High-fidelity turbulent flow field data are essential for understanding complex flow dynamics, optimizing thermal-fluid systems, and developing effective flow control strategies. However, complete flow fields are often unavailable in practical applications due to sparse measurements, limited sensing capabilities, and the storage burden associated with large-scale simulation data. Recovering missing flow fields while preserving turbulent structures and physical statistics remains challenging because conventional reconstruction approaches usually suffer from structural distortion and inaccurate statistical representations under highly incomplete observations. Therefore, a novel multi-source constrained conditional latent diffusion model (MSCLDM) is proposed to reconstruct incomplete flow fields. To ensure physical and structural integrity, a multi-source conditional guidance module is designed. This module mainly consists of a flow-condition encoder block, an energy-based guidance block, and a global information constraint block. A context-adaptive discrepancy (CAD) mechanism based on optimal transport gradient correction is introduced to improve structural continuity between the reconstructed and known regions and facilitate the recovery of fine-scale turbulent features. Moreover, to ensure data fidelity, a mask consistency mechanism is introduced to preserve the known regions throughout the diffusion process. Experiments conducted under various missing scenarios on three datasets comprising isotropic turbulence, channel flow, and cylinder wakes demonstrate that the proposed MSCLDM significantly outperforms current methods. Specifically, it reduces the mean squared error (MSE) by 4.1
Direct-forcing immersed boundary methods (IBMs) are widely used to enforce no-slip boundary conditions in incompressible flow solvers owing to their flexibility and ease of implementation. In this work we introduce a semi-discrete framework that encompasses both Lagrangian and Eulerian direct-forcing IBMs. Within this formulation, classical algorithms can be interpreted as approximate block-LU factorizations of the fully coupled monolithic system. This perspective reveals two distinct sources of error in the enforcement of the no-slip condition: a reciprocity error, associated with the non-exact adjoint relationship between interpolation and spreading operators, and a splitting error, arising from the pressure predictor used in projection-based time integration. The framework clarifies how the relative importance of these mechanisms depends on the interpolation kernel, the time step, and the physical configuration. Representative numerical experiments illustrate the behavior of the two error mechanisms and confirm the predictions of the analysis. In particular, a simple oscillating-balloon configuration is proposed as a diagnostic benchmark that directly exposes inconsistencies between boundary motion and the Eulerian mass balance. The analysis provides a unified interpretation of previously observed inaccuracies in direct-forcing IBMs and suggests possible directions for improvement, including the development of stable higher-order pressure predictors and improved constructions of interpolation–spreading operators.
We propose installing grid-type filters at compact axial-flow fan inlets to suppress flow instability and prevent degradation of fan performance in environments with an upstream obstacle. The effectiveness of this approach was investigated primarily through experiments, complemented by computational fluid dynamics analyses to investigate the flow-instability suppression mechanism. We examined the effects of different combinations of the main filter design factors such as mesh size (the area of a unit grid cell) and porosity (open-area ratio) on (i) the flow-rate range over which flow instability occurs and (ii) the performance and efficiency curves of the compact axial-flow fan in the presence of an upstream obstacle. As a result, installing grid-type filters at the fan inlet successfully suppressed the flow instability. An annular grid‑type filter was more effective in reducing pressure loss than a circular grid‑type filter. Furthermore, both single‑ and multi‑objective Bayesian optimization were applied to the experimental results, and we identified optimal specifications for the grid‑type filters, which were subsequently validated by experiments.
In hydrogen-fuelled scramjet combustors, rapid mixing and reliable flame-holding within a millisecond combustor residence time are mandatory but must not be accompanied by excessive stagnation-pressure loss. The effects of sliding-cavity orientation and the effects of curving the strut and integrating the wake at the supersonic combustor corners on the coupled shock–wake–shear–recirculation–reaction flow system in a dual-strut supersonic combustor is examined numerically. A 2D-steady RANS simulation with the SST (k-ω) turbulence model and the finite-rate hydrogen–air chemistry was performed for three principal arrangements: a cavity-free base configuration, a backward-slide cavity, and a front-slide cavity. As a supplemental comparison, a study of base and curved struts was undertaken in the front-slide cavity. After sensitivity analysis, a 0.15 mm mesh was chosen, and validation with the OH* chemiluminescence measurements and transverse temperature measurements was carried out, reproducing the flame length and width and spreading angle to within 4
We investigate the Lagrangian rotational and deformation dynamics of long flexible fibers in a turbulent channel flow at a shear Reynolds number equal to 300, using one-way coupled Euler-Lagrange direct numerical simulations. The fibers, modeled as chains of sub-Kolmogorov rod elements with prescribed bending stiffness, span more than one order of magnitude in length and are analyzed in three wall-normal slabs covering the near-wall, log-layer and bulk regions. We compute autocorrelations of the tumbling and spinning rates, power spectra of the fiber’s end-to-end vector, and waiting-time distributions of intense tumbling events. Spinning decorrelates on a fixed multiple of the local Kolmogorov time, independent of fiber length, whereas tumbling develops a memory that grows with fiber length by factors of two to five, reversing the ordering of the two timescales between the shortest and longest fibers. Flapping spectra separate into a high-frequency regime in which all fiber lengths carry the same energy in absolute terms and a low-frequency regime in which the energy increases with fiber length. Intense tumbling exhibits approximately exponential, weakly clustered waiting-time statistics whose rescaled form is universal across slabs, fiber lengths and bending rigidities, whereas the event rate depends on both fiber length and rigidity.
