The acoustic emission technique enables the detection of elastic waves released from fretting-induced micro-level events. Proper processing techniques are essential for interpreting and analysing the complex, high-volume acoustic emission data. In this study, unsupervised machine learning algorithms are applied as the processing techniques to study the correlation between fretting-induced damage and acoustic emission signals in a flat-on-flat fretting contact configuration. As a first step, the correlation coefficient method is employed to identify the most significant features from the measured acoustic emission data. Then, training and clustering are performed using self-organizing map and k-means algorithms on the selected features. To achieve effective clustering, the optimal number of clusters is determined through quality index evaluations. Finally, the trained self-organizing map–k-means model is applied to data from tests with varying load cycles to evaluate the clustering performance of the trained model and facilitate interpretation of clusters behaviour. Clustering results, supported by material characterization techniques (optical imaging, profilometry, and scanning electron microscopy), demonstrate that the trained model accurately classifies data from various specimens and provides valuable insights into the relationship between acoustic emission features and fretting features and damage mechanisms such as crack initiation and propagation, adhesive wear, abrasive wear, and debris accumulation.
Khoshroo et al. (Wed,) studied this question.