An inverse approach for piping networks monitoring

Spills and leakages of hazardous fluids from piping networks may pose a significant safety risk to population, industrial plants and the environment. Therefore in fluid distribution the problem of monitoring the network status in order to identify abnormal conditions and locate leakages arises. In the paper an inverse approach resorting to a multi-layer perceptron back-propagation Artificial Neural Network (ANN) is proposed, in order to locate leakages based on pressure and flow rate information. Strategies for generating input data and for correlating by ANN such data to the fluid distribution system status are presented. A two-level architecture is selected, composed by a main ANN at the first level and several branch-specific second-level ANNs in cascade to the main one. The branch in which the leakage occurs is identified, resorting to the ANN operating at the first level, while the specific second-level ANN is activated to estimate accurately the magnitude and location of the leakage in the selected branch.