Characterizing Severity of Barely Visible Impact Damage (BVID) with Machine Learning

This work uses a machine-learning-based approach to qualitatively assess BVID with features generated by guided waves propagating in a damaged material medium. Low-velocity impacts causing BVID in composite materials are a concern while performing in-field service inspections on aerospace structures. Furthermore, BVID may increase in size with continued service operations, compromising the structure’s integrity. Identifying such damage remains a challenge with the current nondestructive evaluation (NDE) techniques practiced in the industry. The use of guided waves for damage identification has been a proven concept for a long time; however, extracting qualitative or quantitative information on the damage from the received wave signals is still a demanding process involving many obscurities caused by interactions of guided waves with damage. In fact, the complex patterns arising from guided waves as they pass through a damaged medium can present a unique opportunity if treated appropriately. These patterns can be viewed as a digital fingerprint of damage and employed as an input for machine learning models to extract features for categorizing damage severity. The current work uses data sets generated for various realizations of BVID via finite element models where parameters signifying different intensities of BVID are varied. Then the digital features extracted from machine learning models are used to categorize the BVID into low, medium, and high damage classes. This work successfully identifies the severity of BVID, providing a methodology that can be adapted to quantify BVID during preliminary diagnosis and to make necessary repairs.