Real-Time Monitoring of Gas Pipeline through Artificial Neural Networks

Considering the importance of monitoring pipeline systems, this work presents the development of a technique to detect gas leakage in pipelines, based on acoustic method and on-line prediction of leak location using neural artificial networks. Audible noises generated by leakage were captured by a microphone installed in a 60 m long pipeline. The sound noises were decomposed into sounds of different frequencies: 1kHz, 5kHz and 9kHz. The dynamics of these noises in time were used as input to the neural model in order to determine the occurrence, magnitude and location of a leak (outputs of the model). The results have shown the great potential of the technique and of the developed neural models. For all on-line tests, the neural model 1 (responsible for determining the occurrence and magnitude of the leak) showed 100% accuracy, except when the leakage occurred through a small orifice (1 mm), with leak located at 3 m from the microphone. In all cases where neural model 1 detected the leak, the neural model 2 (responsible determining the location) could accurately predict the exact location of the leak, except for an orifice of 3 mm, with leakage occurring at the inlet end of the pipeline, showing an error of approximately 1.2 m.