LAUSR

Abstract One of the most important issues arising in the use of acoustic emission (AE) for nondestructive process monitoring is the accurate identification of potential process malfunctions to avoid premature failure. In some cases, the AE signals from malfunction sources are relatively weak, with high levels of background noises. Thus, these signals could easily be submerged, making it rather difficult to separate them. Therefore, it is of critically importance to find a solution to the problem of weak emission source identification and obtain a correct representation of original waveforms. The present work proposes a new feature representation method based on similarity of probability distributions from raw AE waveforms. The Bhattacharyya coefficient is used for this purpose. A standard procedure for calculating similarity is formulated. Both an instantaneous similarity and a relative similarity are defined. The influences of the choice of some key parameters are discussed in detail. Tests on filament breakage detection in an additive manufacturing process reveal the feasibility and effectiveness of the proposed method. This method is believed to be appropriate when the target malfunction emission signal amplitude is less than the environmental emission signals generated by other stationary sources, and threshold methods fail to perform properly. It could also be used as an alternative feature representation method for AE signals in other fields.