2026-07-30 | | Total: 12
The aerodynamic design of turbomachinery is critical to the performance of the overall energy system, yet it is challenging due to the complex non-linear flow physics and the presence of multiple-target design compromises. Denoising diffusion model, as one of the leading approaches in generative machine learning, has shown its advantages of high design solution accuracy and diversity in many engineering applications. In this study, we bring it to the 3D inverse design problem of turbomachinery, using centrifugal compressors as a classic representative, to demonstrate the new methodology for complex geometry designs. A diffusion model-centred design framework has been developed in this study. By specifying the desired design condition (mass flow rate and rotational speed) and targeted performance (pressure ratio and efficiency), the trained diffusion model returns directly the 3D compressor geometry that satisfies the condition inputs. Compared to traditional deterministic forward design approaches, the proposed method not only generates accurate geometry solutions to inverse design problems, but also enables effective exploration of the entire design space, providing a diverse set of candidate solutions. In addition, this paper presents the first study to directly train on 3D blade geometry coordinates rather than parametrised representations, demonstrating the feasibility of coordinate-based learning while enabling a highly flexible framework applicable to a wide range of designs. The trained diffusion model achieves excellent design capability, with solution accuracy up to 99% and unfeasible designs less than 1%. Furthermore, the solution diversity of the trained diffusion model is also quantitatively verified by means of comparing the distribution of the solution sets generated from the diffusion model and from direct sampling of physical parameters.
Intermittency in turbulence manifests as intense vortices and sharp peaks of dissipation. Causing the breakdown of Kolmogorov's simple self-similar theory, it leads to anomalous scaling, multifractality and so far remains beyond the scope of a complete theoretical description. How intermittent flow structures influence these different measurements is not known quantitatively. With a simple filtering procedure-thresholding vorticity and inverting the Biot-Savart law to generate filtered velocity fields-we show the effects of intermittent flow structures can be disentangled. As extreme vorticity contributions to the velocity field are filtered out, the energy spectrum scaling persists, while the bottleneck is flattened, and structure function scalings tend towards their Kolmogorov values. The approach is more rapid for transverse exponents, revealing the selective importance of intensely swirling flow regions. Similarly, the extent of multifractality reduces as intermittency is filtered, shrinking the range of roughness singularity exponents. The residual fields are curiously more multifractal, but their structure begins to break away from an underlying turbulence skeleton. The effects on vortex stretching and strain self-amplification are quantified. Our work shows that a Biot-Savart approach can selectively remove the effects of intermittency from turbulence, and hence from its scalings.
We address the problem of locating a chemical source in a flow. Based on the duality between the concentration field and Lagrangian tracer trajectories, we interpret concentration detections as evidence of paths connecting the source to the detection points. This Schrödinger bridge formulation between plausible emission positions and detection points leverages the backward propagator of passive tracers to frame source localization as the sampling of candidate emission locations via Langevin dynamics. The associated drift reveals classical chemotaxis and cast-and-surge as complementary behaviors emerging from a single transport-based principle. Applied to olfactory search in two-dimensional turbulence, the proposed backtracking framework outperforms classical strategies across varying wind regimes using a single, Galilean-invariant learned propagator.
Data-driven subgrid-scale closures for large eddy simulation are of significant interest in many engineering and geoscience applications. In this context, several important questions remain about the role of rotational equivariance as an inductive bias for learned tensorial mappings. We investigate whether equivariance improves accuracy, parameter efficiency, and generalization for subgrid-scale modelling at realistic filter ratios. For turbulent channel flow, we compare data-augmented non-equivariant architectures to those with equivariance as an inductive bias. We compare both pointwise and nonlocal versions of these two model classes. All models are evaluated at matched parameter counts across spatiotemporal, anisotropy, and Reynolds number generalization. We show that non-augmented models learn a small degree of equivariance directly from turbulence data, especially when that data is more isotropic. The equivariant nonlocal architecture attains the highest correlation coefficient on every generalization test at approximately half the parameter count of its non-equivariant counterpart, while the pointwise architectures do not improve on the analytical Clark baseline. Additionally, the equivariant model is more data-efficient than a non-equivariant model. The benefit of equivariance grows with the receptive field of the model, indicating that equivariance and nonlocality are both useful for the subgrid-scale closure task at realistic dataset size, parameter counts, and filter size.
We present an experimental investigation of mixing dynamics within a laboratory-scale 3-liter stirred tank reactor (STR) equipped with two Rushton turbines and three baffles. Using time-resolved, four-dimensional particle tracking velocimetry, we successfully capture trajectories of up to 40,000 tracer particles in the full reactor volume despite obstructions by stirrer and baffles, providing unprecedented time-resolved flow and mixing information. From these Lagrangian data, we analyze velocities, accelerations, and spatial dispersion, revealing anisotropic mixing. By utilizing novel network-based analysis methods on the experimental particle trajectories, we identify coherent fluid compartments that exhibit strong internal mixing but weak exchange with neighboring compartments. We uncover five distinct compartments acting as transport barriers, which have a high impact on substrate distribution in chemical and biochemical processes. Our approach thus realizes and extends early thought experiments from Danckwerts and Levenspiel by providing detailed insight into the behavior of single fluid parcels and Lagrangian mixing withing chemical and biochemical reactors, offering a valuable approach for evaluation and optimization of chemical and biochemical processes. The trajectory data are made freely available to serve as an experimental reference for further research.
We study the effect of surfactant adsorption kinetics on water electrolysis performance using Surfynol 465, an ultrafast-kinetics surfactant. High-speed optical diagnostics reveal that Surfynol 465 reduces bubble residence time by an order of magnitude. Compared to the slower surfactant Triton X-100 and the surfactant-free baseline, it also prevents the growth of large bubbles ($>200\ μ$m). We validated these results on an anion-exchange membrane electrolyzer (AEMEL) test bench. Adding Surfynol 465 to the $1\text{ M KOH}$ electrolyte decreases cell overpotential by 140--150 mV. It nearly triples the current density (from 0.13 to 0.37 A/cm$^2$ at 2 V) and increases average power by 40\% during a week-long test. Furthermore, downstream corrosion analyses reveal a strong alloy-dependent response. The surfactant reduces degradation in stainless steel and brass but accelerates corrosion in carbon steel.
This study considers extracting causally important vortical structures from the extreme vortex gust-airfoil interaction at a chord-based Reynolds number of $5000$. This extraction is achieved by decomposing a given vortical flow snapshot into its informative and residual components based on the contribution to an arbitrary future target variable with convolutional information-theoretic learning. For the current vortex-airfoil interactions that exhibit transient and multiscale flow characteristics, we first examine the important vortical structures with respect to a future lift coefficient. While the vortex cores are primarily highlighted before vortex impingement, the emerging shear layers are additionally captured after the massive separation, which is evident from a comparison to an instantaneous force-element analysis. We further take the scale-dependent energy transfer as a future variable of interest to examine its impact on the extracted informative structures compared to the lift-associated structures. They are distinct from the lift-based structures in the early stage of the gust encounter yet become similar after impingement, revealing an analogy between informative structures across different transient aerodynamic mechanisms. The present data-driven approach selectively extracts the specific important flow structures responsible for the physics of interest, which can support studying a range of transient aerodynamic flows from the causal, data-driven perspective.
Recent experimental observations revealed that clean spherical bubble chains generate a nearly uniform upward liquid flow. The physical origin of this liquid flow, however, remains unclear. Hydrodynamic interactions in aligned bubble chains were therefore investigated using high-accuracy embedded-boundary simulations that resolve the interfacial boundary layer while capturing long-range interactions among multiple clean spherical bubbles. The simulations were used to quantify the evolution of hydrodynamic interactions within bubble chains and to identify the conditions under which bubble-induced liquid flow can be represented by the superposition of isolated-bubble far wakes. Based on these findings, a reduced-order model was developed and applied to experimentally measured bubble trajectories. The model successfully reproduced the nearly uniform upward liquid flow observed in the experiments. To clarify the role of bubble dispersion, the predictions were compared with those for a uniformly dispersed bubble arrangement having the same overall dispersion width. While the uniformly dispersed arrangement produced a center-peaked velocity distribution, only the experimentally observed bubble trajectories reproduced the nearly uniform upward liquid flow. These results demonstrate that the liquid flow is governed not simply by the dispersion width but by the bubble trajectories that establish the spatial distribution of far wakes. The present study provides a physical framework linking hydrodynamic interactions among bubbles, bubble dispersion, and bubble-induced liquid flow, together with a reduced-order model for predicting the liquid flow generated by bubble chains.
We study by means of direct numerical simulations the influence of the dimensionality of convection on flow properties. We call attention to a few general principles from considering in totality the results from 1D, 2D, 3D and 4D. In particular, we explore two practical aspects: (1) The transient time, or the amount of time it takes for the flow to reach the steady state; and (2) possible implications for the so-called ultimate state.
In reciprocating flapping-like motions, wing-wake interaction plays a crucial role in fluid force generation. While this effect's existence has been acknowledged, particularly in explaining discrepancies between measured forces and quasi-steady approximations, fundamental research on the mechanism underlying this interaction and its scaling remains limited. To address this, we investigate the excess drag force, relative to quasi-steady estimates, acting on a flat plate during the reversal phase of a forward and back translational motion. The flow produced by this motion, studied at insect-flight-relevant Reynolds numbers, serves as a simplified analogue to biological flapping. We demonstrate that interaction with pre-existing wake flow indeed generates excess drag. The main parameter governing this interaction is the distance travelled before reversal, which influences both magnitude and temporal dynamics of the peak drag. We link our observations to optimal vortex formation, as the time trace of the additional drag during reversal is qualitatively altered by the detachment of the starting vortex ring: vortex detachment and re-formation lead to two distinct wake-force peaks. Furthermore, as the pre-reversal distance traversed increases, the wake interaction force post-reversal decays more slowly. Flow observations reveal a similar spatial decay of the streamwise velocity in the wake at the moment of reversal, suggesting a direct link. Representing the starting vortex as a point vortex indicates that the wake's spatial scaling, and commensurately the temporal scaling of the wing-wake interaction effect, is primarily governed by the vortex ring position, shape, and circulation. This simplification reveals a dependence on the pre-reversal translation distance that can be described by a combined fourth-root and linear scaling.
Consider a material point in finite dimensions moving under the influence of a potential force according to Newton's laws. Suppose the potential energy function is a generalized well, strictly convex transverse to a smooth submanifold $M$ on which it is minimal. If the potential is steep, one expects the particle will oscillate rapidly about this submanifold and, if the initial displacement is not too great, should approximately move along $M$ as if it were ideally constrained (e.g. geodesic). This expectation is true if the initial conditions are very well prepared but may fail otherwise - additional potential forces determined by how the Hessian of the potential varies along $M$ may be present. The origin of this force is that the transversal motion acts as a simple harmonic oscillator with a slowly varying frequency, which approximately conserves action, not energy. In this work, we regard continuum mechanical systems such as the elastic thread or compressible fluid as material points moving in an infinite dimensional space according to Newton's laws for appropriate potential energy functionals. We show how to arrive at ideally constrained systems such as the inextensible thread and incompressible fluid as a limit of a strong potential force, computing also corrections to the naive predictions when the data is not very well prepared. For example, for the thread we find a resistance to bending emerge from a strong resistance to compression/expansion. For the fluid, the effective incompressible dynamics may be driven by a remnant acoustical wavefield. Both of these emergent features are nonlinear and non-local. Finally, we give examples of some limits for which the naive models robustly hold because the additional force is trivial. These include the homogeneous incompressible Euler, anelastic Euler, as well as the lake and great lake equations.
Swimming microorganisms often operate in complex confinement, where an interplay of long-ranged hydrodynamic interactions and a short-ranged repulsive interaction can give rise to interesting dynamical behaviors. Here, we theoretically investigate the trajectories of microswimmers - modeled as squirmers - in the presence of periodic boundaries. The latter modify their swimming velocity, leading to behaviors that differ qualitatively from swimming near planar walls. Using a perturbative approach based on bispherical coordinates and the Lorentz reciprocal theorem, we characterize the interaction between a squirmer and a periodic surface in the limit of small surface amplitude and systematically explore its dependence on the boundary corrugation wavelength, squirmer type, orientation, and swimmer-surface distance. Most importantly, our results reveal that pullers become trapped in the valley of the surface corrugations, in contrast to their sliding motion near planar walls. Furthermore, the near-surface dynamics of pushers display oscillations, reflecting the periodicity of the surface structure. A tilt of the swimmer orientation with respect to the surface corrugations results in a wave-length dependent drift that sorts pushers from pullers. These findings highlight the impact of hydrodynamic interactions in shaping microswimmer transport near structured boundaries with potential implications for microbiological phenomena, such as biofilm formation, and technological applications.