Machine learning-based identification and classification of acoustic emission signals from fracture process zones

Since exploring the tensile properties of cementitious materials is an effective method for identifying the damage state of a structure, a concrete cylinder is conducted by an ultimate tensile test using a pressure tension machine, where the fracture processes are monitored by acoustic emission (AE) techniques. Based on the assessment of the proposed external evaluation framework, the infinite Bayesian gaussian mixture model (BGMM) achieves superior performance with high scores across various evaluation metrics, including Precision (0.9915), Recall (0.9871), F1 Score (0.9892), Jaccard Similarity (0.9892), Mutual Information (1.1875), and Fowlkes-Mallows Index (0.9902). These scores are approximately 0.2 points higher than those achieved by GMM and K-Means, highlighting its enhanced clustering capability. The infinite BGMM is therefore introduced to detect and classify the AE signals from fracture process zones (FPZs). As the analysis shows, in the frequency centroid (C-FRQ) domain, the distribution of the AE hits in a FPZ is statistically defined as a two-component Gaussian mixture model composed of a normal distribution of hits from shear cracks alongside a normal distribution from tensile cracks. Concurrently, as the applied pressure increases, the means of the C-FRQ distributions of the shear and tensile cracks within a fracture zone gradually converge.