基于MBW-Faster RCNN的碳纤维复合材料铣削缺陷检测方法OA

When carbon fiber composites were applied to the metro bogies,defects such as burrs,delamination and voids frequently occurred during the milling process,adversely affecting structural performance and service life.In order to efficiently and accurately recognize processing defects,to address the limitations of the Faster RCNN model—namely its weak edge feature extraction capability and high false detection rate,an enhanced defect detection method MBW-Faster RCNN for detecting milling defects in carbon fiber composite material metro bogies was proposed.Multi-Scale Edge Enhancement(MSEE)module was introduced to improve the model's sensitivity to fine defects and edge details in fiber-textured,cluttered regions by combining multi-scale feature extraction with edge information enhancement.In addition,Bidirectional Feature Pyramid Network(BiFPN)was incorporated to strengthen cross-scale semantic fusion and enable the model to extract richer feature information.WIoU V3 loss function replaced the conventional CIoU as the bounding box regression loss,improving localization accuracy and small target recognition performance.Experimental results demonstrated that the proposed MBW-Faster RCNN mAP@0.5∶0.95 was achieved 94.40%,representing a 9.04 percentage points improvement over the baseline Faster RCNN,and showed significant advantages in key metrics such as precision and loss value.