ANN-based Model for Aiding Leak Detection in Water Distribution Networks

Industrial and municipal water distribution networks often have a considerable amount of water lost in transit, particularly in a country like India. Several attempts are being made to detect leaks, of which the physical methods are more commonly employed. However, such methods are reported to be less efficient and time consuming. This paper presents a Neural Network based approach to aid the physical methods. The methodology is demonstrated in a hypothetical unsymmetrical water distribution network. EPANET is chosen to model the hydraulic behaviour of the system. The effect of non-operative/defective pressure meter in the leaking pipe is also studied. It is seen that the proposed methodology performs remarkably well in predicting the exact leaking pipe, the location of leak in the pipe and the leak size. The prediction of leak location in the leaking pipe is more sensitive to the pressure signals from the leaking pipe. Thus, it is concluded that a simple simulation study can be very effective in considerably reducing the time in field methods of leak identification.