A Defect Detection Method for Steel Pipe Welds Based on XXX-DETR Network

Steel pipe weld defect detection is crucial to ensure industrial safety and quality. During the production process, fluctuations in process parameters, human operation deviations and interference from environmental variables can easily lead to complex and variable defects, which can seriously affect the quality. The traditional weld defect detection methods often have the problems of low efficiency and easy to be affected by human factors, so this paper proposes a steel pipe weld defect detection method based on XXX-DETR network, introducing the CSPMogaBlock module combined with multiple types of convolutional operations, global average pooling (GAP) and nonlinear activation function, to enhance the model's ability to characterise complex defects; using the Heavy parameterised BatchNorm instead of LayerNorm, the RepBN module is proposed to optimise the computational efficiency while stabilising the training; the CSP-MSDC module is adopted to integrate the lightweight design, multi-scale feature weighted fusion and efficient up-sampling strategy to strengthen the model's adaptability and flexibility to defects of different sizes. The experimental results indicate that the model achieves a mAP0.5 value of 98.2%, which is a 4.6% increase over the original RT-DETR. Experimental observation shows that the accuracy of the method of this study occupies the first place among all the listed comparison models, which provides a new direction and technical support for the quality control technology of steel pipe welds.