JK-FLOW Japan-Korea Fluid Mechanics Online Workshop

Speaker: Mr. Tianyang Fang (Graduate student, The University of Tokyo)

Abstract: Pulsed V-shaped fins are promising enhancement structures for compact heat exchangers, but their performance is limited by the trade-off between heat transfer and pressure loss. This study combined numerical simulations and Bayesian optimization to optimize pulsed V-shaped fin geometries under laminar flow conditions. Finite-volume simulations were performed in OpenFOAM at Reynolds number of 875 to evaluate the effects of wave height and oblique angle on secondary flow, heat transfer, and flow resistance. Larger wave heights and oblique angles enhanced boundary-layer disruption and heat transfer, but also increased pressure loss through stronger flow separation. Compared with conventional sinusoidal V-shaped protrusions, pulsed fins improved low-performance valley regions by promoting earlier flow reattachment. A periodic simulation framework was validated against theoretical solutions, reducing the mesh requirement to 1.6% of the full-domain model while maintaining trend between heat transfer performance and geometry. Bayesian optimization identified an optimal geometry near A/H = 0.1 and γ = 100°, balancing heat-transfer enhancement and friction penalties. The proposed framework enables efficient design of high-performance pulsed fin heat exchangers.

Speaker: Mr. Yaedalm Son (Graduate student, Korea University)

Abstract: Solar-driven evaporation provides a promising route for decentralized freshwater production and brine management, but its performance is often limited by inadequate water supply, heat losses, and salt fouling arising from insufficient understanding of geometry–transport–performance relationships. Here, 3D-printed multicellular solar evaporators composed of systematically varied lattice unit cells are developed, identifying liquid–solid contact perimeter, porosity and thermal interface area as key geometric parameters governing capillary water delivery and heat transfer. Guided by these insights, an optimized FBCC-+_5 evaporator achieves an evaporation rate of 6.9 kg m-2 h-1 while sustaining zero-liquid-discharge operation. In addition, a hybrid multicellular design combining unit cells with different evaporation rates creates controlled evaporation gradients that localize salt crystallization, enabling stable operation for 5.5 days in 20 wt% NaCl. Outdoor desalination yields up to 48 kg m-2 day-1 of freshwater. This work reveals geometry–transport–performance relationships in 3D-printed multicellular solar evaporators and establishes unit-cell and macro-scale geometry as programmable design parameters for scalable and durable solar desalination.

Speaker: Mr. Jiwon Kang (Graduate student, Inha University)

Abstract: Adaptive mesh refinement (AMR) reduces the cost of unsteady flow simulations by locally increasing resolution where flow structures evolve most actively. In bluff-body wake flows, however, shear layers and vortex-shedding structures move in time and are difficult to capture using hand-defined indicators with fixed thresholds. This study introduces a wake-focused deep-learning AMR framework based on heteroscedastic temporal-residual learning. A dual-head U-Net is trained on circular-cylinder wake data using non-dimensionalized velocity and pressure fields. The network predicts both the temporal residual and a spatially varying uncertainty map that identifies where future flow evolution is difficult to estimate. During simulation, this uncertainty map is transferred to a modified OpenFOAM solver and used as a refinement indicator for dynamic mesh adaptation. The same trained model is applied to square and diamond cylinders without retraining or geometry-specific threshold adjustment, showing its potential for cross-geometry wake refinement. This framework demonstrates how learned uncertainty can serve as a physically interpretable and transferable AMR criterion for CFD simulations.

Speaker: Dr. Takahiro Ushioku (Assistant Professor, Tokyo University of Agriculture and Technology)

Abstract: Understanding the three-dimensional (3D) stress field in non-Newtonian fluid flows is essential for clarifying the mechanisms of cerebral aneurysms and designing functional soft materials. Recent advances in flow birefringence, where retardation and orientation data are acquired as 2D optical projections of the shear stress field, have offered pathways to access the 3D fluid stress field. However, reconstructing the full 3D stress tensor from these integrated optical projections remains a highly challenging, underdetermined inverse problem. To address this limitation, we have recently proposed Unsupervised Flow-Integrated PhotoElastic Tomography (U-Flow PET), a framework for reconstructing 3D fluid stress fields using a 3D convolutional neural network coupled with physical constraints, namely the incompressible Cauchy momentum equation and boundary conditions. In this presentation, we show reconstruction results based on synthetic flow datasets under various flow conditions. We then demonstrate that our proposed framework successfully recovers the full-component 3D fluid stress tensor field from 2D optical projections.

Speaker: Mr. Minseo Lee (Ph.D. student, POSTECH)

Abstract: The helical coil steam generator (HCSG) has one of the most complex flows among reactor components. Computational fluid dynamics (CFD) has been used to compute such flows, but its long computation time limits real-time use. Recently, machine learning has been studied to accelerate CFD, and the neural operator is one such approach. This study applies and compares neural operators for the Kármán vortex flow field in the HCSG. The latent DeepONet (L-DeepONet) couples a reduced-order model with DeepONet to learn high-resolution CFD data efficiently. To mitigate the spectral bias in predicting the strongly periodic Kármán vortex, a multi-scale technique is applied to L-DeepONet to construct the Multi-scale L-DeepONet. The FNO and Multi-scale FNO are applied for comparison. Even under unseen conditions, the Multi-scale L-DeepONet reproduces the Kármán vortex and its shedding, whereas the FNO family fails to capture them and recovers only the mean flow. DTW values for the velocity magnitude time histories at several points are the smallest for L-DeepONet. All proposed models are at least several thousand times faster than CFD, with potential for real-time HCSG simulations.

Speaker: Dr. Yasuhiro Yoshida (Postdoc Researcher, The University of Tokyo)

Abstract: Reducing skin-friction drag in wall-bounded turbulent flows is an important issue for engineering applications. Traveling-wave control, in which a sinusoidal wall-normal deformation propagates in the streamwise direction, is a promising drag-reduction technique. This study investigates its performance and underlying mechanisms in a spatially developing turbulent boundary layer through wind-tunnel experiments. The drag reduction was evaluated using the von Kármán momentum integral equation, indicating a global drag-reduction rate of 9.4% over the controlled length. Turbulent statistics show that the drag reduction is achieved through suppression of turbulent momentum transfer. Measurements at multiple downstream stations reveal that turbulence suppression is initially confined to the near-wall region and subsequently spreads toward the outer region. The reduction in turbulent intensities exhibits a progressive diffusion from the near-wall region, following a power law with downstream distance. Two-point correlations indicate reduced streamwise coherence of buffer-layer streaks induced by the traveling wave.

Speaker: Mr. Juhyeong Lee (Ph.D. student, Hanyang University)

Abstract: In molten salt reactors the fuel itself flows, carrying delayed-neutron precursors out of the core; a change in pump speed then redistributes them around the loop and alters the reactivity. Capturing this demands both the system-wide flow and the detailed in-core precursor field. We present a multiscale scheme that couples a 1D system code, resolving the whole-loop flow, to a CFD solver that resolves the in-core distribution of the six precursor groups and weights them by an adjoint importance from OpenMC; the reactivity then follows from this importance-weighted in-core source. The scheme is validated against the MSRE pump-transient benchmark using corrected experimental data. The measured pump-coastdown reactivity, a swing of about 200 pcm, is reproduced in agreement with the data, while a small phase lead at the onset of startup reflects the uncertainty of the estimated flow itself and appears in existing system codes as well. Adjoint weighting proves essential to this agreement, and in axisymmetric transients the 3D solution collapses onto the 1D one, confirming the consistency of the coupled multiscale scheme.

Speaker: Dr. Shaghayegh Zamani Ashtiani (Postdoc Research Associate, Tohoku University)

Author list: Shaghayegh Zamani Ashtiani, Soshi Kawai, and Kai Fukami

• Department of Aerospace Engineering, Tohoku University 

Abstract: Many unsteady aerodynamic flows appear complicated with a range of spatiotemporal length scales, yet the dynamics that govern them are often far simpler. Reduced-order models can extract this simplicity, but their design depends on the flow physics of interest. This talk centers on two-dimensional transonic buffet, a flow described by millions of degrees of freedom whose periodic nature leads to low-dimensional underlying dynamics. We compress the flow with proper orthogonal decomposition followed by a nonlinear autoencoder, whose structured latent space yields a three-dimensional embedding encoding buffet phase, frequency, and angle of attack. Inspired by the supercritical Hopf bifurcation, this embedding supports a parametrized reduced representation for quasi-steady pitching as well as flow-field reconstruction from sparse sensors. Towards the end of the talk, we also briefly introduce our recent efforts on time-dependent basis-based modal analysis that linearly evolves the bases directly in time rather than fixing them as in conventional approaches with an example of extreme vortex gust encounter. These complementary strategies show that encoding prior knowledge of flow physics in the model design is key to parametrization and, ultimately, surrogate modeling of unsteady aerodynamic flows.

Speaker: Mr. Hanbi Kang (Ph.D. student, Dong-A University)

Abstract: Computational fluid dynamics has become an essential tool for solving a wide range of engineering problems. However, the substantial cost of repetitive high-fidelity simulations limits its use in applications requiring rapid or long-term predictions. This challenge has motivated the development of reduced-order models (ROMs), which capture the dominant dynamics of high-dimensional systems in low-dimensional spaces. In particular, Koopman operator theory provides a promising framework for representing nonlinear dynamics through linear evolution in a space of observable functions. Building on this framework, Koopman autoencoders (KAEs) use neural networks to learn nonlinear latent representations while modeling their temporal evolution with a linear Koopman matrix. Despite their potential, conventional KAEs may suffer from accumulated errors and unstable long-term predictions when the spectral properties of the Koopman matrix are not properly constrained. In this work, we present KAEs that impose structural constraints on the latent dynamics. Using unsteady flow-field data, we compare the proposed approaches with a conventional KAE and examine how these constraints affect long-term prediction accuracy and stability. This presentation highlights the potential of structured latent dynamics for developing stable data-driven ROMs.