A Pipeline Leak Classification and Location Estimation System With Convolutional Neural Networks

Abstract

An accurate pipeline leak classification and location estimation method can help to control and reduce the damage to the environment when spills happen. Some of the current research on this topic rely on the direct analysis of target frequencies from the monitoring of sensors. However, this assumes that the frequencies may be known before hand and such analysis can be very cumbersome. In this article, we propose convolutional neural network-based classification and location estimation methods, which use raw data instead of prefiltered (or preconditioned) information. The design approach is fully described and the network structure is discussed. Finally, analysis of experimental results validate the proposed network demonstrating that the classification and location estimation can be done with good accuracy.