Deep Learning Based Leak Detection of Oil Transmission

Leakage of oil pipeline pose significant challenges in the area of petroleum transport and infrastructure management, with far-reaching implications encompassing economic, environmental, and public safety concerns. While machine learning has developed quite rapidly in recent years and has gained prominence in various sectors, its adoption in petroleum engineering domain remains relatively rare. This paper explores the application of machine learning, specifically Artificial Neural Networks (ANN) method, in predicting the location of oil leaks in a pipeline, given the outlet pressure of the pipe and the expected leak rate. Due to very low number of leak occasion in the real practical case, it is quite challenging to obtain enough balanced leak data to train the learning model properly. To manage that issue, we generate the data manually by simulating the leak process using Beggs-Brill pressure drop correlation in various combination of different experimental cases. The generated data incorporates outlet pressure and estimated leak rates, to serve as input features for the ANN model. We fine-tune the ANN architecture to optimize model performance and analyzes its learning capabilities. This paper underscores the potential of ANN technology in addressing the complex issue of oil pipeline leaks and contributes to the ongoing efforts to enhance pipeline safety.