Leak Detection in Transportation Pipelines Using Feature Extraction and KNN Classification

A methodology for software based pipeline leak monitoring using feature extraction and classification is presented here. For feature extraction, time domain features of the measured flow, pressure and temperature of the pipe at its inlet and outlet are used. The time domain features are calculated for every time window. Information in the time domain signals is converted to matrix of features. The leak and no-leak signals are classified using k nearest neighbor and Bayesian classifiers. This approach is tested on a real multiphase pipeline with various leak positions and sizes. Also the ability of the classifiers for distinguishing operational changes from leakages is evaluated. Results shows that feature extraction and classification can be applied successfully to the problem of pipeline leak detection.