The effect of initial conditions on the turbulent structures of various scales in a near-wake has been investigated for two wake generators, i.e., a circular cylinder and screen of 50% solidity, of the same characteristic dimension. Velocity data were obtained at $x∕h$ ($x$ and $h$ are the streamwise distance downstream and the height of the wake generators, respectively) $=20$ using 16 X wires, eight aligned in the plane of mean shear and eight in the plane parallel to both the cylinder axis and the flow direction. A wavelet multiresolution technique was used to decompose the velocity data into a number of wavelet components based on the central frequencies, which correspond to the scales of turbulent structures. It was found that the behaviors of intermediate- as well as large-scale structures depend on the initial conditions. In the circular cylinder wake, the large- and intermediate-scale structures make the largest (and comparable) contribution to Reynolds stresses, but in the screen wake there is a significant difference in the contribution between the two types of structures. The two-dimensionality of the wavelet component is also examined.

A new pneumatic conveying system that applies soft fins mounted vertically on a center plane of pipe in the inlet of the gas-particle mixture is developed to reduce power consumption and conveying velocity. The effect of different fin’s lengths on a horizontal pneumatic conveying is experimentally studied in terms of the pressure drop, conveying velocity, power consumption, particle flow pattern, and additional pressure drop. The test pipeline consisted of a horizontal smooth acrylic tube with an inside diameter of 80 mm and a length of about 5 m. Two kinds of polyethylene particles with diameters of 2.3 mm and 3.3 mm are used as conveying materials. The superficial air velocity is varied from 10 to 17 m/s, and the solid mass flow rate is from 0.20 to 0.45 kg/s. Compared with conventional pneumatic conveying, the pressure drop, minimum and critical velocities, power consumption, and additional pressure drop can be reduced by using soft fins in a lower air velocity range, and the efficiency of fins becomes more evident when increasing the length of the fins or touching particles stream by the long fins. The maximum reduction rates of the minimum velocity and power consumption by using soft fins are about 20% and 31.5%, respectively. The particle concentrations of using fins are lower than those of non-fin near the bottom of the pipe and are higher than those of non-fin in the upper part of the pipe in the acceleration region. Based on analyzing the frequency features of the fin’s oscillation, the Strouhal number of more efficient fins is about St ≈ 0.75 in the air velocity range of lower than 13 m/s.

To address the challenges of limited labeled data and insufficient global feature extraction in flow field prediction, this paper proposes a modeling approach that combines self-supervised learning and Graph Transformer. The self-supervised learning module leverages feature reconstruction tasks and contrastive learning tasks to fully exploit the latent information in unlabeled data, thereby enhancing the joint modeling capability for local and global features. The Graph Transformer incorporates a self-attention mechanism, enabling effective modeling of long-range dependencies and multiscale features in complex flow fields. Experimental results demonstrate that, under 100% labeled data conditions, the proposed method reduces the root mean squared error achieved by graph convolutional network and a multiscale graph neural network model on the cylinder flow and airfoil flow datasets from 0.970 and 0.561 to 0.616 and 0.305, achieving significant accuracy improvements of 36.5% and 45.6%, respectively. Under 50% labeled data conditions, the method still exhibits outstanding robustness, with RMSEs of 0.792 and 0.390, respectively. The ablation studies reveal that the feature reconstruction and contrastive learning tasks exhibit strong complementarity, achieving optimal performance when jointly employed. Furthermore, the self-attention mechanism significantly enhances the modeling of global features, demonstrating its effectiveness in capturing complex dependencies. The proposed method demonstrates superior prediction accuracy and robustness under limited labeled data and complex flow field conditions, providing an efficient solution for flow field modeling with broad application potential.