Using neural networks to model detonations in aluminum-air mixtures

摘要

This study investigates combustion modeling in AI-air mixtures using a hybrid model that combines both diffusion-limited and kinetics-limited regimes of burning. Neural networks are trained on the combustion mass-transfer rates as a function of five independent nondimensional numbers, which are obtained from three training cases. Specifically, particle sizes in the range 1-5 μm, and fuel lean, stoichiometric, and rich configurations are considered, to demonstrate the wide range of applications of the neural-network-based combustion modeling. Four different neural-network architectures, each with two hidden layers, are trained, and one of them is used to make predictions on four test cases not part of the training set. The neural network performs well, predicting the right detonation speed, and pressure and velocity behind the detonation front. This study demonstrates that neural networks can be a useful tool to model the combustion of A1 particles in detonation applications.