Denoising and regularization in NAH for turbomachinery noise source reconstruction

The identification/localization of propulsion noise in turbo machinery plays an important role in its design and in noise mitigation techniques. Near field acoustic holography (NAH) is the process by which all aspects of the sound field can be reconstructed based on sound pressure measurements in the near field domain. Identification of noise sources, particularly in turbomachinery applications, efficiently and accurately is difficult due to complex noise generation mechanisms. This paper discusses the application of data smoothing techniques to better condition the measurement data for improved source reconstruction in NAH. A boundary element model which describes the non-linear transfer function between measurement data (acoustic pressure measurements acquired through microphones placed at close proximity to the test object) and acoustic noise sources within the turbofan duct is used for source reconstruction. The measurement data is smoothed prior to inversion, and two separate smoothing (or denoising) techniques - exploratory data analysis and wavelet denoising - are compared for accuracy. Acoustic noise sources are subsequently estimated from the smoothed data using truncated singular value decomposition (TSVD) based regularized inversion techniques. Preliminary results indicate that the use of smoothing prior to inversion, and regularization for inversion, provide higher accuracy in acoustic source identification.