A deep learning-based approach for flow field prediction in a dual-mode combustor

Abstract

Accurate acquisition of the distribution of flow parameters inside the supersonic combustor is of great significance for hypersonic flight control. It is an interesting attempt to introduce a data-driven model to a supersonic combustor for flow field prediction. This paper proposes a novel method for predicting the flow field in a dual-mode combustor. A flow field prediction convolutional neural network with multiple branches is built. Numerical investigations for a strut variable geometry combustor have been conducted to obtain flow field data for training the network as a flow field prediction model. Rich flow field data are obtained by changing the equivalent ratio, incoming flow condition and geometry of the supersonic combustor. The Mach number distribution can be obtained from the trained flow field prediction model using the combustor wall pressure as input with high accuracy. The accuracy of flow field prediction is discussed in several aspects. Further, the combustion mode detection is implemented on the prediction flow field.

Keywords

• Flow field prediction
• Deep learning (DL)
• Convolutional neural networks (CNNs)
• Data-driven model
• Dual-mode combustor
• Variable geometry combustor