Machine learning supported acoustic emission technique for leakage detection in pipelines

Abstract

Acoustic emission (AE) based method is a very promising passive measurement technique for detection of faults and incipient damage in in-service structures. Considering the advantage of detecting even the weak acoustic signals emitting from in-service critical infra-systems for characterizing the fault/damages/leakage in the structures, AE technique is considered to be one of the efficient NDT techniques. In the present work, acoustic emission technique has been utilised to detect leakage in the pipelines by systematically analysing the signal parameters. The leakage in the pipeline is simulated by means of pressure release valves provided at identified locations. Leakage detection in the pipe is carried out for different rate of leakage through valve. AE signals are measured from the sensors attached to the pipeline and the measured signals are analysed to extract the leakage sensitive acoustic wave features. The AE features evaluated from the acoustic signals are further processed to identify- and localize-the leakage (varying flow rates) in the pipe. Out of all the AE features, AE counts, cumulative AE energy, and signal strength are found to be very sensitive parameters to indicate the leakage in the pipelines. Further, support vector machine (SVM) learning and Relevance Vector Machine (RVM) pattern recognition algorithms are employed to develop the hyperplanes and to classify the leakage by using binary- and multiclass-classifications. Results of the study clearly showed that the SVM and RVM enabled AE features can effectively be utilised for identification and localization of leakage in the pipelines.