DEEP LEARNING METHOD FOR FLOW FEATURE RECOGNITION BASED ON DIMENSIONLESS TIME HISTORY

The characteristics of the flow field influence the flow-induced vibration state of structures, it thusly has a crucial significance to study the flow features around structures. However, in the case of the flow field with medium and high Reynolds numbers, the wake is highly complex, and it is not easy to extract and recognize complex features through traditional mathematical and physical methods. This paper proposes a deep learning method that uses the dimensionless time history of the flow variables to identify the flow features. The method eliminates the influence of different incoming flow speeds and only uses the time-varying features of the dimensionless time history for feature recognition, which improves the scope of the method. Two different flow time-history deep learning models are used to extract and identify the wakes of three types of prisms. Comparison results proved that unified time history maintains the critical features of wake caused by objects of different shapes. Furthermore, the unified time history can be used for the feature extraction of the flow field and improve the accuracy of the flow field feature extraction by the model. It is a feasible new method for flow field feature extraction.