Explainable Artificial Intelligence Modelling of Ignition Delay Times of Pure Ammonia and Hydrogen-doped Ammonia

This research focuses on developing an easy-to-use mathematical relation that can be employed in predicting the ignition delay times of pure ammonia and ammonia doped with different fractions of hydrogen. An artificial neural network model trained on 31332 synthetic data was developed, statistically evaluated and validated against experimental results from shocktube experiments with an accuracy of R-squared 0.99. the developed model produces results in 420ns, a 98% reduction in the time required by experiments. In the attached data, the first sheet is the generated synthetic data used in ANN modelling, the first four columns are the inputs and the fifth column is the output expected(IDT). Columns K to O are the normalised values of columns A to E. In the second sheet, the mathematical relations for the 50 neurons are shown where the final predicted IDT value is shown in cell B5.