Volume 46: Energy Transitions toward Carbon Neutrality: Part IX

Enhanced Gas Pipeline Multiple Leak Detection Using Artificial Neural Networks in Complex Multiphase Flow Conditions Wahib A. Al-Ammari, Ahmad K. Sleiti

Abstract

When fast and accurate detection is crucial, such as in cases of pipeline leakage accidents, machine learning is a powerful tool for integrating real data with artificial intelligence. In this context, this study introduces reliable artificial neural network (ANN) models that are intended to precisely locate single or several leaks in undersea gas pipelines. Using OLGA multiphase software, thorough data for numerous leak scenarios are generated under multiphase flow conditions and based on an actual offshore pipeline profile. These data cover measurable flow variables (mass flow rate, temperature, and _x000D_
pressure) at the inlet and outlet of the monitored pipeline. The developed ANN models were trained to estimate the leak’s location(s) and size(s) under single leak as well as multiple leak conditions, which have not been investigated before in open literature. The findings show that for single leak scenarios, the ANN models can detect and identify the leak size and location with an error of less than 1.0% across all phase flow scenarios. The leak localization error, however, is higher than 8.0% under multiple leak conditions. The outcomes of this study also show that the accuracy of the ANN models is significantly influenced by the phase flow conditions and number of ANN input parameters. To improve ANN performance and minimize the impact of noise signals, it is suggested to employ a multi-stage leak identification technique. To do this, an ANN model with two inputs must be utilized first, and subsequently, a model with six inputs must be progressively added. In this study, our approach and methods for generating leak-based data and identifying leaks in multiphase flow and various leak conditions can serve as a foundation for enhancing the efficiency and reliability of future machine learningbased leak detection systems.

Keywords offshore gas pipelines; leak detection; leak localization; multiple leaks; multiphase flow; artificial neural network