Research on time sequence prediction of the flow field structure of supersonic cascade channels in wide range based on artificial neural network

Abstract

Accurate and comprehensive flow field prediction is indispensable for promptly monitoring the flow state of supersonic cascades. This paper proposes a time sequence prediction architecture based on the full convolutional neural network (FCNN) to predict the future flow parameters of supersonic cascades based on the wall pressure at the previous moment. Considering the complicated spatial-temporal characteristics of the time sequence prediction of flow field structures, FCNN embeds the convolution into the long short-term memory (LSTM) and replaces the fully connected layer with a convolution in the output layer. Wind tunnel experiments with different flap rotation rates were performed to obtain the dataset required for model training and verification. For pressure-to-schlieren time sequence prediction, FCNN takes the cascade wall pressure at the previous moment as input and the future schlieren of the flow field structure captured by a high-speed camera as its output. The experimental results show that FCNN can accurately predict the position of the leading edge and that the maximum relative error is less than 4.4%. Moreover, for pressure-to-pressure time sequence prediction, the bidirectional LSTM (BiLSTM) was used to predict the wall pressure of the cascade channel. Results show that the BiLSTM can accurately capture the nonlinear characteristics of the wall pressure changing over time. Thus, the results of this study lay a solid foundation for the subsequent determination of the stability margin of flow fields.