Detection of Hydrogen Leakage Using Different Machine Learning Techniques

Abstract

When employing pure hydrogen, its leakage poses a serious safety risk since it can cause fire or explode if it comes into contact with the air. In this study, hydrogen leakage in a form of a buoyant jet is investigated using machine learning approaches. As the experiments used to explore hydrogen leaks are extremely dangerous, and there is a limitation of data, we instead construct an artificial dataset using a traditional numerical model. The dataset was produced using a combined empirical-analytical-numerical model. Investigations into dataset preparation, feature significance, correlation, and hyperparameter adjustment are conducted. Artificial neural networks, random forests, gradient boosting regression, and decision trees are the machine-learning approaches that have been used to forecast the distribution of hydrogen leaks in the atmosphere. Different error metrics and R2 correlation have been used to assess the prediction accuracy. The RF method was found to be the most effective approach for forecasting the dispersion of hydrogen leaking into the air.