Automated detection of subsurface defects in GFRP by integration of barker coded thermal wave imaging and deep learning

Abstract Automating industrial production processes with deep learning-based image processing improves detection efficiency, real-time performance, and accuracy, while enhancing product quality throughout the manufacturing cycle. Integrating deep learning algorithms with image processing enables automatic detection of subsurface defects, such as scratches, spots, and holes, with high precision. Glass fiber reinforced polymer (GFRP) composites are vital in the aviation, automotive, and electronics industries because of their low weight, corrosion resistance, and long lifespan. Monitoring the condition of GFRP materials is crucial for improving safety, serviceability, and operational limits. Non-destructive testing and evaluation is widely used for damage detection and material characterization due to its non-invasive nature. Thermal image processing (thermograms) is essential for improving data quality, but challenges such as noise and non-uniform backgrounds complicate defect identification. These challenges require deep learning-based image processing techniques. An experimental study is conducted on GFRP samples containing Teflon inserts at varying sizes and depths. In this paper, a novel approach is presented that integrates image processing with deep learning algorithms to automatically detect defects. The faster region-based convolutional neural network, which integrates a feature pyramid network and a region proposal network, is used for robust feature extraction, precise region identification, and efficient defect detection through region of interest pooling and fully connected layers. The proposed method significantly improves defect detection performance, achieving an average precision (AP)@0.50 of 0.86, mean AP @[0.5:0.95] of 0.612, and average recall @100 of 0.87 on thermographic GFRP data. The presented framework provides a promising solution for automating material condition monitoring, offering improved reliability and efficiency in defect detection in various industrial applications.