Fault Detection System for Long-Distance Gas Mixture Pipelines Using Statistical Features

Integrity management of gas pipelines can be enhanced by incorporating state-of-the-art failure detection, diagnostics and prediction tools. A plethora of methods are available for real-time monitoring and leak detection, the majority of which reported for liquid pipelines that can be quickly introduced for real-life applications. The present paper, however, proposes dynamic principal component analysis (DPCA) for it has not been tested for gas pipelines under transient conditions. Mass flow rate, temperature and pressure values are used separately and in combined form to establish the reference models. Measured data are projected into the new dimension based on selected principal components. For leak detection, Hotelling’s T2-statistics and Q-statistics are monitored in real time. The validation tests for simple as well as dynamic PCA show that both methods successfully detect a leakage that has an opening of 10% of pipeline diameter. DPCA significantly magnified the information on leak in terms of T2-statistics, thus reducing the probability of missed faults. T2-statistics is found to be more sensitive to small leaks than Q-statistics. Overall, it can be said that the proposed technique has the potential to accurately identify small leaks under transient conditions.