Dynamic characterization of the A400M acoustics fuselage demonstrator

Dynamic characterization of the A400M acoustics fuselage demonstrator René WINTER 1 ; Jörn BIEDERMANN 2 ; Marc BÖSWALD 3 ; Martin WANDEL 4 1,2,3 Deutsches Zentrum für Luft- und Raumfahrt e.v., Germany 4 AIRBUS Operations GmbH, Germany ABSTRACT Turboprop engines generate vibrations in the mid frequency range which are transmitted as structure-born sound to an aircraft fuselage. There it is radiated into the cabin as noise. To manage the acoustic impact validated simulation models and analysis methods are necessary. The A400M Acoustic Fuselage Demonstrator, a full scale aircraft fuselage, was used to develop test methods providing the validation data needed. The mid frequency range provides great challenges for such tests: To expand the well-established method of experimental modal analysis (EMA) into higher frequencies a very large number of measurement points is necessary. To ensure high quality results testing procedures from ground vibration tests (GVT) were adapted to the mid frequency range. The resulting sensor grid is a mixture of an optimized sensor distribution for low frequency analysis and a high density grid in the propeller plane of the A400M. To measure the approximately 2800 sensor positions a roving sensor method was used. Using several sensors permanently fixed to the structure the overall consistency of the data was analyzed. The methods presented here were able to provide high quality data used in later works for both model updating and method validation Keywords: Modal analysis, vibro-acoustic testing INCE Classification: 73.6, 75.6 1. INTRODUCTION The A400M acoustics fuselage demonstrator (see figure 1) is setup at the Helmut Schmidt University of the Federal Armed Forces in Hamburg. It is used as a test structure for new acoustic and vibro-acoustic methods and developments. The demonstrator is an actual Airbus A400M fuselage based on a predecessor design slightly different to the final aircraft. The German Aerospace Center (DLR) performed two measurement campaigns spanning a total time of seven weeks to acquire a set of vibrational data. The data contains information for the low and mid-frequency range on thousands of measurement points using dozens of excitation configurations. The one goal was the generation of high quality frequency response data of the A400M fuselage structure. Another was the development as well as the validation of methods suitable to conduct similar tests in the future within shorter time frames or with higher data quality. The acquired data was used for several applications: Experimental modal analysis (EMA) was performed in the low frequency range using commercially available tools (LMS and the PolyMAX algorithm). With the results of this analysis a modal model of the structure was created. This model was used to update a finite element model of the A400M acoustics fuselage demonstrator. The latter was derived from the finite element models originally employed during the A400M development phase (1). The EMA analysis itself allowed for several observations concerning its limitations when approaching the mid-frequency range. Finally the measurement data was also utilized to perform energy correlation (as described in (2) and (3)) employing the previously mentioned updated FE-Model. 1 rene.winter@dlr.de 2 joern.biedermann@dlr.de 3 marc.boeswald@dlr.de 4 martin.wandel@airbus.com 3296

INTER-NOISE 2016 c AIRBUS S.A.S. 2014. All rights reserved Figure 1: The A400M acoustics fuselage demonstrator during the first measurement campaign without acoustic boundary elements. As mentioned above the data was acquired over the course of two measurement campaigns: A first campaign with a strong focus on experimental modal analysis and a second campaign focused on acoustic excitation. During the first campaign four electrodynamic shakers and as many excitation points were employed. The second campaign used a 132 channel loudspeaker array for acoustic excitation. The array created a sound pressure field resembling the one generated by the A400M engines. Both campaigns had the exact same sensor grid layout but different excitation methods. Due to a scheduling conflict a significant structural difference was introduced in between the two campaigns by the addition of acoustic boundaries to the fuselage cavity. To evaluate the impact of these boundaries a electrodynamic shaker and excitation point was added to the second campaign. This created comparable data sets for both campaigns. Presented in here are the methods developed and used to acquire and process the data and an analysis of the data s quality as well as its usability for experimental modal analysis (EMA) is performed. Additionally a comparison of the modal model acquired using the first and second campaign is presented to give an estimation of the impact of said modifications on the structure. Since the measurements were conducted within a very limited time frame an optimized sensor grid was developed. It combines algorithmically placed measurement points for the measurement of the lower frequency vibrations and a very dense grid of sensors around the propeller plane. The latter is the most significant part of the structure for mid-frequency analysis of the operational deflection shapes. The realization of said sensor grid with limited resources was done using a roving grid of sensors. Subsequently, the data was processed and merged to get a single set of frequency responses for each excitation configuration and all measurement points. The quality of the acquired data was checked both for the signal-to-noise ratio (SNR) and for data consistency. The latter being a crucial piece of information when merging data acquired over the course of weeks with slight changes to the structures weight loading due to the moving grid of sensors as well as environmental conditions. The final results presented here are an overview of the experimental modal analysis performed, the implications for its usability in the mid-frequency range of aircraft fuselage structures. Also a comparison of the modal data of the structure before and after the addition of the acoustic boundaries is presented. 3297

2. A400M ACOUSTICS FUSELAGE DEMONSTRATOR In vibrational testing the structure to be analyzed is the driving force for designing the test. The A400M acoustics fuselage demonstrator at the Helmut-Schmidt-University of the Federal Armed Forces (HSU) in Hamburg (HSU) is an actual Airbus A400M fuselage. It was build prior to design changes to the A400M line of aircrafts, making the fuselage accessible for testing purposes. It was suspended on two portals within a testing facility at the HSU. The suspension was fixed to the fuselage at three points creating a statically determined system. The A400M acoustics fuselage demonstrator and its setup at the HSU is described in detail in (4). Table 1 lists some of the key characteristics of the structure: Table 1: Key characteristics of the A400M acoustics fuselage demonstrator Length 30 m Diameter 6 m Weight 13 t No. of frames 54 No. of stringers in circumference 99 No. of stringers above cargo floor 72 No. of skin fields in circumference 99 No. of skin fields above cargo floor 71 As can be seen in figure 1 an aircraft fuselage is a ribbed cylindrical structure. Within this text the stiffeners in longitudinal direction are called stringers. The ones running perpendicular to the stringers are called frames. In this case the frames shape above the cargo floor identifiable in figure 1 is roughly circular.the parts of the fuselages aluminum skin in between the stiffeners are called skin fields. 3. EXPERIMENTAL SETUP The tests conducted at the A400M acoustics fuselage demonstrator were designed to provide the best quality of data possible while taking limitations imposed by available equipment and time constrains into account. To ensure a positive outcome equipment, measurement procedures and techniques originating from DLRs ground vibration testing (GVT) experience at passenger aircrafts like the AIRBUS A350XWB MSN1 (5) were adopted to the problem at hand. 3.1 Measurement Equipment The sensor grid used during the campaigns at the A400M acoustics demonstrator was subjected to several limitations: The overall number of sensor positions that could be measured simultaneously was determined by the number of channels available on the measurement system (a LMS SCADAS III system in a 3 frontend configuration with 288 channels in total) and by the number of sensors available (230 ICP acceleration sensors, manufacturer PCB, model 352C65). Disregarding other limitations a total of 230 sensor positions could be measured before moving the sensors to new parts of the fuselage. 3.2 Sensor placement and grid design Besides hardware and time constrains, the sensor grid design was driven by these test objectives: Determination of the upper limits of modal analysis on a full-scale fuselage structure. Realization of a high resolution and equally spaced sensor grid in a part of the fuselage which is essential for further development and application of a correlation criterion based on kinetic energy distribution. Collection of stringer and skin field vibrational data in the propeller plane to verify supposed local vibration effects of stringers and skin fields observed in earlier tests. 3298

An extensive pre-test suggested an installation time of 4 h for 80 sensors given a two person installation team. On site testing time was limited to 15 workdays with a 5 person team for each campaign. The tworing configuration of the sensor grid allowed four persons to work on installation simultaneously. To ensure reproducibility of the roving sensor positions during both campaigns threaded sensor adapters were manufactured from POM (polyoxymethylene) and glued to all measurement positions directly on the fuselage. The hardware limitation suggested a setup with moving sensors on a predefined grid. Considering the setup time and space limitations in the A400M fuselage no more than 2800 sensor positions could be measured during the campaigns. To accomplish all the goals laid out above within the given constraints the sensor grid was composed of three subsections: The high frequency analysis was focused on the area between the front of the fuselage and the wing box (i.e. very stiff frames in the area where the wing is attached to the fuselage) and above the cargo floor. In that area every skin field was installed with a sensor (71 sensors per ring) and every frame of the structure was measured by an equally spaced sensor ring (72 sensors per ring). Figure 2: Sensor grid configuration. In addition the stringers between the 6 frames closest to the propeller plane were also installed with sensors (72 sensors per ring). In addition, every ring configuration in the front part of the fuselage was installed with 7 sensors below the quasi-rigid cargo floor. Due to lack of access these sensors were installed from the outside of the fuselage. In total, this installation results in approximately 80 sensors per ring. The rest of the fuselage was installed with sensors according to the results of the effective independence sensor placement method (6) using a finite element model of the structure for the low frequency range provided by Airbus. Due to known problems of mode visualization with an optimized sensor grid (7) and the deficiencies of the FE-model originally designed for overall aircraft dynamics, the calculated grid for the mid section of the fuselage was overlaid with a grid of 8 sensors for every frame while 5 sensors were added to every third frame of the rear cone section. The resulting sensor grid is shown in Figure 2. The gray area was used for global modal analysis. The yellow area is an equally spaced grid for the further development of higher frequency correlation techniques. The red area is additionally fitted with sensors on the stringers of the structure. A subset of 50 sensors from the configurations above were attached to fixed positions on the fuselage. These sensors were used to check the signal amplitudes between different measurements of the roving grid and to apply corrections if necessary. All sensor positions originating from the effective independence sensor placement method were part of this subset. In theory the fixed sensor positions are sufficient for modal analysis up to 50 Hz, even though the results would be hard to interpret visually. 3299

3.3 Excitation Signals During the tests two kinds of vibration signals were used: Sinusoidal sweep and random excitation. The sweep excitation was chosen because it should provide the best results for a linear system (8). The random excitation was used due to its simpler application and a lower sensitivity to structural defects. These were present in the form of mounting parts for the systems that usually go into a finished A400M aircraft. Not all of these could be removed before the measurements. Sinusoidal and random signals were used in a single and a multi-point excitation configuration. To obtain a high signal-to-noise ratio (see 4.1) all signals used for the mid-frequency range were force limited (see (5)). This was done to ensure a homogeneous force input into the structure over the excited frequency band of 50 Hz to 500 Hz. The signals used for the low-frequency experimental modal analysis were not modified using force-notching. Instead several measurement runs at different force levels using only sinusoidal sweep excitation were performed. Combining all measurements conducted during both campaigns seven shaker positions were measured using two different excitation signals for the mid-frequency range and one signal for the low-frequency range. 3.4 Test procedure The sensor grid consisted of 30 rings of approximately 80 sensors in the front part of the fuselage with each ring containing only frame, skin field or stringer sensors and of 160 sensors spanning the mid and rear section. This grid was separated into 15 double ring configurations plus the so called global configuration. In order to reduce systematic mass loading effects in the roving accelerometer tests, the double ring configurations did not use two consecutive frames, but instead it consisted of the first and sixteenth ring, the second and seventeenth ring, etc. This pattern allowed for easy cable routing with one cable harness leaving the structure to the front and one to the back. Also it separated the two rings of a single configuration by about 3 m, allowing for simple parallel installation. Figure 3: Measurement process for a single sensor configuration. 3300

To reduce errors during sensor repositioning all measurement points were fitted with a custom made sensor adapter. This adapter was directly glued to the structure and enabled a threaded connection of the sensor. Furthermore, unique labeling of each sensor adapter was used denoting the unique sensor id and configuration name before any sensors were installed. This reduced the effort of moving the sensor grid to a simple repositioning of accelerometers along matching numbers on labels and sensors. The adapters were also used to attach dummy masses to the unmeasured measurement points during the test lessening the error introduced by the shifting mass of sensors and cables in between measurement. During adapter installation each adapter was fitted with a dummy mass approximating the mass of a sensor and the cable directly attached to the structure near the measurement point. To assure a high signal quality of the measured data a multi-point excitation pattern was used. Four exciters of two different types were available for the test: Two Prodera EX 520 C50 modified for application in the high frequency range and two RMS SW-122. The Prodera exciters were used at opposing sides of the propeller plane and made up the first exciter pair while the RMS exciters at the front of the structure and the wing box section made up the second exciter pair. All exciters were attached to the outside of the fuselage pointing in radial direction. Only the push-pull rods were removed during configuration changes, everything else was left unchanged to ensure consistent data and a faster reconfiguration process. Both crest optimized random and sinusoidal sweep signals were used for excitation (8). The random signal was used for fast on site checks of the signal quality needing only a single excitation run and no external post processing to generate FRFs. The sinusoidal sweep was used to generate the final data set providing the highest quality of FRFs. The process used to setup and measure a single configuration is detailed in Figure 3. Original planning assumed an expenditure of 4 hours per configuration measured with a five person team. After the team got used to the process the time needed was reduced to about three hours freeing time to measure 4 additional configurations of 80 sensors each after the planned measurements were done. 3.5 Signal processing The frequency response functions were calculated for each individual run and later merged into FRF sets. One set per exciter and signal type combination. The processing was done using conventional post processing methods (9). To obtain the FRFs the H1 estimator was used. The auto- and cross-power spectral densities used in the H1 estimator were calculated using Welch s method with a Hanning window and an overlap of 75%. In case of the sweep excitation, due to the correlated excitation signals the spectra from the symmetric and anti-symmetric runs had to be combined to a single input matrix before the H1 estimator was used. The data from the non-coherent crest-optimized random signals was directly processed thanks to the uncorrelated excitation signals. Due to the huge size of the resulting data and different requirements in the analysis methods it was necessary to do the post processing twice: Once with a frequency resolution of 0.195 Hz but limited to a range of 50-200 Hz. This data was used for modal analysis. And once with a resolution of 0.5 Hz in the full range of excitation (50-500 Hz) used for energy calculations. 4. RESULTS The acquired data was checked for both quality and consistency. To get a measure of the data quality the signal to noise ratio was evaluated. The consistency of the data was checked by calculating the deviation of the signals measured by the fixed sensors. 4.1 Data Quality and Data Consistency The signal-to-noise ratio (SNR) was calculated to evaluate the necessary excitation forces and number of shakers for future tests of similar structures. During the test procedure as outlined in 3.4 the noise floor was measured for each configuration by measuring all channels in an unexcited state. Specifically the SNR for dual shaker excitation and single shaker excitation and for multi-sine and sine-sweep excitation were compared. An overview of the results is shown in Table 2. The results shown are averaged both over a wide frequency range of 50 Hz to 400 Hz and over all 50 fixed position sensors. According to the test results the SNR increases by approximately 5 db for dual-point excitation compared 3301

Table 2: SNR comparison for different excitation setups and signals. The SNR given below is averaged between 50 Hz and 400 Hz and over all fixed sensor positions. Signal Type No. Exciters SNR [db] Sweep 1 25.8 Sweep 2 31.4 Random 1 20.4 Random 2 25.8 to single-point excitation and by approximately the same when using a sine-sweep instead of a random excitation signal. These results point toward a multi-point sine-sweep excitation as optimal, if SNR is the driving concern. On the other hand multi-point sine sweep excitation proved to be quite time consuming, requiring a measurement run per shaker. Random excitation requires just a single run for all shakers combined. Fewer shakers mean less equipment to be transported and setup. The most time efficient method of excitation is single-point random excitation. Combined with the results in Table 2 the trade-off to be made is one between signal quality and time. A critical aspect of the test procedure was the measurement of a roving sensor grid. These measurements were performed over the course of weeks. The test facility was not air conditioned leading to temperature fluctuations. A total of 160 sensors and associated cables was moved through the structure between measurement runs changing the mass load slightly. Additionally the push-pull rods of the electromagnetic exciter had to be disconnected before anyone could enter the structure to reconfigure for the next measurement run. To Figure 4: Averaged FRF of a single fixed sensor acquired over all 20 measurement runs of the first campaign and its 2 σ standard deviation. check the data consistency the signals of 50 sensors placed at fixed locations on the fuselage was evaluated by calculating the standard deviation over all 20 measurement runs conducted in a single campaign. One example of this shown in figure 4 and table 3. The consistency of the data decreases with rising frequency. Especially the amplitude deviation rises while the phase stays relatively constant. As mentioned in 3.5 all FRFs where calculated using a Hanning window. This window function favors phase over amplitude information and is thus the most widely used window function for modal analysis. A study of the effects of different window functions like a flattop favoring correct amplitudes on data consistency remains to be performed. 3302

INTER-NOISE 2016 Table 3: Amplitude and phase error with increasing frequency for a single fixed sensor Frequency Range 8-100 Hz 100-200 Hz 200-300 Hz 300-400 Hz Amplitude Deviation 11.7 % 16.9 % 44.0 % 43.0 % Phase Deviation 25 30 20 25 4.2 Experimental Modal Analysis To investigate the limits of experimental modal analysis (EMA) for aircraft fuselage structure when approaching the mid-frequency range EMA was performed using the data of the first measurement campaign. Some general observations were made during modal analysis of the FRFs obtained from this test. These observations concern the data quality and the possibility of picking modes. These results were obtained using LMS TestLab with the PolyMAX estimator for modal analysis. Figure 5: All 2850 FRFs calculated for a single excitation point using sweep excitation. Up to 50 Hz modal analysis using PolyMAX works flawlessly. The synthesized FRFs fit the measured data very well. At 70 Hz the resonance peaks begin to widen but the process of modal analysis still works well. At approximately 90 Hz the peaks of the measured FRFs begin to blur into each other but modes can be identified with reasonable accuracy. At 120 Hz it is still possible to pick some modes but the damping is strongly underestimated using the PolyMAX algorithm. At 140 Hz and up no reasonable modal analysis is possible due to the resonance peaks blurring into each other. Figure 5 shows all FRFs obtained from notched sweep excitation calculated from a single excitation point. The behavior described above can easily be seen. The decrease in peak separability and the general rise of response towards higher frequencies is easy to spot. Figure 6 shows some exemplary modes identified from the data shown in Figure 6. The figure shows the high density front part of the sensor grid only. The modes shown there were identified in the frequency range between 50 Hz and 100 Hz and are sorted according to their eigenfrequency, i.e. fa<fb<fc<fd. The damping ratio of these modes can be found between 1.1% and 1.5% critical damping. It is important to note that each of the displayed modes was calculated from an FRF data set that was collected in 20 separate measurement runs with different sensor configurations. The occasional seemingly dead sensors originate from parts of the fuselage where installation of a sensor was simply not possible due to irregularities of the structure. 3303

INTER-NOISE 2016 Figure 6: Some modes identified from the FRFs shown in figure 5. The modes are sorted according to their eigenfrequency from lowest (A) to highest (D). 4.3 Influence of the acoustic boundaries As mentioned earlier the two test campaigns were performed on slightly different configurations of the A400M acoustics fuselage demonstrator. For the second campaign the fuselage was closed using foam absorber elements. These elements are attached to a rigid frame affixed to the testing hall in which the fuselage demonstrator is located. To ensure minimal change of the dynamic characteristics of the fuselage itself the cavity within the fuselage is attached to the frame carrying the boundary elements by an flexible rubber connection only. The setup is shown in figure 7. (a) Front end acoustic boundary (b) Rear end acoustic boundary Figure 7: Attachment of the acoustic boundary elements added to the fuselage structure between the measurement campaigns. The rubber band connecting the edges of the fuselage cavity to the frame can be seen as a thin black line. An analysis of the lower frequency dynamic mode shapes between 8 Hz and 80 Hz resulting from shaker excitation of the two different campaigns shows that this setup changed the demonstrators dynamic characteristics more than expected. While a change of the rigid body modes of the structure located below 8 Hz was expected, the dynamic modes should have been mostly comparable. As can be seen by the MAC matrix (10) 3304

Figure 8: The influence of the acoustic boundaries added to the A400M acoustics fuselage demonstrator. According to the MAC criterion there is only a weak correlation between the modes identified before and after the modification. calculated for the modal model of the two different structural configurations in figure 8 there is next to no correlation of the mode shapes. The structural changes introduced by the absorber elements and their bearing are so significant that they resulted in fundamental changes of the dynamic characteristics of the A400M acoustics fuselage demonstrator. 5. SUMMARY The methods developed to test the A400M acoustics fuselage demonstrator in the low- and mid-frequency range have been proven to work well. The overall quality of the acquired data is excellent and extensive tests with different excitation signals and shaker configurations will allow future test campaigns on similar structures to balance the time and effort spent with the quality of data necessary for the problem at hand. Using a roving sensor grid method with dozens of sensors moving along the structure proved to be no problem for the low-frequency results. Both the analytical consistency check and the EMA analysis proved a high data consistency up to 150 Hz. With rising frequency the errors introduced by changing sensor positions and environmental changes over the course of the test campaigns impacted the results more significantly. This makes methods reliant on direct single sensor data unusable. Methods relying on averaged results, either spatially or in the frequency domain, are still usable though. The methods developed during the testing of the A400M acoustics fuselage demonstrator form the basis of further test procedures planned on a similar structure: The CRAFD (Cror Acoustics Fuselage Demonstrator) located in the ZAL TechCenter in Hamburg. In addition, the data was used for several subsequent studies: A finite element model based on the A400M was updated to reflect the properties of the A400M acoustics fuselage demonstrator as setup at the HSU, the data was used as a basis for planning the Acoustic FlightLAB platform (1) and to advance the development of a new correlation criterion for experimental and simulated data (11) in the mid-frequency range of structural vibrations. 3305

ACKNOWLEDGEMENTS The authors gratefully acknowledge the valuable support of Prof. Delf Sachau and Mr. Christian Köhne from the Helmut-Schmidt-University of the Federal Armed Forced in Hamburg for providing the A400M acoustics fuselage demonstrator and an ideal infrastructure for conducting vibro-acoustics experiments. The work presented here was conducted in the course of the COCLEA project (Comfortable Cabin for Low Emission Aircraft) funded by the German Federal Ministry for Economic Affairs and Energy (BMWi) under the LuFo-4 Framework program. REFERENCES 1. Wandel M, Scheel H. Design Requirements of Acoustic Flight LAB Platform. In: Proceedings of Inter-Noise 2016; 2016.. 2. Winter R, Norambuena M, Biedermann J, Böswald M. Experimental characterization of vibro-acoustic properties of an aircraft fuselage. In: Proceedings of ISMA International Conference on Noise and Vibration Engineering. Katholieke Universiteit Leuven; 2014.. 3. Biedermann J. Energiebasierte Korrelation von strukturdynamischen Messungen mit numerischen Modellen für Strukturen mit hoher modaler Dichte [Dissertation]. Technische Universität Braunschweig; 2016. 4. Sachau D, Köhne C, Renger K, Scheel H. Vibroacoustic test bed with Airbus A400M-fuselage. In: Proceedings of the International Conference on Noise and Vibration Engineering ISMA 2014. Leuven, Belgium; 2014. p. 55 62. 5. Govers Y, Böswald M, Lubrina P, Giclais S, Stephan C, Botargues N. AIRBUS A350XWB Ground Vibration Testing: Efficient techniques for customer oriented on-site modal identification. In: Proceedings of the International Conference on Noise and Vibration Engineering ISMA 2014. Leuven, Belgium; 2014. p. 2495 2507. 6. Kammer DC. Sensor Placement for On-Orbit Modal Identification and Correlation of Large Space Structures. In: American Control Conference, 1990; 1990. p. 2984 2990. 7. Böswald M, Will B, Schulze B. Theoretical and practical implementation of sensor optimization for ground vibration testing. In: Proceedings of the International Conference on Noise and Vibration Engineering ISMA 2012. Leuven, Belgium; 2012. p. 2055 2069. 8. Pintelon R, Schoukens J. System identification: a frequency domain approach. John Wiley & Sons; 2012. 9. Füllekrug U, Böswald M, Göge D, Govers Y. Measurement of FRFs and modal identification in case of correlated multi-point excitation. Shock and Vibration. 2008;15(3-4):435 445. 10. Allemang RJ, Brown DL. A Correlation Coefficent for Modal Vector Analysis. In: Proceedings IMAC I - 1st International Modal Analysis Conference; 1982. p. 110 l16. 11. Biedermann J, Winter R, Böswald M, Wandel M. Advanced correlation criteria for the mid frequency range. In: Proceedings of Inter-Noise 2016; 2016.. 3306