Extracting Quantitative Streamline Information from Surface Flow Visualization Images in a Linear Cascade using Convolutional Neural Networks

Abstract

Surface flow visualization (SFV), specifically surface oil flow visualization, is an experimental technique that involves coating the surface with a mixture of oils and dyes before applying the flow to the subject. While investigating the surface flow, the surface topology must be analysed to determine the flow field near the surface. For this, numerous flow visualization and image processing techniques have been proposed, showing good performance. Nonetheless, their accuracy is largely contingent on human expertise, and the overall processing cost is elevated because they necessitate the trial-and-error optimization of thresholding parameters, which are not applicable universally across all experimental conditions. Convolutional Neural Networks (CNN) are deep learning models designed primarily for tasks involving grid-like data, particularly image and video analysis. Inspired by the outstanding feature extraction performance of deep neural networks, in this work, we trained a CNN-based model to develop an automated streamline detection and flow field reconstruction tool which works well on surface oil flow visualization images. The accuracy of streamline detection can be customized through the use of a threshold function. The predictive outcomes of the flow field and the distribution of shear pressure will be compared to the results obtained through Computational Fluid Dynamics (CFD) in the same case.