MetaHeuristics
Previous Projects

Detection of leaks in pipelines by artificial neural networks



A leak detector system was developed for a liquid ammonia pipeline that was able to locate within 100 seconds leaks down to 1% of flow rate. With regard to positionning the segment between break valves where the smallest detectable leak were occuring was 50% successful, but 100% successful in locating 1% leak within 3 segments. The detection system uses a physical model for the pipeline flow rate and a Metacomp based neural model is then trained against the difference betwwen the physical model and measurements of pressure, temperature at various line-break stations, and inlet and outlet flow rates. In long pipelines the inlet flow rate is not equal to the outlet flow rate for nonstationary flows due to compressibility effects. The neural networks perform very well compared to conventional leak detection technology in there circumstances.

PERFORMANCE OF LEAK DETECTION SYSTEMS