Precision Pipeline Leakage Detection through Deep Learning and Infrared Sensing

Abstract

Leaks of chemicals and gas pipes left unrepaired or lost in distribution networks can result in hazards with significant harm to the environment, people, and property, causing economic losses amounting to billions of dollars annually. Conventional inspection techniques, which depend on sporadic inspection schedules, routine checks, and localized sensors, are inadequate since they frequently miss tiny breaches because of their low sensitivity. This paper presents an innovative approach that leverages machine learning algorithms combined with infrared thermal imaging to enhance early detection of minor leaks. We developed a dataset comprising 1035 thermal images of leak scenarios and 1036 images of no-leak scenarios, captured using a thermal camera in a controlled setup. The final dataset has 1035 images of leakages and 1036 images of no-leakage scenarios. Our experiments with a Multi-Layer Perceptron model achieved the highest accuracy of 95.41%. To augment the system’s detection capabilities, we integrated it with traditional flow sensor-based methods. The resulting portable device merges thermal imaging and deep neural network analysis, offering a powerful tool for detecting leaks in both household and industrial settings, as well as in agricultural applications.