Tests in aeroacousitcs wind tunel was conducted in the main subsonic wind tunnel at Ecole Centrale de Lyon, LMFA. Different test configurations have been realized, for varying flow speeds from 10 to 50m/s, and for different acoustic source configurations. The consortium got the necessary database for academic validation of developed methods. The measurements were done with (a) a floor-mounted array of MEMS microphones, (b) a rotating linear array in the floor. For comparison, a case with 30 m/s flow speed and three speakers mounted in the ceiling were considered. Two of these speakers were driven coherently.
An expert assessment of the stochastic denoising tool, as developed within ADAPT has been made. The purpose of the assessment was to apply state-of-the-art denoising techniques to the measurements and to compare them with the results of the newly developed denoising techniques, especially those that are based on stochastic modelling, like Probabilistic Factor Analysis (PFA) and bayesian focusing. A comparison was reported between several existing denoising techniques. It was found that CLEAN-SC beamforming and Source Power Integration (SPI) were the most successful “traditional” methods for obtaining denoised microphone data. Then the state-of-the-art beamforming methods ISPI and CLEAN-SC were applied to wind tunnel measurements made available in the ADAPT project and compared with newly developed denoising methods.
The following conclusions has been drawn.
• CLEAN-SC works better than ISPI.
• The quality of the CLEAN-SC denoised data is comparable to the denoised MEMS array results shown in ADAPT Deliverable D1.2.
• The denoised MEMS array spectrum obtained with PFA, is better than the CLEAN-SC reconstruction (Figure 27).
• The denoised linear array spectrum obtained by integration over axial wave numbers, is much better than the CLEAN-SC reconstruction.
• The Bayesian approach provides a very promising denoising tool, especially if re-propagation is applied. This, however, requires knowledge about the sources. For blind denoising (no knowledge about the sources), DRCM or PFA may perform better. The MCMC-method is a very promising tool for assessing the reliability of the results.
Furthermore, a method has been proposed to automatically extract tones from aeroacoustic signals. Compared to the state-of-the-art, the proposed solution if flexible enough to extract very large numbers of tones, even when distributed at incommensurate frequencies, and in the presence of moderate random modulations. The proposed methodology has proved very efficient when processing several datasets, including the CROR data which are known for being challenging. As a byproduct, the extraction of tones also returns the broadband contribution of the signals. Then a companion method has been proposed to decompose broadband noise into a cyclostationary part, typically attached to the operation of a rotor, and a stationary part related to background noise.
An important goal of this work was to illustrate the successive steps involved in the separation process as well as providing guidelines as how to set up parameters in the algorithms. Criteria have been devised for allowing automatic selection of critical parameters, so to relieve as much as possible the user intervention. The methodology has been illustrated on aeroacoustic signals recorded in front of a CROR engine in a wind tunnel. The broadband signals obtained after removing the tonal components have been further decomposed into cyclostationary contributions due to rotation of the rear and front propellers and the stationary background noise due to inflow and turbulent boudary layer. It was demonstrated that the cyclostationary approach developed are capable of identifying cyclostationary CROR signals, both tonal and broadband. At present there is no comparable method in the field.
Finally, the denoising techniques has been applied to microphones signals measured on the fuselage of an inflight aircraft. Results are conclusive and promising.
