Simulation assisted leak detection in pressurized systems using machine learning

In this paper, we utilize Computational Fluid Dynamics (CFD) generated data to train a Recurrent Neural Network (RNN) for detecting leaks in pressurized fluid distribution systems. The obtained results support the validity of implementing Machine Learning techniques in approximating active leak locations in a single pipe setup. This paper also discusses the validity of implementing these techniques for implementation on a fluid distribution network.

Results obtained utilizing the RNN model show adaptive behavior with the system’s consistent response to different configurations of pipes and boundary conditions. The predictions for the leak localities are more accurate and more economically feasible than those obtained with currently used methods.