Early Leak Recognition and Classification in Two-Phase Flow Systems Based on Machine Learning Using Dynamic Pressure Signals
IEEE Open Journal of Instrumentation and Measurement · 2026 · 5 · 2500314-2500314
DOI: 10.1109/ojim.2026.3706192 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Open Journal of Instrumentation and Measurement, 2026. Verfasst von Alamera Nouran Alquennah und 5 weiteren Autor:innen. Die Arbeit nennt 43 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Qatar National Research Fund (a member of Qatar Foundation) through Qatar National Library.
Abstract (Originalsprache)
Gas–liquid two-phase flows in oil and gas pipelines exhibit highly nonlinear, transient, and regime-dependent behavior, which poses substantial challenges for reliable monitoring and fault detection. Accurate leakage identification in such systems is critical for operational safety, environmental protection, and asset integrity; however, conventional physics-based and empirical methods often struggle under dynamic multiphase conditions. While machine learning (ML) techniques have demonstrated strong performance in flow regime identification and pattern recognition, their application to leakage detection in multiphase pipelines, particularly under offshore operating conditions, remains limited, with most existing studies focusing on single-phase systems. This article proposes a pressure-based artificial intelligence framework combined with a sliding-window feature extraction strategy for leakage detection and classification in gas–liquid two-phase pipelines. The methodology targets three operational scenarios: no leakage, single-point leakage, and multiple leakage points. Statistical features are extracted from windowed pressure measurements to capture transient flow behavior, and multiple data-driven classifiers are evaluated under both low-frequency experimental data and high-frequency numerical simulation data. The approach is designed to enhance robustness against flow regime fluctuations, reduce false alarms, and improve discrimination between leakage scenarios. The results demonstrate that cascaded ML models are well suited for low-frequency measurement systems, while neural network (NN) models exhibit high performance under high-frequency sampling, achieving online inference times of less than 1 s on a low-cost embedded platform. Overall, this study offers actionable guidance for the design and implementation of intelligent leakage monitoring systems, supporting early fault detection and improved integrity management in offshore oil and gas operations.
Förderung
- Qatar National Research Fund (a member of Qatar Foundation) through Qatar National Library
Zitieren
APA
Alquennah, A. N., Ferroudji, H., Rahman, M. A., Ghrayeb, A., Abu-Rub, H., & Barooah, A. (2026). Early Leak Recognition and Classification in Two-Phase Flow Systems Based on Machine Learning Using Dynamic Pressure Signals. IEEE Open Journal of Instrumentation and Measurement, 5, 2500314-2500314. https://doi.org/10.1109/ojim.2026.3706192
Vancouver
Alquennah AN, Ferroudji H, Rahman MA, Ghrayeb A, Abu-Rub H, Barooah A. Early Leak Recognition and Classification in Two-Phase Flow Systems Based on Machine Learning Using Dynamic Pressure Signals. IEEE Open Journal of Instrumentation and Measurement. 2026;5:2500314-2500314. doi:10.1109/ojim.2026.3706192
BibTeX
@article{Alquennah2026,
author = {Alquennah, Alamera Nouran and Ferroudji, Hicham and Rahman, Mohammad Azizur and Ghrayeb, Ali and Abu-Rub, Haitham and Barooah, Abinash},
title = {Early Leak Recognition and Classification in Two-Phase Flow Systems Based on Machine Learning Using Dynamic Pressure Signals},
journal = {IEEE Open Journal of Instrumentation and Measurement},
year = {2026},
volume = {5},
pages = {2500314--2500314},
doi = {10.1109/ojim.2026.3706192}
}
EUVIMED