Pipeline Leakage Detection: By Employing the Euler-Bernoulli Beam Using an Iterative Algorithm with A Machine Learning Approach

Abstract

Pipelines play a crucial role in transporting liquids, gases, and slurries across extensive distances in sectors such as oil and gas, petrochemicals, and water treatment. However, factors like aging infrastructure, joint or seal failures, corrosion, and pressure fluctuations make pipelines vulnerable to leakage. The present work aims towards a simulation based supervised learning approach for a real time leakage monitoring system in pipelines using MATLAB Simulink with an embedded classification learner tool to predict potential leaks in pipelines. Using MATLAB Simscape, an embedded physical model for a pipe system was developed for modelling fluid flow and related wall vibration in healthy and faulty conditions. The Euler Bernoulli beam theory is used to build a lumped parameter model to simulate the vibration response of a pipe due to leakage. Furthermore, the pressure fluctuation in the pipe due to turbulent flow is introduced through a Von Karman spectrum based model. The model was coupled with an accelerometer to acquire vibration data from the pipe wall near the leak. This data is input for the leak detection mechanism. The time series data of vibrations were collected, pre processed, and further utilized to train the Support Vector Machine (SVM) and Naive Bayes machine learning models to classify and predict instances of leakage and the magnitude of the leakage. The results obtained from the SVM model and the Naïve Bayes indicate an accuracy rate of 85% and 79%, respectively. Hence, the results from these two methods show good agreement.