Damage Progression of C/SiC Composite by Acoustic Emission Analysis With Supervised and Unsupervised Learning Strategies

ABSTRACT Acoustic emission (AE) monitoring was combined with supervised and unsupervised machine learning algorithms to investigate the effect of cyclic loading on the residual mechanical behavior and damage progression of a C/SiC composite. Tensile tests with AE monitoring were performed on samples before and after fatigue tests with maximum stresses corresponding to 60% and 90% of the material strength. Two approaches were conducted to analyze and quantify the type of damage: A supervised learning approach based on the k‐Nearest Neighbor algorithm, and an unsupervised learning approach using K‐Means. While the unsupervised approach does not require previous knowledge of the material, the supervised approach has a higher specificity for different mechanisms. A comparison of the two methods highlighted that supervised learning generally results in more accurate classifications, especially for distinguishing similar damage mechanisms like fiber debonding and pullout. The unsupervised approach, though less precise, effectively captures broad damage trends and can be advantageous for initial exploratory analysis. Both analyses indicate that the fatigue damage weakens the composite interface, leading to more fiber pullout and individual fiber fracture. In turn, this results in an increase in tensile strength after the fatigue loading, albeit with a decrease in elastic modulus.