Machine Learning for Acoustic Emission Signatures in Composite Laminates

Development of new composite material systems and structures require detailed understanding of failure mechanisms and failure progression. The identification of damage initiation on a damage-mode basis is essential for validation of progressive damage analysis techniques. Damage progression is the industry pathway towards expanding the composite design space to limit weight and increase confidence in structural performance. The challenge in validating these techniques is that the initial damage events predicted by analysis occur at a scale that are smaller than the defects measured by standard laboratory level inspection techniques (x-ray, ultrasound, digital image correlation). Of the techniques, acoustic emission has been found to be one of the most sensitive methods for identifying the onset of damage. However, characterizing the nature of the damage event from the acoustic emission signal has been challenge for many years. This paper describes a method for classifying acoustic emission signals that applies machine learning to a feature called the frequency distribution function. The method is demonstrated on composite laminates specimens and compared to observations from in-situ testing inspection techniques.