A note on the improvement of Artificial Neural Network models for detonation cell size and critical tube diameter prediction
In this note we present strategies to improve a deep Artificial Neural Network (ANN) to predict the dynamic parameters of gaseous detonations in hydrogen- and other hydrocarbon-based mixtures. These new strategies involve using only non-dimensional features for the model, which have been created usi...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Nuclear Engineering and Technology |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1738573325003924 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In this note we present strategies to improve a deep Artificial Neural Network (ANN) to predict the dynamic parameters of gaseous detonations in hydrogen- and other hydrocarbon-based mixtures. These new strategies involve using only non-dimensional features for the model, which have been created using thermochemical and chemical kinetic parameters from the steady reaction zone structure commonly used in detonation studies, as well as a non-dimensional target, obtained by dividing the experimental cell size with the induction length ΔI. In addition, the ANN model's structure has been supplemented with dropout layers, thus improving the training process and also leading to a better determination of the model's uncertainty. Apart from predicting the detonation cell size, this updated model creation approach is implemented to the critical tube problem, combining thermochemical and kinetic parameters with experimental data to create an accurate model that predicts the critical tube diameter DC. The optimal structure and combination of features for the ANN are thoroughly assessed. The source codes of the ANN models are readily available on GitHub. |
|---|---|
| ISSN: | 1738-5733 |