Float in the Air: Defect Detection in Pipelines With Electromagnetic Levitation Technology

Internal surface defects (ISDs) in pipelines are critical factors affecting their service life and performance. Therefore, accurately detecting ISDs in pipelines during the manufacturing process is important. However, due to the enclosed nature of the interior of the pipeline, existing detection methods often require manual assistance, and existing automated detection systems typically necessitate the pipeline to be reversed after the detection, which greatly reduce the detection efficiency. These limitations make it challenging to meet the demands for efficient and intelligent detection. To address these problems, this article innovatively introduces the electromagnetic levitation technology (ELT) into the task of detecting ISDs in pipelines, constructing a vision-based detection system that achieves intelligent, efficient, and noncontact detection. In addition, we construct the first dataset for ISDs in tube blanks (TBs), TB1500, which includes six types of defect and diverse defect attributes. Furthermore, we propose a local and layer-wise information propagation network (L<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula>IPNet), utilizing the dimensional attention propagation module (DAPM) and vertical semantic propagation module (VSPM) to achieve local and layer-wise information propagation. L<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula>IPNet achieves the best detection result on the TB1500 dataset, outperforming 19 state-of-the-art methods from the past five years. The source code and the dataset are publicly available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/VDT-2048/L2IPNet</uri>.