Detection and Estimation of Natural Gas Leakage Using UAV by Machine Learning Algorithms

Abstract

Presently, unmanned aerial vehicles (UAVs) have a broad spectrum of applications, in which regular monitoring of critical infrastructure assets is an important risk mitigation usage. This paper presents a novel approach for modeling leakage from natural gas pipelines using machine learning algorithms operating on the gas leakage detection (Methane) data collected using an UAV having a gas sensor and a light detecting and ranging (LiDAR) payload. The detection system has a lightweight design along with a small form factor to ensure compatibility with most autonomous mobile platforms like UAVs or wheeled robots. Two experiments were conducted to collect the gas leakage detection data at various loitering heights in real atmospheric conditions before applying the estimation algorithms. The first experiment measured natural gas leakage data in varying loitering radiuses and altitudes for leakage pressures around one pound per square inch (PSI), while the second experiment did the same for leakage pressures around two PSI. Real atmospheric conditions were incorporated along with additional aspects of propeller down-wash and environmental air movements that can influence leakage detection. Two estimation algorithms, viz., reduced support vector machine (RSVM) and artificial neural network (ANN) were applied to the leakage data collected from the UAV experiments. It was found that the ANN approach resulted in more faster and accurate detection than RSVM that heavily depended on the kernel function for its performance. The efficacy of leakage detection of natural gas using a UAV payload was demonstrated, which is a faster and cost-effective alternative to the manual inspection process.