Abstract
Mechanical damage is a leading cause of serious pipeline incidents worldwide. Anomalies caused by it, including those associated with third-party damage, are challenging to detect, classify and size. In particular, in-line inspection (ILI) systems have had difficulty inspecting dent regions for coincidental gouging and distinguishing gouging from corrosion. This is a problem as gouging, which occurs when the pipeline is damaged by mechanical or forceful removal of metal from a local area on the pipe surface, tends to be more susceptible to cracking.
Key to achieving proper mechanical damage assessment is a comprehensive inspection of the pipeline using multiple technologies plus the implementation of advanced data analysis algorithms, including machine learning techniques.
Using the T.D. Williamson (TDW) Multiple Dataset (MDS) platform to provide comprehensive pipeline assessment along with the recent development of advanced machine learning models has resulted in the first gouge classifier backed by an industry compliant performance specification.
Through this innovative development, the pipeline industry has access to accurate classification of dents with coincident gouging. Once accurately classified, validation of depth sizing performance for metal loss, both gouges and non-gouges, located coincident with a dent is achieved.
This paper will outline the history of the gouge classifier development, recent improvements and enhancements made to the classifier, and an overview of the classifier performance when applied to a recent ILI project. Classifier model performance metrics, specifically precision and recall, will be reviewed. Post-classification metal loss depth sizing accuracy of gouge and corrosion features will also be presented. Finally, an overview of the operational benefits of the enhanced gouge classifier results will be provided.
