A Method for Semi-automatic Mode Recognition in Acoustic Emission Signals

Abstract

Acoustic emission (AE) is a non-destructive technique that relies on monitoring naturally occurring sources of high frequency ultrasound in components and structures. Ultrasonic waves propagate in the form of different wave modes—for instance Lamb waves in thin plates, or Rayleigh and P- and S- waves in bulk structures. Those wave modes have different properties, but also contain information regarding the source of the naturally occurring wave. Manually, the wave modes can be recognized by comparing a time–frequency representation of the signal to the dispersion curves expected in the tested object. For analyzing a large number of signals, this manual mode recognition becomes a tedious process. This paper proposes a method to automate the wave mode recognition based on some minimal knowledge of the occurring wave modes. As inputs, only the propagation speed of the possible wave modes and the source position need to be provided along with a limited set of reference wavelets for each wave mode. Cross-correlation of a signal with a reference wavelet of a mode reduces the signal to a limited number of peaks that may delineate the start of the mode. Using other signals from the same event but from different sensors, velocities are calculated for each peak in order to select the peak that corresponds to the arrival of the mode under investigation. To validate the method, a dataset was recorded based on four types of out-of-plane sources: Hsu-Nielsen sources of 0.3 and 0.5 mm, sensor pulse signals and AEs from melting ice. Since the presented dataset was recorded on a plate, the aim of the validation was to recognize the zero-order symmetrical and anti-symmetrical Lamb modes. The results of the proposed mode recognition method applied to this dataset are compared with results from manual mode recognition. For Hsu-Nielsen sources, the succes rate is found to be above 95%. For narrow-band pulsed signals or for AEs from melting ice with a low signal-to-noise ratio, succes rates between 75 and 80% relative to manual mode recognition are reported.