A lightweight strategy based on knowledge base reasoning for weld seam detection

With the widespread application of deep learning in weld defect detection, existing object detection models such as the YOLO series and Transformer-based architectures typically rely on abundant texture, illumination, and surface detail features. This dependence results in substantial computational overhead and high energy consumption during both training and inference. In this paper, a lightweight weld defect detection strategy based on knowledge base reasoning is proposed. By retaining only the geometric shape features of defects and eliminating redundant texture information, the proposed method achieves efficient and low-power defect identification. Experimental results demonstrate that the proposed approach significantly reduces the number of model parameters and inference time while maintaining detection accuracy. Consequently, computational resource usage and system energy consumption are effectively minimized. This interpretable and lightweight detection framework provides an efficient solution for green motor manufacturing and industrial energy conservation, and it is particularly suitable for industrial scenarios with stringent requirements on energy efficiency and real-time performance, such as quality control of welding in new energy motors and maintenance of energy-saving equipment.