Optimization of Auto-ignition Time for a Skeletal Mechanism in a Piloted Jet Flame Using Artificial Neural Networks

Abstract Combustion modeling is a process that must strike a balance between accuracy and efficiency. Detailed mechanisms, such as GRI-3.0, offer a high level of fidelity, but are expensive and require skeletal mechanisms that reduce complexity, which can compromise accuracy. The present work addresses this challenge by optimizing a skeletal mechanism for methane combustion using an artificial neural networks (ANNs) approach. The skeletal mechanism selected for optimization comprises 22 species and 30 reactions, named 30R, and is derived from the detailed mechanism GRI-3.0. The kinetic parameters of mechanism 30R were optimized to generate a new skeletal mechanism, named 30RANN, which exhibited enhanced accuracy for combustion simulations, particularly in predicting auto-ignition delay times. The skeletal mechanism 30RANN was validated against both the detailed mechanism GRI-3.0 and the skeletal mechanism 30R in combustion simulations of a jet flame using Chemkin-Pro and openfoam. The results demonstrated that the mechanism 30RANN exhibited a superior capacity to replicate the predictions of the mechanism GRI-3.0 in comparison to mechanism 30R, resulting in a reduction of error from 26.4% to 3.8%. This enhancement highlights the importance of optimizing skeletal mechanisms to enhance the predictability of combustion properties. Furthermore, the results highlight the crucial role of neural networks in augmenting the predictive and memory capabilities of the mechanism, thereby demonstrating their potential in optimizing skeletal mechanisms to achieve improved accuracy and efficiency in combustion applications.