Efficient selection of time domain features for leakage detection in pipes carrying liquid commodities

In this paper, a classification approach is proposed for the leakage detection in pipes carrying liquid commodities in the pipeline network of an oil refinery. Leak detection is treated as a binary classification task. Time domain features are computed from acoustic signal measurements using accelerometers mounted on the surface of the pipes. An efficient feature selection procedure is applied, combining correlation feature analysis and feature ranking. The root mean squared power and the zero crossing rate of the signals are shown to be the most discriminative among a set of candidate time domain features, which are subsequently used by a k-th nearest neighbor classifier, allowing for successful leakage detection at an affordable computational cost. The performance of the proposed scheme is evaluated using real measurements from oil refinery pipeline systems.