Modal analysis of in-duct fan broadband noise via an iterative Bayesian inverse approach

Abstract

An advanced duct microphone array analysis based on a user friendly iterative Bayesian Inverse Approach (iBIA) has been successfully applied to assess the modal content and the Sound Power Level of fan/outlet guide vanes (OGV) broadband noise. It is shown that iBIA provides considerable reduction of artefacts associated to array sidelobes leading to an improved dynamic range in the reconstructed mode distribution plots. The method also benefits from a data-dependent and fully automated regularisation procedure. Moreover, it allows to control the sparsity degree of the reconstructed modal content: thus different sparsity levels can be tailored for specific noise components, such as fan tones or broadband mode components. It also applies directly to the cross-spectral matrix of measurements, thus advantage can be taken of averaging the data over several snapshots whenever the assumption of stationarity and ergodicity holds. These features facilitate its application to industrially relevant configurations such as the UFFA fan test rig that was operated by AneCom AeroTest in order to deliver reference data for broadband noise modelling in the frame of the European Project TurboNoiseBB. The iBIA has been applied to perform a full azimuthal and radial mode detection as well as axial-wavenumber and azimuthal decompositions. Two configurations of fan/OGV-spacing along two working lines have been tested. The radiated noise has been measured by in-duct microphone arrays located at the intake and downstream of the fan module. Longer rotor–stator gaps are shown to reduce high frequency levels at approach conditions for both inlet and exhaust noise over a large frequency range. The benefit of long gaps tends to vanish for higher fan speeds and even leads to noise increase at high frequencies. Duct sound power estimates obtained from a full azimuthal and radial decomposition have been compared to estimates from both the wavenumber and the azimuthal decompositions, the latter two requiring further assumptions regarding the modal energy distribution. Based on this approach, the widely used equal energy per mode and equal energy density per mode assumptions have thus been evaluated on a large industrially relevant experimental data set.