Smart and Sustainable Leakage Monitoring for Water Pipeline Systems

Abstract

It is estimated that about 20% of treated water is lost through leakages in the distribution pipelines in the U.S. and this percentage is much worse in developing nations. It is vital to minimize such leakage given the growing scarcity of freshwater in many regions across the world and the possibility that a leaking pipeline could eventually fail resulting in significant consequences. Furthermore, minimizing leakage will reduce the operational costs of water utilities. Modern leakage detection techniques are capable of locating leaks when there is prior knowledge of the leak to be located in a certain section of the system. Many such techniques require human intervention to successfully locate leaks. There is growing interest in developing sustainable leakage monitoring systems that are embedded in the pipeline infrastructures and are selfpowering. This paper presents the experimental evaluation of a novel framework of surface vibration-based pipeline leakage detection that could work with self-powering sensors which harvest locally available energy. The alternatives to local energy harvesting are either batteries or wired power sources, which are environmentally harmful and inconvenient, respectively. This paper will specifically outline the: (a) development and testing of a robust vibration-based leakage detection technique on buried water pipelines using an experimental test bed built on the Clemson University campus; (b) investigation of leveraging deep learning algorithms to detect leakage using vibration signal data collected from multiple locations along the pipeline length; and (c) investigation of the availability and variation of the harvestable energy in water pipeline systems to support leakage monitoring sensors. A leakage detection index (LDI) is formulated to determine the leakage presence and its relative severity. LDI quantifies the changes in the cross-spectral density of acceleration signal measured at two locations on the leaking pipeline relative to a baseline (i.e., non-leaky) state. The results show that the LDI method is effective in detecting the small leakages in the buried pipelines, which can go undetected for a long time because of their minimal impact on vibration signal and pressure. Subsequently, it is shown that leakages may be detected using machine learning algorithms by bypassing the knowledge of the engineering system dynamics. As a preliminary investigation, an artificial neural network (ANN) model is developed to predict the leakage size and location using the LDI values. The ANN model is able to predict the leakage size and location with more than 80% accuracy. A convolutional neural network (CNN) model has been trained using the layers of AlexNet (a pre-trained network) on scalogram images that are created from acceleration signal data collected across multiple location along the pipeline length to detect the leakages. The CNN model is able to detect the leakages on buried and unburied pipeline with approximately 95% accuracy. Furthermore, various experimental configurations are tested across the pipeline network to evaluate the energy harvesting potential of flow-induced surface vibrations by changing the soil bedding conditions and matching the resonance frequency of the piezoelectric films with the natural frequency of the pipeline system using appropriate tip masses. It is hypothesized that an energy harvesting system with multiple piezoelectric films connected in parallel may be sufficient to power a small leakage detection sensor mounted on the buried pipeline. The findings of this study offer promise for continuing the development of long lasting smart and sustainable leakage monitoring systems for buried pipeline infrastructures and eventually lead to reduced water losses.