Clustering-based Crack Growth Characterisation using Synchronised Vibration and Acoustic Emission Measurements

Abstract

Monitoring and predicting fatigue crack growth is a critical part of the prognosis step of Structural Health Monitoring (SHM). It is also arguably the most difficult step owing to the uncertainties associated with future loading conditions and fatigue processes. For this study, the authors focus on the use of machine learning, and more specifically clustering, to find groups of features in measured data that are able to predict the current state of fatigue life. The data, in this case, is a combination of vibration and Acoustic Emission (AE) measurements. Individually, both of these measuring techniques have been previously used to characterise crack growth. Here, it is shown that features that are particularly useful for clustering, and potentially other machine learning techniques can be extracted if AE data is synchronised to the main excitation frequency during harmonic loading.