[4A2] Advanced defect detection and classification in sandwich composite structures using statistical and AI-driven techniques
J Aigbotsua¹, B Drinkwater², R Smith² and T Marshall³
¹Baugh & Weedon Ltd, UK
²University of Bristol, UK
³Tom Marshall NDT Ltd, UK
Non-destructive testing (NDT) of sandwich structures with low-frequency vibration is common for ensuring structural integrity. However, industrial practices often rely on limited data and inspection sensitivity focused on a single defect type, which can result in other defects being overlooked and limits detailed characterisation. This study explores advanced strategies for defect detection and characterisation by integrating statistical methods and artificial intelligence (AI)-based techniques. The methods evaluated include a baseline contrast frequency (CF) technique, a root mean square deviation (RMSD) algorithm and a machine learning (ML) approach using an isolation forest model, all analysed within a 12-19 kHz bandwidth. It was found that both the RMSD and ML approaches outperform the baseline method, demonstrating improved detection reliability. Further work investigated the classification of inspection responses into categories such as pristine, delamination and disbonds at both the skin-to-adhesive and adhesive-to-core interfaces. These classes exhibit distinct spectral features that can be captured using machine learning. A model trained on these features showed promising classification results, highlighting its potential for automated defect type identification in sandwich composites.
