Pressure Sensor Placement for Leakage Detection and Calibration of Water Distribution Networks Based on Multiview Clustering and Global Sensitivity Analysis

Abstract

Most previous studies in the field of sensor placement have focused on only one aim. In this study, pressure sensor placement is done for calibration and leakage detection simultaneously to use pressure data optimally to save money and time. The sensor placement method implemented in this paper consists of two main parts: clustering the nodes of a water distribution network (WDN) and determining the representative node of each cluster as the sensors’ location. A new multiview clustering approach is implemented to cluster nodes of a WDN based on two pressure sensitivity matrices. In fact, two different aspects of nodes’ characteristics are used for clustering. The representative node from each cluster is also chosen based on global sensitivity and the number of detection criteria. The Sobol method is used for global sensitivity analysis, and the number of detections is calculated with the local sensitivity matrices. The performance of sensor placement is evaluated individually and collectively for different goals in the Anytown network. The accuracy of network calibration with the sample design proposed in this study is equal to 0.0707 m, which is the lowest value between previous studies. Leakage detection also has a significant performance ratio than random pressure sampling. Furthermore, this new method of sensor placement is evaluated in the more extensive networks of the C-town and the Modena, which have high complexity. The performance of the presented pressure sampling is significantly different from the random pressure sampling, and the pressure data collected at the selected nodes can be used for the calibration procedure efficiently.