Convolutional neural network for automatic defect detection in composites

• Main Authors: Prasanthi, Y. Naga, Ghali, V. S., Vesala, G. T., Suresh, B.
• Format: Conference Proceeding
• Language: English
• Subjects: Aerospace industry; Anomalies; Artificial neural networks; Carbon fiber reinforced plastics; Classification; Defects; Fiber reinforced polymers; Machine learning; Nondestructive testing; Thermal imaging; Thermal wave imaging; Thermography
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0111836

Fiber-reinforced polymers (FRP) are widely recommended in the aerospace and automotive industries because they are lightweight and have superior mechanical qualities to metals. Poor manufacturing conditions or cyclic usage leads to defect generation in these materials, and a proper non-destructive testing (NDT) method assesses integrity without impairing their future application. Infrared NDT (IRNDT) is gaining interest with machine learning-based advanced processing approaches in the recent past, among numerous NDT techniques. However, the thermographic data is highly imbalanced with fewer defect region thermal profiles than their non-defective counterparts. The present article introduces a deep one-class classification (Deep-OCC) model in quadratic frequency modulated thermal wave imaging for automatic anomaly detection. A carbon fiber-reinforced polymer (CFRP) sample with flat bottom holes of different sizes at various depths is used to certify the proposed methodology. Few machine learning (ML) and thermographic metrics are used to evaluate the suitability of the deep OCC by comparing it with conventional deep one-class classification models for the thermographic data.