A Novel Feature Representation Method Based on Similarity Between Statistical Distributions of Acoustic Emission Waveforms

Acoustic emission (AE) has been increasingly favored for nondestructive testing and process monitoring due to its exceptional sensitivity to malfunctions. Nonetheless, excessive background noise can often mask crucial information, making it challenging to detect defects. Hence, it is crucial to resolve the problem of identification of the weak emission source and to obtain an accurate representation of the initial waveform. The present work proposes a novel feature representation method using similarity between statistical distributions of original AE waveforms. In contrast to existing AE signal processing methods, in this work, the waveform is considered and treated innovatively as a connection of many harmonics from a microview. Therefore, only the amplitude and half-period of the harmonic are extracted as a feature for representing the harmonic. This means that it is possible to significantly reduce the data size. On this basis, the statistical distribution of the harmonic or the joint distribution of the amplitude and half-period is calculated to characterize the statistical properties of the signals. Finally, to evaluate the similarity of the statistical distribution, one uses the Bhattacharyya distance. A standardized procedure for calculating the similarity between statistical distributions has been developed. The test for polymer flow state monitoring in fused filament fabrication (FFF) validates the availability and efficacy of the proposed method. The impact of the key parameters of window length and cell size is examined. As an alternative AE feature representation method, the proposed method is believed to be suitable for practical process monitoring, fault diagnosis, and other scenarios.