2024 – 07

Amir Behbahanian1, Adrian Belanger2, Robert Coleman1, Paul Dalfonso1, Ron Lundstrom1

1T. D. Williamson, Salt Lake City, USA., 2T. D. Williamson, Houston, USA

1T. D. Williamson, Salt Lake City, USA. 2T. D. Williamson, Houston, USA

Abstract

 Machine learning tools have been used for over a decade to process the large amounts of in-line Inspection (ILI) data and to report accurate sizing of metal loss features.  With the explosion of software solutions that came with the advancement of graphic processing units, (GPU) power and memory available to the commercial market, the use of machine learning in processing magnetic flux leakage (MFL) tools for corrosion sizing now standard in the industry.  An aspect of automation that does not get as much attention is the identification of features such as fixtures.  Traditionally this task has been done either manually or with a semi-automated process based on signal pattern recognition by utilizing operator provided survey data.   As fixtures are commonly used as a reference point when performing dig verifications, fast and effective methods for locating them are helpful.

This paper will present an approach using a machine learning method called semantic segmentation.   There is an enormous amount of information contained in an MFL survey that has an image like structure with magnetic measurements sampled both axially and circumferentially on a grid. The powerful technique of semantic image segmentation, which is used in applications like autonomous driving, medical imaging, urban planning, manufacturing, robotics, etc., is ideal for analyzing this data.

Four different models will be presented to demonstrate how a combination of them can perform the task of identifying and classifying pipeline features.  This work will lay the foundation of not only automating a semi-manual task but training models to focus on features of concern to the operator.   Due to concerns about aging infrastructure, identifying certain types of fixtures can help with integrity threat management.  The use of these models has the potential of identifying specific features that have known integrity concerns and facilitate a pipeline owner’s remediation plan.

Read/Download PaperGrade