Multiphase Pipeline Leak Detection Based on Fuzzy Classification

摘要

this study investigates a new approach for computational pipeline monitoring. Flow rate, pressure and temperature of the pipe at both ends are measured. Time domain features are extracted from windowed signals. They form separable patterns in the feature space for the normal and leak conditions. Fuzzy classifier is trained with data from a medium leak in the middle of the multiphase pipe. The trained classifier is used to detect leaks with different positions and sizes. Also the ability of the classifier to distinguish operational changes from leakages is evaluated. Results of fuzzy classifier are compared with a neural network classifier and a linear classifier. Results show that this approach obtains good results for pipeline leak detection.