Fatigue damage monitoring using un-supervised clustering method of acoustic emission signal on SAE 1045 steel

Abstract

This paper described the capability of acoustic emission (AE) technique in monitoring the fatigue damage level using unsupervised clustering technique. As fatigue damage is being a major contributing factor to component failure, it is essential to evaluate the level of damage caused by fatigue load in order to prevent the catastrophic failure of the structure. It is a concern in this study to differentiate the AE signals according to the fatigue damage stages by implementing an unsupervised clustering technique. In this study, the AE signals were collected on specimens made of medium carbon steel SAE 1045 that underwent the axial fatigue testing. The test was run at three loading values of 600, 640 and 680 MPa. The pattern behaviour of AE signals was recorded using a piezoelectric sensor in a form of time domain history signal. Later, the AE signals collected were analysed and clustered using K-means technique. Five clusters of K1, K2, K3, K4, and K5 have been found for the specimens subjected to stress value of 600-680 MPa. The optimum numbers of K clusters were determined using the smallest objective function in their group which ranges between 2.6 to 3.0. This pilot investigation s