Leaks in the pipe or pipeline networks has direct financial and environmental consequences. It is required to have timely and accurate leak detection. Sensors placement plays an important role toward accurately finding the leaks. For the pipes with various lengths, finding the optimal places to deploy the sensors is a difficult problem. The leak detection problem may be divided into two stages which complement each other. The first stage is devoted to sensor placement and next is leak detection through analyzing the data gathered from the sensors.
In this paper, however, we will propose a new method to facilitate detection of leak with higher precision. Assuming there are m potential nodes in the water distribution network (WDN) where required sensors for leak detection can be deployed. What we have proposed in this paper is finding min(n) nodes where deploying sensors there, guarantees the required accuracy for leak detection. To identify the minimum n, in the second stage initially we use all the m sensors and by defining various leak scenarios that are defined based on model-based methodology, we calculate leak detection accuracy. In order to minimize the number of required sensors, we relied on the combination of hydraulics equations and data mining models. We utilize hydraulics equations to find sensitive features for leakage and using data mining classifiers to find the good placements with high performances. At the second stage, iteratively by reducing the number of sensors, and using trial and error, we find the optimal distance between the two sensors that provide an acceptable result. When the optimal number of sensors are found, the costs of sensors, installation, maintenance, communication and processing are decrease. This would improve the leak prediction costs. This idea can also be extended in other fluid transporting by pipelines.