Reconstruction of the fluid velocity field measured by SPIV via artificial neural networks

One of the important and main tools for studying flow and evaluation criteria of other methods such as numerical methods is experimental data of the fluid mechanics. Thus, the appropriate quality of the data, that are measured in the laboratory, is important. One of the important information of any flow is fluid velocity field, that is measured by different instruments. One of those tools is the SPIV tool. This tool provides sheet information from the flow velocity components. Generally, the data extracted from this tool will have big errors in some points of the velocity field, for various reasons and laboratory conditions, and the values obtained in these points will be eliminated, which are called gappy points. Therefore, in order to reconstruct the velocity field at these gapps, methods are needed. In this regard, in the present study, we will use artificial neural networks, such as MLP and CNN. The optimization of the number of neurons in the MLP network was performed by the mean error of the test data and the matching of the images. The final error was obtained for each of the methods, which due to the errors and accommodation between the reconstructed velocity field and experimental data, it was obtained that for both velocity components, the CNN neural network had the best performance.