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

Three-dimensional particle reconstruction with limited two-dimensional projections is an under-determined inverse problem that the exact solution is often difficult to be obtained.In general,approximate solutions can be obtained by iterative optimization methods.In the current work,a practical particle reconstruction method based on a convolutional neural network(CNN)with geometry-informed features is proposed.The proposed technique can refine the particle reconstruction from a very coarse initial guess of particle distribution that is generated by any traditional algebraic reconstruction technique(ART)based methods.Compared with available ART-based algorithms,the novel technique makes significant improvements in terms of reconstruction quality,robustness to noise,and at least an order of magnitude faster in the offline stage.

3D Reconstruction Technique for Tomographic PIV

Tomographic particle image velocimetry(Tomo-PIV) is a state-of-the-art experimental technique based on a method of optical tomography to achieve the three-dimensional(3D) reconstruction for threedimensional three-component(3D-3C) flow velocity measurements. 3D reconstruction for Tomo-PIV is carried out herein. Meanwhile, a 3D simplified tomographic reconstruction model reduced from a 3D volume light intensity field with 2D projection images into a 2D Tomo-slice plane with 1D projecting lines, i.e., simplifying this 3D reconstruction into a problem of 2D Tomo-slice plane reconstruction, is applied thereafter. Two kinds of the most well-known algebraic reconstruction techniques, algebraic reconstruction technique(ART) and multiple algebraic reconstruction technique(MART), are compared as well. The principles of the two reconstruction algorithms are discussed in detail, which has been performed by a series of simulation images, yielding the corresponding reconstruction images that show different features between the ART and MART algorithm, and then their advantages and disadvantages are discussed. Further discussions are made for the standard particle image reconstruction when the background noise of the pre-initial particle image has been removed. Results show that the particle image reconstruction has been greatly improved. The MART algorithm is much better than the ART. Furthermore, the computational analyses of two parameters(the particle density and the number of cameras), are performed to study their effects on the reconstruction. Lastly, the 3D volume particle field is reconstructed by using the improved algorithm based on the simplified 3D tomographic reconstruction model, which proves that the algorithm simplification is feasible and it can be applied to the reconstruction of 3D volume particle field in a Tomo-PIV system.

Dual-basis reconstruction techniques for tomographic PIV

As an inverse problem, particle reconstruction in tomographic particle image velocimetry attempts to solve a large-scale underdetermined linear system using an optimization technique. The most popular approach, the multiplicative algebraic reconstruction technique(MART), uses entropy as an objective function in the optimization. All available MART-based methods are focused on improving the efficiency and accuracy of particle reconstruction. However, those methods do not perform very well on dealing with ghost particles in highly seeded measurements. In this report, a new technique called dual-basis pursuit(DBP), which is based on the basis pursuit technique, is proposed for tomographic particle reconstruction. A template basis is introduced as a priori knowledge of a particle intensity distribution combined with a correcting basis to enable a full span of the solution space of the underdetermined linear system. A numerical assessment test with 2D synthetic images indicated that the DBP technique is superior to MART method, can completely recover a particle field when the number of particles per pixel(ppp) is less than 0.15, and can maintain a quality factor Q of above 0.8 for ppp up to 0.30. Unfortunately, the DBP method is difficult to utilize in 3D applications due to the cost of its excessive memory usage. Therefore, a dual-basis MART was designed that performed better than the traditional MART and can potentially be utilized for 3D applications.