Research And Comparison Of Noise Source Identification Algorithms Based On Acoustic Holography Method

In order to research the sound field reconstruction performance of acoustic holography method, near field acoustic holography(NAH), statistically optimized near field acoustic holography(SONAH), near field acoustic holography based on equivalent source method(ESM-based NAH) and wide band holography(WBH) were deeply studied. The sound field reconstruction performance of the four algorithms and influence factors on reconstructed results were analyzed through simulation experiments. Algorithm comparison and optimization were conducted and validation experiment was carried out, obtaining the accurate reconstructed results and excellent reconstruction performance.First of all, the sound field reconstruction formulas of NAH were deduced based on STSF and the reconstruction errors caused by finite aperture effects were illustrated. To reduce the errors, different window functions were added in both special and k-space domain. Then, the reconstructed results of point source and plate source were calculated based on NAH, the results indicate that the low frequency noise sources can be significantly identified, but the medium and high frequency noise source identification results are not satisfactory. Besides, NAH can reconstruct the sound field radiated by plate sources accurately.Next, to explore the influence of array geometry and array element failure on sound field reconstruction results, the sound field reconstruction results of grid array, circle array and approximate evenly distributed irregular array, based on SONAH theory, were simulated. Results showed that grid array and approximate evenly distributed irregular array could accurately identify the sound source position, while circle array could not be used in SONAH sound field reconstruction. At the same time, the sound field reconstruction performance of SONAH and NAH was compared. Analysis showed that the two algorithms both could identify the noise source while SONAN reflected the true sound field more accurately. Further, SONAH was used to identify noise sources of a diesel engine. The identification results indicated that the main noise sources were the oil pump shaft and the oil pump governor in 50~1 000 Hz frequency band.Then, the reconstruction formulas of the ESM-based NAH were deduced based on the equivalent source theory. In order to choose an optimal regularization parameter of Tikhonov for near ESM-based NAH, Bayesian regularization criterion method was introduced to solve the inverse problem. Comparison simulations were conducted with Generalized cross validation method(GCV), L-curve method and Hald regularization formula method at various sound source frequencies, hologram distances, and Signal-to-Noise Ratios(SNR). The results show that GCV, L-curve, and Hald regularization formula method cannot achieve the optimal regularization parameter when at large hologram distances, high noise levels, high frequencies and input SNR errors. While, Bayesian regularization criterion method is not affected by these restrictions and has an excellent reconstruction performance in wide frequency band. Further experiment validates the correctness and effectiveness of the application of Bayesian regularization criterion method in ESM-based NAH. On this basis, iteratively reweighted least square method(IRLS) was introduced to compare with ESM method. The results show that IRLS method is more accurate and effective than ESM method in terms of sound field reconstruction, providing a new reference method for noise source identification.Finally, the reconstruction formulas of wide band holography(WBH) was deduced based on the equivalent source and the steepest descent method. On this basis, an algorithm was designed to research the reconstruction performance of WBH, also, the reconstructed results of WBH was compared with that of ESM and IRLS method, and the influence of holographic distance and SNR on WBH reconstructed results was analyzed. The results indicated that WBH had a high accuracy of noise source identification and excellent reconstruction performance in wide frequency band, moreover, holographic distance and SNR were insensitive to the reconstructed results of WBH. ESM and IRLS were suitable for low noise source identification, while IRLS had similar reconstruction performance to the WBH in low frequency band. Further, in order to choose more effective eigenvalues and corresponding principal components to improve the calculation efficiency, the principal component of WBH was truncated to optimize the algorithm, and the results of optimized WBH was compared with that of primary WBH. The simulated and validation experiment both showed that the noise sources were identified accurately and reliably by the truncated WBH algorithm.