Automatic flaw detection of carbon fiber prepreg using a CFP-SSD model during preparation

Abstract

As an intermediate material for carbon fiber composites, surface flaws inevitably occur during carbon fiber prepreg preparation, which will seriously affect the quality of carbon fiber composite products. The current approaches for identifying flaws on carbon fiber prepreg have the drawbacks of being labor-intensive and inefficient. This research puts forward a novel model for identifying surface flaws on carbon fiber prepregs using an improved single-shot multibox detector (SSD), called CFP-SSD model. A machine vision-based platform for surface flaws identification on carbon fiber prepreg is created. Additionally, the modified-Resnet50 backbone employed in the proposed CFP-SSD model can enhance the effectiveness of network feature extraction. Then, the multi-scale fusion remote context feature extraction module is designed to efficiently fuse the information from the shallow and deep layers. The findings of performance comparison experiments and ablation experiments indicate that the proposed CFP-SSD model achieves 86.63% mean average precision and a detection speed of 47 frames per second, which is sufficient for real-time automatic identification of carbon fiber prepreg surface flaws.

Attention mechanism based on deep learning for defect detection of wind turbine blade via multi-scale features

“…Object detection algorithms of deep learning could be broadly classified into two types: single-stage and two-stage detection algorithms. Single-stage algorithms, such as the YOLO series [23][24][25][26][27] and SSD [28,29], make direct predictions of the categories and positions of targets. Two-stage algorithms are employed by the R-CNN series [30][31][32][33], which includes R-CNN, Fast R-CNN, Faster R-CNN, and Mask-R-CNN.…”