197

Victor Ferrer1, Gerard Jover1, Katja Träumner2, Thomas Meinzer2, Spain.2

1NDT Global, Barcelona, Spain. 2NDT Global, Stutensee, Germany

Ultrasound In-Line Inspections (ILI) technologies are a well-established inspection method providing pipeline vendors with a high probability of detection, identification, sizing and reporting of anomalies. In recent years, several automatic algorithms have been proposed to detect and identify relevant anomalies while trying to avoid flagging nonrelevant indications or noise (analysis false positives). Regardless, manual analysis is still needed to discard those false positives and to adjust highlighted areas to the feature itself.

Despite the high efficiency analysis step that these automatic algorithms provide, in the presence of a large number of false positive indications, meeting the required priority notification timelines is challenging.

In this paper, we will present an automatic ranking algorithm based on predicted feature depth for ultrasound metal loss technology. By 1.) leveraging deep learning (DL) technology, 2.) biasing the algorithm for deep but rare features (more precise on critical deep features) and 3.) mimicking human analysis, a conservative and easy to use algorithm (no parameters) was developed. It allows manual analysis to focus on a subset of initial indications ranked as potential actionable anomalies and thus reducing priority notification time. By knowing beforehand which are the most important actionable anomalies, a stricter quality checking can be used during human analysis.

Read/Download PaperGradeAvailable after the event