Water Pipeline Leakage Recognition and Localization Using Machine Learning and Deep Learning Techniques

Abstract

Water distribution systems often face problems with leaks, causing water loss and environmental worries. In literature, applications of machine learning (ML) and deep learning (DL) algorithms in detecting a pipeline leak are tremendous, which helps to avoid wastage of water and environmental worries. For pipeline leak location, there are a few DL-based techniques, but ML techniques are not available. This work’s primary goal was to investigate various machine learning and deep learning techniques for leak identification and localization utilizing data gathered from an acousto-optic sensor to determine a more effective and precise approach. ML algorithms explored in this study are k-nearest neighbors (KNN), decision tree (DT), random forest (RF), categorical boosting (CatBoost), eXtreme Gradient Boosting (XGB), and adaptive boosting (AdaBoost). DL models explored in this study are the recurrent neural network (RNN), convolutional neural network (CNN), VGG16, and region-based convolutional neural network (RCNN). For ML algorithms, 10 traditional features were extracted from the raw one-dimensional time series data. For DL methods, the collected data underwent preprocessing. The preprocessing included data augmentation and normalization to ensure high-quality and consistent results. XGB provided the highest leak detection and localization accuracy among the ML algorithms. All the DL methods were tested at two different pressures, namely 200,000 Pa (2 bar) and 300,000 Pa (3 bar), to check the stability. Furthermore, the RCNN provided the highest leak detection and localization accuracy among the DL methods at pressures of both 2 and 3 bar.