higher frequencies, generally reducing the amplitudes of perturbations originating in the combustion chamber by a factor of 10 to Differing from the behaviour encountered in previous atmospheric campaigns, the burner shows a second anchored flame mode additional to the lifted mode already known at high equivalence ratios and pressures at and above 2.5 bar. This mode shows a brighter, more compact flame with higher heat release density and noise production, which is attached to the injector exit into the combustor. 4. At certain operating points featuring low frequency excitation (10 Hz), the shift in flame mode from lifted to anchored and back locks to the excitation. The reason is found in a variation of air excess ratio with the excitation. This behaviour is not encountered for higher frequencies, since the velocity fluctuations responsible for generating the equivalence ratio changes are generally low for high frequencies due to the attenuating behaviour of the burner. 5. Generally, the flame response measured in radiation intensity and flame location is much stronger at low frequencies. It seems like the flame is not able to follow high frequency perturbations. 6. The position of the flame under varying pressure and air excess ratio has been investigated. A dependence on pressure was not detected, but a strong variation with air excess ratio?. 7. A resonance mode in the combustion chamber peaking at about 380 Hz has been detected with amplitudes varying with pressure and air excess ratio, but rather small compared to the pressure drop over the injector. This mode is associated with the amplification of flame noise by the combustor acoustics rather than with a classic thermoacoustic instability due to the very little amount of flame activity involved. Page 28 of 42