Hydraulic Transient Analysis and Leak Detection on Transmission Pipelines: Field Tests, Model Calibration, and Inverse Modeling

The use of hydraulic transients for leak detection is theoretically possible assuming that water pipelines respond elastically and that current transient models are capable of replicating measured responses from real pipelines. This paper presents results for tests using hydraulic transients with and without a leak on a typical transmission main in South Australia. The size of the leak artificially introduced to the pipeline was set at the maximum limit of interest to South Australian Water Corporation operators. Based on the results of the field tests and modelling performed using a quasi-steady friction transient numerical model it was found that it was difficult to model the response of the pipeline, without and with the introduced leak, because of unsteady friction and mechanical dispersion and damping of the transient waveforms. Inverse analysis was performed using the quasi-steady friction transient model and it was found that leak could not be successfully detected. The transient model was improved by including unsteady friction and a "viscous" damping mechanism that was calibrated for inelastic mechanical effects using no-leak measured responses. Inverse transient analysis was performed using this improved model focussed on reflection information over 2L/a seconds of the measured leak responses and over an extended period. The small size of the direct reflections from the artificial leak made them difficult to discern amongst other reflections from elements not related to the leak. The inverse transient analysis performed over an extended period made use of leak damping information but was also affected by sources of damping not related to the leak. It was found that the improved forward transient model, in combination with prior information regarding the leak discharge (commonly available for flow monitored transmission pipelines), gave the best estimate of the location and size of the leak. However, the "true" leak was not identified as the optimal candidate following the inverse transient analysis because of persistent inadequacies in the replication of all the physical complexities affecting the measured transient responses.