Pattern recognition and acoustic emission

Abstract Development of a pattern-recognition method to discriminate between emission from different acoustic emission (AE) sources is described. The method is based on extracting features of AE waveforms from an unknown source using computer analysis, and then comparing these features with those predicted for possible sources using theoretically and experimentally determined models of sources and calibration studies. The feasibility of using pattern-recognition studies to establish a basis for distinguishing between waveforms from a damage-related source and from several extraneous sources is demonstrated: features obtained from the autocorrelation function and duration of an AE signal are used to separate AE from inclusion fracture and AE from gas flow, electromagnetic interference, mechanical noise and transient electronic noise. The importance of calibration studies when distinguishing between different locations for a particular damage-related source is also illustrated. For the future, the analysis should be extended to incorporate other damage-related sources and additional extraneous sources such as fretting.