Acoustic emission classification for failure prediction due to mechanical fatigue

Acoustic Emission signals (AE), generated by the formation and growth of micro-cracks in metal components, have the potential for use in mechanical fault detection in monitoring complex-shaped components in machinery including helicopters and aircraft [2]. A major challenge for an AE-based fault detection algorithm is to distinguish crack-related AE signals from other interfering transient signals, such as fretting-related AE signals and electromagnetic transients. Although under a controlled laboratory environment we have fewer interference sources, there are other undesired sources which have to be considered. In this paper, we present some methods, which make their decision based on the features extracted from time-delay and joint time-frequency components by means of a Self-Organizing Map (SOM) neural network using experimental data collected in a laboratory by colleagues at the Georgia Institute of Technology.