Using a spectral entropy criterion for improved onset picking in ultrasonic testing

Abstract

In ultrasonic testing, the time of flight (ToF) of a signal can be used to infer material and structural properties of a test item. In dispersive media, extracting the bulk wave velocity from a received signal is challenging as the waveform changes along its path of propagation. When using signal features such as the first peak or the envelope maximum, the calculated velocity changes with the propagation distance. This does not occur when picking the signal onset. Borrowing from seismology, researchers used the Akaike information criterion (AIC) picker to automatically obtain onset times. In addition to being dependent on arbitrarily set parameters, the AIC picker assumes no prior knowledge of the spectral properties of the signal. This is unnecessary in ultrasonic through-transmission testing, where the signal spectrum is known to differ significantly from noise. In this contribution, a novel parameter-free onset picker is proposed, that is based on a spectral entropy criterion (SEC) to model the signal using the AIC framework. Synthetic and experimental data are used to compare the performance of SEC and AIC pickers, showing an improved accuracy for densely sampled data.