A statistical model of in-situ vibration sensor detection performance in water distribution pipe networks

ABSTRACT

Detecting leaks in underground water networks is essential for reducing water loss and improving the management of this critical infrastructure. Extensive previous studies have explored the use of vibration sensors in detecting such leaks. Effective deployment of such sensors requires characterising their ability to detect leaks of different sizes, through pipes of different materials, and at various distances from the leaks. We present a statistical model that predicts whether a sensor can detect a leak as a function of pipe materials, sensor-leak distances and leak sizes. Parameter estimates are obtained using hierarchical bootstrap sampling. The proposed statistical model is based on logistic regression. The model is trained and evaluated using ground-truth leak and sensor response data from a large urban water distribution network. The sensor sensitivity model shows a predictive accuracy of around 70% for our test cases. Additionally, the model coefficients are consistent with the expected attenuation based on the physical properties of the different pipe material types. These results provide a quantitative basis for planning sensor deployment in water distribution networks.