Noise prediction from the measured surface-vibration of a casing structure by using compressed sensing and finite element method

Abstract

In a complex structural system, such as a gearbox or a motor, with partially or completely unknown set of excitation forces, the exact prediction of the vibrations of a casing encapsulating the complex system, is not possible, especially at high frequencies. For the spatial recording of the complete vibration field on the casing, it would be necessary to scan the casing surface using several vibration sensors, especially at higher frequencies. Such a high-resolution rate of sensors or recording points on the casing surface is rarely achievable in industrial tests. Moreover, usage of several optical scanning devices to scan the vibration field from every possible direction is very expensive, cumbersome and mostly ineffective. Therefore, a novel technique is developed here, which enables the calculation of the participation factors of the casing eigenmodes using a reduced number of vibration sensors on the casing, with the aim to estimate the radiated noise levels generated by the vibrating casing. This technique is based on a Compressed Sensing approach which solves an under-determined system of equations. The calculated participation factors of the eigenmodes are then used as inputs into a Finite Element Method (FEM) to calculate the radiated noise levels. The noise predictions are then compared to the measurements to achieve enough confidence level in this method for its industrial usage.