Feature extraction in pressure signals for leak detection in water networks

Abstract

Techniques based on signal analysis for leak detection in water supply systems typically use long pressure and/or flow data series of variable length. This paper presents the feature extraction from pressure signals and their application to the identification of changes related to the onset of a leak. Example signals were acquired from an experimental laboratory circuit, and features were extracted from temporal domain and from transformed signals. Statistical analysis of features values and a classification method were applied. It was verified the feasibility of using feature vectors for distinguish data acquired in the absence or presence of a leak.