PROGRESS OF CONVOLUTION NEURAL NETWORKS IN FLOW FIELD RECONSTRUCTION

In recent years, with the rapid development of deep learning in image processing, speech recognition, automatic driving, natural language processing, and other fields, this technology is also more and more widely used to process fluid mechanics direction with the characteristics of complex non-linearity, high latitude and a large amount of data. Traditional methods can not effectively deal with these huge data. Due to its powerful function fitting ability, deep learning can mine useful information from a large amount of data. At present, the deep learning technology of fluid mechanics has some preliminary research results, it has important engineering value in flow information feature extraction, multi-sensor data information fusion and intelligent reconstruction of flow field, and its potential of application has been gradually confirmed. How to use the data obtained from ground wind tunnel test, numerical simulation and flight test to carry out in-depth mining, fast intelligent perception and reconstruction of flow field can provide important guidance for active flow control. Starting from different types of network structures of deep learning, this paper discusses the research progress of convolutional neural network in flow field reconstruction. Firstly, In this paper, we introduce some basic concepts and basic network structure of convolutional neural network, and then we briefly introduce the basic structure and theory of flow field super-resolution reconstruction network, end-to-end mapping network and short-term memory network (LSTM), a series of research progress and achievements of their improved forms in the field of flow field reconstruction are summarized in detail. Finally, we summarize the article and discuss the challenges and prospects of deep learning technology of flow field reconstruction.