Close shock detection using time-frequency Prony modeling

In many cases, modeling a mechanical process may require a good understanding of signals issued from the system, as vibration accelerations. This is particularly the case when shocks are responsible of the vibrations. In the case of critical systems, each shock induces natural modes excitation with damped sines amplitudes. Identification of the shocks (occuring instants and induced vibrations) is a very important step of the analysis. However, when successive shocks are very close, their separation and their individual identification are not easy. In that paper, we adapt the well-known stationary Prony model to this non stationary context. We propose a method where shocks instants detection and parameters shocks estimation are separated. We illustrate the performances of the method on COR curves. In a last part, we apply it to a real acceleration signal recorded on a chairlift running over a compression tower in a rope transport plant where 48 shocks are expected, some of them being separated by only a few milliseconds.