This study presents and tests several approaches for predicting fatigue failure in metals using acoustic emission data analysis. We demonstrate that conventional single-valued AE features (event rate, amplitude, power-law exponents) fail to correlate with fatigue damage progression in selective laser melted additively manufactured AlSi10Mg specimens. Instead, we develop two complementary methods based on spectral density analysis of complete acoustic emission waveforms: (1) a physics-based approach that identifies fatigue-related events through their enhanced excitation of an axial resonant mode, and (2) a data-driven method using dimensionality reduction combined with statistical metrics. Both methods demonstrate the ability to predict approaching failure, with the data-driven approach showing superior consistency. Furthermore, both methods are unsupervised, require no extensive training datasets, and are grounded in physical principles, suggesting potential applicability across different materials and geometries. These findings offer promising pathways for nondestructive fatigue monitoring.

Close

PrevNext

View figure in article

• Fracture
• Mechanical & acoustical properties
• Machine learning