Noise source identification for mechanical systems generating periodic impacts

For mechanical systems generating periodic impacts, classical noise source identification approaches are faced with several difficulties. An alternative approach has been developed. First, the radiated sounds under impact excitation of the various system elements are measured. Second, their contribution to the overall system noise is obtained using a signal processing method to extract the various impact source signatures from the overall signal. The approach has been validated progressively in three steps: first, with a numerical simulation, second, with a first experimental set-up where impact sources can be easily added or removed, and, third, with a second experimental set-up closer to industrial reality. The prediction error was found to increase between the simulation and the experimental validations, from <1 dB to values between 1 and 3 dB, which are still acceptable for noise source identification in practice.