Acceleration signal characteristics for loose part impact.

Abstract

Characteristics of acceleration frequency spectra for an impact signal were studied and loose part mass was estimated, using variation in the frequency spectral patterns. T h e frequency spectral pattern for the acceleration induced by impact depends on the loose part mass and not on the impact energy and the loose part shape. It was found that the frequency spectral pattern for an impact which occurred at the lower plenum vessel wall is not influenced by the impact point. On the other hand, the maximum acceleration amplitude depends not only on the loose part mass but also on the impact energy, the loose part shape and the distance from the impact point. It was also found that the foremost portion of the accelerometer output signal is influenced by the impact force between a loose part and the vessel. However, a second portion is influeuced by the sensor resonance and structural resonances. Loose part mass estimation, using "FR" values (ratio between low frequency component and high frequency component), is useful. The uncertainty range for the estimated loose part mass is about 0.7 decade.

Mass Estimation of Impacting Objects Against a Structure Using an Artificial Neural Network Without Consideration of Background Noise

“…It is well known that the mass of a loose part considerably changes the APSD curve of an impact signal [10]. For this reason, the APSD of an impact signal is considered an important factor for estimating a loose part mass.…”

“…As stated earlier, the APSD changes considerably with the impact mass. The main peak On a theoretical basis, the main peak is related to the dominant frequency (fmain) that occurs when the sphere collides against the plate and is determined by the contact duration (Tcontact) given by Hertz's theory as follows [10]:…”

“…Among the deterministic models for detecting and recognizing a specific signal pattern, the ANN has been widely used for cases in which the input-output relationship is strongly nonlinear [9,10]. In general, the ANN has a strong theoretical background and performs feature mapping, with high accuracy, from input parameters in an ambiguous form to the expected domains [13].…”