Pipeline leak diagnosis based on wavelet and statistical features using Dempster-Shafer classifier fusion technique

Leaks in hydrocarbon transporting pipelines cause major problems including environmental hazards and financial losses. Many leakage diagnosis methods try to detect the leaks with a small False Alarm Rate (FAR). However, they are not capable of identifying leakage location and size. In this paper, a novel leakage diagnosis method is introduced which not only detects the leakage occurrence, but also determines its location and size. The inlet pressure and outlet flow signals at different leakage conditions are generated using the OLGA software. Different feature extraction methods including statistical techniques and wavelet-based approaches are used to extract the features from the signals. The statistical and wavelet features are then individually used as inputs to a Multi-Layer Perceptron Neural Network (MLPNN) classifier to determine the leakage state. Finally, the outputs of two MLPNN classifiers are fused by the Dempster-Shafer (D-S) technique. The proposed leakage diagnosis method is applied to the first 20 km of the Golkhari to Binak pipeline located in the south of Iran. Simulation results show that the Correct Classification Rate (CCR) of the simultaneous detection and identification of the leakage location and size is about 95%.