A novel pipeline leak detection approach independent of prior failure information

Abstract Condition monitoring of pipelines is of importance to detect fluids leakage and associated financial losses and accidents. Artificial intelligence (AI) techniques have been widely used for the pipeline condition assessment. A major limitation of the currently-used supervised AI methods is that they heavily rely on sufficient pipeline failure historical data for their training. To cope with this issue, this paper proposes a health index-oriented approach based on multiscale analysis, Kolmogorov-Smirnov (KS) test, and Gaussian mixture model (GMM) for determining the leakage situation in pipelines. GMM is an unsupervised AI method capable of training itself with pipeline normal condition data. In this study, acoustic emission (AE) signals are first acquired from the pipeline at different pressure conditions. Then, the multiscale analysis and KS test are deployed to extract suitable features from the AE signals. The feature samples corresponding to the pipeline normal condition are used to train the GMM. Finally, the feature samples to be tested are supplied to the GMM and the desired health indicator is obtained. The results confirm the effectiveness of the proposed approach in discriminating the normal and leak conditions as well as the severity of leaks. Further, the GMM classifier trained with features derived from multiscale analysis and the KS test outperforms the GMM trained with crest factor, and mean frequency.