LEAKAGE DETECTION AND QUANTIFICATION TECHNIQUES USING VARIOUS METHODS OF NEARFIELD ACOUSTIC HOLOGRAPHY
MetadataShow full item record
This thesis proposes an acoustic technique to detect and relatively quantify leakages in buildings and enclosures using various methods of nearfield acoustic holography (NAH). This laboratory study was performed on a scaled, wooden building model. Known leakages can be created in the wooden model and the acoustic method was tested to localize and relatively quantify these known leakage areas. An acoustic source was placed inside the building model and a planar hologram measurement was performed near the surface of the building model. Various methods of NAH were applied on the hologram data to reconstruct the sound pressure field on the wall of the building model. The detection and quantification capabilities of four different NAH methods, namely, discrete Fourier transform (DFT) based NAH, equivalent source model (ESM) based NAH, boundary element method (BEM) based NAH and statistically optimized NAH (SONAH), were compared in this study. It was shown that the NAH methods were able to successfully locate and relatively quantify the area of the leakages using the reconstructions. Although all the four algorithms produced comparable results in the very nearfield, at larger hologram distances, ESM and SONAH reconstructions were more accurate than the reconstructions using the other methods. Although, ESM and SONAH produced similar results for most of the cases, ESM is more preferable due to its simplicity in implementation and less computational time requirements. Lower frequency reconstructions were found to be more accurate and advantageous in the context of leakage detection and quantification. When the hologram distance was increased more than a particular limit, all the four algorithms arrive at inaccurate reconstructions due to the very ill-conditioned propagation matrices. New filtering methods to alleviate these larger reconstruction errors were introduced and the results were demonstrated. Effects of large sensor phase mismatch were also studied. It was demonstrated that larger phase error in the measurements could result in less accurate reconstructions. Performances of various regularization parameter choice methods applied to different approaches of nearfield acoustic holography were compared at various distances of reconstructions. Generalized cross validation and Morozov methods were implemented to arrive at filtering parameters to regularize the NAH reconstructions. Morozov method did not provide any significant filtering for the geometries considered in this study. GCV method produced very accurate reconstructions when a very nearfield measurement was supplied. Four new parameter choice methods were introduced to obtain the appropriate regularization parameters for very ill-conditioned inverse problems such as NAH. These methods work very well even at larger hologram distances and when the matrix dimension is very large where other available methods fail. These new parameter choice methods are not specific for the NAH problem. They can be applied to any ill-conditioned inverse problem. The advantages of each parameter choice method were explored and discussed in detail. Effects of signal quality on the NAH reconstructions were also studied. Patch NAH was implemented successfully to extend the aperture of computational domain more than that of the measurement. Also, the challenges in obtaining a smooth solution through patch NAH were discussed. A unique, MEMS based microphone array was designed, fabricated and tested keeping the future field tests in mind. The tests show that this array produces reasonably accurate measurements that can be used for the NAH methods. GCV method was found to work well for the reconstructions from the array even at larger distances because of the smaller propagation matrix due to the less number of microphones. This portable array can be used for field tests due to its portable form factor and reasonably accurate reconstructions.