Machine learning approach for blockage detection and localization using pressure transients

Pipe networks in industries and water distribution systems are prone to the formation of blockages resulting in higher pumping requirements. Thus, detection and localization of blockages are of prime importance to maintain the operational efficiency of the system. In this work, a method is proposed to detect the existence, location, and magnitude of discrete blockages in a single pipeline using pressure transients. The pressure transients interact with blockages and thus the data obtained from them contain necessary features to identify pipeline blockage conditions. Features are extracted from experimentally obtained pressure signals using the TSFresh algorithm and stationary wavelet transform. Recursive feature elimination and support vector machines are used for feature selection and building the classifiers respectively. Three classifiers were built for classifying between blocked-unblocked conditions, locations of blockages, and magnitude of the blockages which resulted in the accuracies of 93.44%, 89.96%, and 80.67% respectively. Another classifier built for the combination of all the above cases resulted in 86.58% accuracy.