Innovative input-driven ANN approach for the prediction of hydrogen flame length.

Hydrogen's role as a clean energy source necessitates accurate modeling of its combustion properties, particularly flame length, to ensure safety and efficiency in industrial applications. Traditional methods that rely solely on Froude number correlations often fail to capture the complexities of hydrogen jet flames, especially under-expanded conditions. This study overcomes these limitations by employing artificial neural networks (ANNs) with novel input parameters to predict hydrogen flame lengths. Two ANN models were developed: the Multi-Layer Perceptron (MLP) and the Radial Basis Function (RBF) network. The MLP model achieved a regression coefficient (R2) of 0.990, while the RBF model demonstrated superior performance with an R2 of 0.996. Sensitivity analysis identified the Froude number as the most critical parameter influencing flame length. Experimental data from various nozzle diameters and flow rates were used for model training and validation. The application of Response Surface Methodology (RSM) further enhanced the model's predictive accuracy and addressed the limitations of single-parameter approaches. [Display omitted] • ANN models predict hydrogen flame length using dimensionless parameters. • RBF networks outperform MLP in accuracy and generalizability. • Sensitivity analysis highlights the Froude number's critical influence. • RSM improves model generalizability for hydrogen flame predictions. [ABSTRACT FROM AUTHOR]