Detonation cell size prediction based on artificial neural networks with chemical kinetics and thermodynamic parameters

Abstract

In this paper, we develop a series of Artificial Neural Networks (ANN) using different chemical kinetic and thermodynamic input parameters to predict detonation cell sizes. The feedforward neural networks are trained and validated using available experimental data from the Caltech detonation database covering a wide variety of gaseous combustible mixtures at different initial conditions. For each combination of input parameters, a multiple-stage process is followed, which is described in detail, to first determine the best hyperparameters of the ANN (hidden layers, nodes per layer, etc.) and secondly to establish through a fitting process the optimal parameters for each specific network. The performance of the artificial neural networks with different input features is assessed using data from the same source, but that is kept independent and separate from the training and validation process of the ANN. It is found that ANN with three features can provide an accurate estimation of detonation cell size, while increasing the number of features does not improve the accuracy of the ANN. It is also found that the input parameters with the best performance relate indirectly to the stability parameter χ.

Keywords

• Machine learning
• Artificial neural network
• Gaseous detonation
• Cell size
• Chemical kinetics
• Stability parameter