Nonlinear frequency modulation thermography for defect detection in CFRP composites based on deep learning YOLO-V8 network model

This paper proposes the application of nonlinear frequency modulation (NFM) thermography to detect delamination defects in carbon fibre reinforced polymer (CFRP) composites. A series of artificial flat-bottomed holes (FBHs) were prepared to simulate the delamination defects in a CFRP specimen. An NFM signal was generated for stimulating CFRP specimens for defect detection. The image sequences were processed by several data dimensionality reduction techniques, including principal component analysis (PCA), independent component analysis (ICA), linear discriminant analysis (LDA) and quadratic discriminant analysis (QDA). The signal-to-noise ratio (SNR) was calculated to assess the performance of above techniques. QDA remarkably improves the SNR, outperforming PCA, ICA, and LDA techniques. Furthermore, a Faster R-CNN, YOLOv5 and YOLOv8 object detection model was applied to analyse the QDA-processed results. The result showed that the YOLO-V8 network model can effectively identify and annotate defects. YOLO-V8 was successfully identified as small-sized defects with a diameter of Φ = 4 mm at a depth of h = 0.4 mm and h = 0.8 mm in CFRP composites. By combining signal processing with the deep learning YOLO-V8 model, their respective advantages are effectively brought into play, providing an effective method for detecting defects in CFRP composites.