Probabilistic leak detection in pipelines using the mass imbalance approach

Typically, models of pipelines and pipe networks are calibrated to metered data by optimizing the choice of parameters according to some penalty function. This approach does not provide a natural way to assess the predictive uncertainty when these models are used to infer the presence and description of a leak. This paper describes a fully probabilistic approach in which the activities of calibration and prediction are unified, using the massimbalance approach to leak detection as an example. The resulting probability distribution over leak location and size can be presented graphically, or it can be used within an optimal decision framework to compute an effective response taking uncertainty into account. The approach is generalized to different leak-scenarios, which could include multiple leaks.