- Explosion characteristics of hydrogen mixtures have been extensively investigated at different conditions. Due to the intensiveness of the explosion characteristics experimental determination, empirical and semi-empirical models are commonly used to predict explosion limits in dependance of conditions: temperature, pressure, and mixture composition. However, unevenly distributed, and limited empirical data and the complex non-linear relationship of these explosion characteristics’ present significant challenges to empirical explosion limits prediction methods under various mixture conditions. Moreover, some empirical models and semi-empirical models are not comprehensive and limited in scope of application. To address these issues, the present study adapts a machine learning approach for improving the hydrogen mixtures explosion characteristics prediction at different conditions, offering a fast, flexible, and comprehensive accurate prediction approach. A Multi-Layer Perceptron modelExplosion characteristics of hydrogen mixtures have been extensively investigated at different conditions. Due to the intensiveness of the explosion characteristics experimental determination, empirical and semi-empirical models are commonly used to predict explosion limits in dependance of conditions: temperature, pressure, and mixture composition. However, unevenly distributed, and limited empirical data and the complex non-linear relationship of these explosion characteristics’ present significant challenges to empirical explosion limits prediction methods under various mixture conditions. Moreover, some empirical models and semi-empirical models are not comprehensive and limited in scope of application. To address these issues, the present study adapts a machine learning approach for improving the hydrogen mixtures explosion characteristics prediction at different conditions, offering a fast, flexible, and comprehensive accurate prediction approach. A Multi-Layer Perceptron model was trained, validated, and tested using key input features such as flammability state, initial mixture temperature, inert gas concentration, adiabatic flame temperature, and Lewis numbers. Data augmentation techniques were conducted to supplement and improve the predictive capability of the model. The model’s performance was compared with a separate experimental dataset. This machine learning approach offers a cost-effective and robust alternative to existing empirical explosion limit prediction method, thus also reducing the experimental effort for explosion limits determination.…

