Leak detection model for pressurized pipelines using support vector machines

Aging infrastructures, specifically pipelines, that were installed quite a while back and currently operating under poor conditions, are highly susceptible to the threat of leaks, which pose economic, health, and environmental threats. For example, in the year 2009, the state of Ontario lost 25% of its water supply solely due to leaks. The amount of lost water is equivalent to the volume of 131, 000 Olympic swimming pools and worth 700 million Canadian dollars. Therefore, a need arises to develop an approach that allows condition monitoring and early intervention. This article proposes a model for a real-time monitoring system capable of identifying the existence of single event leaks in pressurized water pipelines. The model relies on wireless accelerometers placed within the network on the exterior of the pipelines. The vibration signal derived from each accelerometer was assessed and analyzed to identify the Monitoring Index (Ml) at each sensor on the pipeline. The data collected from experimentation were analyzed by means of support vector machines (SVM) technique. A leak threshold was determined such that if the signal increased above the threshold, a leak status is identified. Experiments were performed on one inch cast iron pipelines, one inch and two inch PVC pipelines using single event leaks and the results were displayed. The developed models showed promising results with 98.25% accuracy in distinguishing between leak states and non-leak states.