A reconstruction method of detonation wave surface based on convolutional neural network. (1st May 2022)

• Record Type: Journal Article
• Authors: Bian, Jing; Zhou, Lin; Yang, Pengfei; Teng, Honghui; Ng, Hoi Dick
• Is Part Of: Fuel. Volume 315(2022)
• Publication Date: 2022-05-01
• Subjects: Detonation waves -- Wave surface reconstruction -- Convolutional neural network -- Machine learning; Fuel -- Periodicals; Coal -- Periodicals; Coal; Fuel; Periodicals; 662.6
• DOI: 10.1016/j.fuel.2021.123068
• Languages: English
• ISSNs: 0016-2361

Abstract

Highlights: A deep learning method based on CNN is first applied in the study of detonation wave. Compared with traditional MLP method, this CNN approach performs much better. The well-trained CNN shows a certain generalization capability and robustness. Abstract: Detonation wave surface is composed of lead shock and reactive front, which are difficult to be measured simultaneously, so it is necessary to reconstruct the detonation surface. In this study, a reconstruction method is proposed for predicting lead shock from reactive front to obtain a full cellular detonation surface. The reconstruction uses a convolutional neural network (CNN) with the advantages of feature extraction and data dimensionality reduction, and the proposed method has been verified by data from numerical simulations in this work. The results indicate that this method performs much better than the traditional multi-layer perceptron (MLP), benefiting from the advanced architecture of CNN. Furthermore, effects of hyper-parameter choice have been tested, and the generalization capability of trained CNN for different activation-energy cases are also discussed.