Advancements in machine learning, particularly Deep Learning, transform and automatize more and more processes during data analysis of complex In-Line Inspection (ILI) data. These new technologies hold the potential to not only surpass traditional algorithms but also introduce innovative, multi-technology, parameter-free solutions, enhancing the robustness and automation of ILI services.

Girth weld detection is a crucial aspect of ILI processes, traditionally managed and automatized by classical algorithms that despite their reliability, encounter significant challenges particularly when dealing with recording artifacts, e.g. induced by movements or liftoff of sensor carriers when passing girth welds, noisy data, extensive installations, or in the presence of sleeves. The need for tailored implementations across various ILI technologies adds considerable maintenance overhead, increased deployment cost and added usage complexity.

In this paper, we present a novel approach for girth weld detection that overcomes all these limitations by applying Deep Learning technology. Our method employs 1D keypoint detection offering real-time application, higher accuracy compared to classical detection approaches in complex scenarios, and a parameter-free approach for multiple technologies. By leveraging circumferential girth weld characteristics, the underlying Deep Learning technology processes multichannel signals instead of images, speeding up the process by a factor of 100.