Fault detection with Distributed PCA methods in Water Distribution Networks

This paper describes a fault detection and diagnosis method applied to a water distribution system. The main purpose, in this kind of systems with a very big amount of data, is to find a fault detection method which can achieve the best performance, while reducing computation and communication costs. The approach proposed here divides the installation sensors into several and possibly overlapping blocks, in each of which a local principal component analysis (PCA) is performed to detect and diagnose faults. After that, a central processor will receive the minimal possible information from all the nodes to take a global decision about fault detection and identification, i.e., the variable most responsible for the fault, for the whole plant. This distributed PCA method (DPCA) is compared with other distributed PCA methods, as well as a centralized PCA model, in order to get an idea of how good the proposed method is. Experimental results on the water network demonstrate that this DPCA method with local PCA models achieves good performance.