A study on the application of machine vision technology in defect detection of CFRP composites

Traditional damage detection methods for carbon fiber-reinforced polymer (CFRP) composites primarily rely on manual operations, resulting in low efficiency and accuracy. This study proposes a high-precision and rapid damage detection approach for CFRP composites based on machine vision. First, three common types of CFRP damage specimens were prepared through controlled experiments, and ultrasonic C-scan imaging was employed to acquire damage images. A dataset comprising 6650 annotated images was constructed through data augmentation techniques. Subsequently, the Mask RCNN model was enhanced in two key aspects: (1) integrating the Sim-AM attention mechanism to improve feature extraction capability, and (2) replacing the conventional Non-Maximum Suppression (NMS) algorithm with Soft-NMS. This study employs an improved Mask RCNN model to train and test the dataset. Experimental results demonstrate that the model achieves significant enhancements in detecting CFRP (carbon fiber-reinforced polymer) damage, with Precision, Recall, and mAP (mean Average Precision) reaching 97.2%, 94.5%, and 95.1%, respectively. The detection speed attains 4.9 FPS (frames per second). These metrics indicate that the proposed model outperforms both the standard Mask RCNN and other mainstream object detection models, exhibiting superior detection accuracy to meet industrial requirements. Additionally, this work provides a novel research approach for damage detection in this field.