Exploration of the Ignition Delay Time of RP-3 Fuel using the Gradient Descent Optimization Algorithm in a Machine Learning Framework

Ignition delay time (IDT) is a fundamental parameter in fuel combustion, with accurate prediction across temperature ranges remaining a significant research challenge. RP-3, a widely used aviation kerosene, plays a key role in refining chemical kinetics and optimizing aeroengine combustion design. This study presents a BP neural network prediction framework, enhanced through random selection and gradient descent optimization, to achieve high-precision IDT prediction under various conditions.The proposed model features five inputs and one output, with hidden layer structure optimized using a random selection strategy to improve generalization and robustness. Among five gradient descent algorithms evaluated, the conjugate gradient method paired with a [21, 17, 19] three-hidden-layer architecture achieved the best performance (R² = 0.99705, MAPE = 1.2%, MAE = 0.0287, RMSE = 0.0359). The model maintained high accuracy across equivalence ratios (φ = 1.0–2.0) and pressures (10–20 bar).Sensitivity and importance analysis revealed that temperature and diluent mole fraction significantly influence IDT, especially in the lowand high-temperature and NTC regions. Overall, the proposed method demonstrates strong generalization, robustness, and application potential, offering a data-driven tool for modeling combustion characteristics of complex aviation fuels.