Wake mode identification of rotating triangle with machine learning approaches

Abstract

The moving body usually leaves different forms of wake trails in fluids, and these wake fields contain physical information such as the state and geometry of the moving body. Recognizing and extracting these information from the wake can provide new insights for non-acoustic detection technology. Traditional methods do not easily extract the flow state, geometry, and other information directly from the wake structure. This work mainly uses convolutional neural network algorithms for intelligent recognition of the wake types of rotating triangles. Based on the flow field visualization technology of the soap film tunnel, the wake types of the flow around a structure controlled by external excitation of sinusoidal rotation are studied. The winding characteristics of the rotating triangle and the variation rule of the wake with control parameters are analyzed. At last, the recognition rate of the wake types on the test set is above 90%. The recognition rates of the experimental data not involved in the training conditions are all above 80%, demonstrating the generalizability of the model. This method provides a reference for further utilizing artificial intelligence in extracting physical information from wakes, playing a crucial role in advancing wake detection technology.