Sensor fault detection and diagnosis in drinking water distribution networks

Abstract

In this work, the local PCA approach is used as a statistical process control tool for drinking water distribution(DWD) systems to detect and isolate sensor faults. The multivariate statistical process monitoring task is carried out by learning a finite mixture model to describe the local statistical behavior in each cluster, followed by the determination of the local statistical confidence limits. The objective of a water distribution system is to convey treated water to consumers through a pressurized network pipe. The aim is diagnosing sensor faults in DWD. Experimental results using a model of an actual water distribution network illustrate the effectiveness of the proposed approach.