Experimental study of NH3/H2/n-heptane combustion and reduced mechanism optimization via a CNN-augmented neural network and the L-SHADE algorithm

Adding H 2 and n-heptane to ammonia improves flame speed and autoignition reactivity, respectively. Using n-heptane as a pilot fuel to ignite NH 3 /H 2 mixtures has emerged as a promising strategy to reduce carbon emissions in engine applications. In this work, ignition delay times (IDT) of NH 3 /H 2 /n-heptane ternary mixtures with n-heptane molar fractions from 0 to 0.12 and NH 3 /H 2 ratios from 90/10 to 0/100 were measured in a rapid compression machine (RCM) at compressed pressures of 15 and 30 bar, compressed temperatures of 650 to 1050 K, and equivalence ratios ( ϕ ) of 0.5, 1.0, and 2.0. Results show that the introduction of n-heptane significantly enhances reactivity and dominates the ignition behavior, thereby diminishing the influence of H 2 on IDTs compared to binary NH 3 /H 2 mixtures. Furthermore, a novel convolutional neural network (CNN)-augmented hybrid model is proposed to predict IDTs by introducing compression-related features in RCM experiments. These features, combined with mixture composition, thermodynamic conditions, and reaction-rate multipliers, serve as inputs for an integrated artificial neural network (ANN). The model accurately captures complex input–output relationships and yields robust predictions. By coupling this surrogate model with the advanced Success-History based Adaptive Differential Evolution with Linear Population Size Reduction (L-SHADE) optimization algorithm and incorporating a variety of experimental data, a robust mechanism optimization framework is developed. The final optimized reduced mechanism, validated against extensive in-house and literature data, demonstrates strong predictive capability and compactness, making it suitable for engine simulations applications.