Filtering enhanced tomographic PIV reconstruction based on deep neural networks

Abstract

Tomographic particle image velocimetry (Tomo‐PIV) has been successfully applied in measuring three‐dimensional (3D) flow field in recent years. Such technology highly relies on the reconstruction technique which provides the spatial particle distribution by using images from multiple cameras at different viewing angles. As the most popular reconstruction method, the multiplicative algebraic reconstruction technique (MART) has advantages in high computational speed and high accuracy for low particle seeding reconstruction. However, the accuracy is not satisfactory in the case of dense particle distributions to be reconstructed. To overcome this problem, a symmetric encode–decoder fully convolutional network is proposed in this paper to improve the reconstruction quality of MART. The input of the neural network is the particle field reconstructed by the MART approach, while the output is the regenerated image with the same resolution. Numerical evaluations indicate that those blurred or irregular particles can be significantly refined by the trained neural network. Most of the ghost particles can also be removed by this filtering method. The reconstruction accuracy can be improved by more than 10% without increasing the computational cost. Experimental evaluations indicate that the trained neural network can also provide similar satisfactory reconstruction and improved velocity fields.

Particle reconstruction of volumetric particle image velocimetry with the strategy of machine learning

“…For example, [23] applied convolutional neural networks (CNN) to PIV and achieved similar effects of traditional cross-correlation algorithms. Liang et al [24] used CNN as a filtering step after several MART iterations for particle reconstruction. However, at the moment, most existing works on applying machine learning to PIV are two dimensional while an investigation of applying machine learning on particle reconstruction, as a fully three-dimensional application, is still lacking.…”

Solving multiple travelling salesman problem through deep convolutional neural network

“…After transferring the city coordinates into an image representation, U-net is trained to establish the mapping from the input images to the label images. The U-net structure includes encoder and decoder paths [27]. The encoder path consists of five units.…”

Machine learning for flow field measurements: a perspective

“…This network is trained to minimize the backprojection error, and it can outperform two-frame reconstruction based on a spatial-filtered version [59] of the multiplicative algebraic reconstruction technique (MART, [60]). Liang et al [61] indicated that the MART tomographic reconstruction can be filtered using an encoder-decoder CNN to suppress ghost particles and regularize the reconstructed particle shape.…”