Assay of Metal Loss in Pipelines With Repaired Sleeves Using Machine-Learning-Assisted Fiber-Optic Distributed Acoustic Sensing

Metallic pipelines are essential in the oil and gas industry. Hence, various technologies have been developed to monitor their integrity, particularly in diagnosing corrosion-induced metal loss. Once a fault is detected, the structure is reinforced with a composite sleeve as a provisional measurement to extend the lifespan of the asset before replacement. However, these temporary solutions significantly hinder the ability of conventional monitoring technologies to track the progression of metal loss. Their inherent characteristics, such as high-vibration insulation, reduce the effectiveness of traditional sensors. To address these limitations, we report on the use of fiber-optic distributed acoustic sensing (DAS) integrated with machine-learning (ML) algorithms. This approach facilitates the creation of a novel distributed monitoring tool capable of accurately detecting and tracking postrepair metal loss. The proposed approach collects temporal and spectral data from operational sounds along an optical fiber wound around the pipeline, while an ML algorithm classifies these signals as indicating intact or faulty operation in unrepaired sections and quantifies the damage progression of repaired sections, providing a comprehensive solution for timely detection, repair, and replacement decisions. As a proof-of-concept demonstration, we induced artificial sounds generated from numerical simulations in real repaired and unrepaired sections of a metallic pipe, mimicking those produced by fluid flow in pipelines under different metal loss etching depths. The system captured samples of these sounds to train and test the performance of convolutional neural network (CNN) models, achieving fault detection accuracy of up to 99.93% and postrepair metal loss quantification accuracy exceeding 85%.