Water supply systems are essential for public health, ease of living, and industrial activity; basic to any modern city.
But water leakage is a serious problem as it leads to deficient water supplies, roads caving in, leakage in buildings, and
secondary disasters. Today, the most common leakage detection method is based on human expertise. An expert,
using a microphone and headset, listens to the sound of water flowing in pipes and relies on their experience to determine
if and where a leak exists.
The purpose of this study is to propose an easy and stable automatic leak detection method using acoustics. In the
present study, 10 leakage sounds, and 10 pseudo-sounds were used to train a Support Vector Machine (SVM) which was
then tested using 69 sounds. Three features were used in the SVM: average Itakura Distance, maximum Itakura
Distance and the largest eigenvalue as derived from Principal Component Analysis. This paper focuses on the Itakura
Distance, which is a measure of the difference between AR models fitted to two data sets, and is found using the
identified AR model parameters. In this study, 10 leakage sounds are used as a standard reference set of data. The
average Itakura Distance is the average difference between a test datum and the 10 reference data. The maximum
Itakura Distance is the maximum difference between a test datum and the 10 reference data. Using these measures and
the PCA eigenvalues as features for our SVM, classification accuracy of 97.1% was obtained.