Quantitative Benchmarking of Acoustic Emission Machine Learning Frameworks for Damage Mechanism Identification

A challenging opportunity in structural health monitoring of composite materials is using machine learning (ML) methods to classify acoustic emissions according to the damage mechanism that emitted the signal. A wide variety of ML frameworks have been developed; however, lack of ground truth datasets in addition to limited overlap between experimental configurations has precluded any direct, quantitative benchmarking of their accuracy. Here, we generate a ground truth dataset comprised of pencil lead breaks with known angles of incidence, \(\theta \). Each angle generates a unique frequency spectrum that changes continuously with \(\theta \), which could be analogous to attributes of acoustic emission signals generated from failure processes, such as those that occur in composites. Five frameworks are then applied to the ground truth dataset and benchmarked according to their ability to discriminate between two sets of signals with a fixed \(\Delta \theta \). A discussion of their performance as related to choice of features is given, and a set of guidelines for best-practices for feature selection and standardized practices are proposed.