High-fidelity immersed-boundary simulation resolves the coupled motion of a deforming swimmer and its surrounding flow, but the resulting cost limits repeated evaluations for engineering design, parameter studies, and control. We develop neural-operator surrogates for temporal prediction of the hydrodynamic fields generated by planar and volumetric eel swimmers. The surrogates are trained on regular-grid fields exported from adaptive fluid--structure simulations and are conditioned on swimmer geometry and Reynolds number. The planar model jointly predicts two velocity components, scalar vorticity, and pressure. On five held-out high-Reynolds-number trajectories, its full-domain global relative L^2 error is 3.51 %. The volumetric formulation uses three target-specific models with a common multichannel input: one model predicts three-dimensional velocity, one predicts vorticity, and one predicts pressure. Their full-domain global relative L^2 errors on five held-out within-range trajectories are 3.44 %, 5.58 %, and 19.2 %. Together, the results demonstrate the feasibility of field-resolved neural surrogates for moving-boundary swimmer flows while identifying pressure accuracy and physical consistency as priorities for further development.
The transient dynamics of the micrometer-scale slipper-pair oil film in axial piston pumps are governed by multi-factor coupling effects, rendering traditional numerical methods inadequate for real-time simulation and rapid design optimization. This paper proposes a physics-enhanced neural operator (PENO) to predict transient lubrication characteristics. PENO combines a motion module that predicts the film thickness at three feature points with a hard-constrained Fourier neural operator for pressure reconstruction. The pressure field is decomposed into a logarithmic, boundary-compatible background field and a masked residual field that analytically enforces the radial Dirichlet pressure boundary conditions. Reynolds-equation residuals and force/moment equilibrium residuals are incorporated into the loss function to improve physical consistency. Evaluations across various operating conditions show that PENO can accurately predict the lubrication characteristics of the slipper-pair oil film. Across the evaluated operating conditions, the full-cycle relative L2 errors of the predicted pressure fields range in order of magnitude from 10−4 to 10−2, while the full-cycle oil-film thickness root mean square errors (RMSEs) range from 10−3 to 10−1 μm. The prediction results meet the physical consistency requirements, with the largest full-cycle RMSEs for the axial supporting force and supporting moments being 7.4 N and 3.8 × 10−3 N m, respectively. Compared to traditional numerical methods, PENO reduces the average full-cycle prediction time from 3.32 × 103 to 6.96 × 10−2 s, representing a speedup of 4.76 × 104. When the mesh size increases from 31 × 121 to 98 × 382, the inference time remains at 7.64 × 10−1 s, indicating favorable inference-time scaling over the tested mesh resolutions.
Hanyu Gao, Hua Fang, Hong Pan et al.· The Physics of Fluids· 0 citations
This study investigates the passive control of backward-facing step flows using rigid and elastic cantilevered plates, with emphasis on flow reattachment, pressure recovery, and fluid–structure interaction effects. The numerical methodology was first validated against an experimental benchmark, yielding a reattachment-length prediction within 1.28% of the measured value. Following validation, simulations were performed using the unsteady Reynolds-averaged Navier–Stokes equations coupled with the Spalart–Allmaras turbulence model. Rigid plate configurations were examined at momentum-thickness-based Reynolds numbers of Reθ=500, 1000, and 5000, while two-way fluid–structure interaction simulations were conducted at Reθ=5000 to evaluate the influence of structural stiffness. For the baseline configuration, the non-dimensional reattachment length and base drag coefficient remained within the ranges of xr/h=6.18–6.34 and cB=0.199–0.205, respectively. The most effective rigid configuration, L/h=2.5, reduced the reattachment length from xr/h=6.34 to xr/h=5.57 and the base drag coefficient from cB=0.205 to cB=0.175 at Reθ=5000, corresponding to reductions of approximately 12% and 15%, respectively. The fluid–structure interaction simulations showed that plate stiffness strongly influences flow-control effectiveness. The stiffest elastic configuration, with E=2×109 Pa, achieved xr/h=6.14 and cB=0.196, whereas more flexible plates exhibited larger deformation and reduced aerodynamic benefit. Overall, the results demonstrate that cantilevered plates provide an effective passive flow-control strategy for backward-facing step flows. Rigid plates deliver the greatest aerodynamic improvement, while elastic plates require sufficient structural stiffness to maintain favorable pressure recovery and flow-reattachment characteristics.
B. Gungordu, Matin Hasanli· Applied Sciences· 0 citations
A fast-converging and asymptotic-preserving adjoint shape optimization method is proposed for drag reduction of multiscale gas flows in vibrating micro-electro-mechanical systems. The convergence of the Boltzmann kinetic equation is accelerated by macroscopic synthetic equations, whose constitutive relations integrate continuum-limit terms and high-order kinetic corrections to faithfully characterize spatiotemporal rarefaction effects. As such, this method maintains near-continuum limit consistency while retaining high kinetic accuracy in rarefied flow regimes. Fourier stability analysis performed in an infinite domain demonstrates that the present method yields a spectral radius below 0.5, indicating that the numerical deviation from the converged solution is halved per iteration. Numerical simulations are conducted on an oscillating cylinder and a comb-shaped resonator. The results verify the high accuracy of the derived adjoint sensitivities and the excellent drag reduction performance of the proposed method across various Knudsen and Strouhal numbers. Compared with conventional kinetic iteration methods, the present method produces convergent primal and adjoint solutions within dozens of iterations and features asymptotic preserving behavior, permitting spatial cell sizes far larger than the molecular mean free path. This facilitates efficient design of vibrating micro-electro-mechanical systems.
Elastoplastic analysis is computationally demanding because its nonlinear, path-dependent constitutive behavior requires incremental loading and repeated iterative solutions. To address this challenge, we propose Plasolver, a physics-informed neural operator framework that combines the efficiency of operator learning with the accuracy and robustness of classical numerical solvers. Plasolver consists of a physics-informed pretraining stage and an optional warm-start stage. During pretraining, the neural operator is trained solely by minimizing the incremental potential energy of elastoplasticity formulated by Simo, without requiring any labeled solution data. It operates directly on unstructured point clouds by encoding spatial coordinates, loading histories, and material properties as unified point-wise prompts. This formulation provides dual invariance to spatial and loading-path discretizations, enabling consistent predictions across different spatial resolutions and different numbers of increments representing the same loading trajectory. The pretrained Plasolver achieves relative errors on the order of 1\% while providing approximately two orders of magnitude acceleration over conventional finite element simulations. In the warm-start stage, the pretrained prediction is supplied as the initial solution to a classical iterative solver, preserving its numerical accuracy, robustness, and convergence properties while substantially accelerating convergence. Numerical results show that Plasolver reduces the required number of iterations by approximately 50\% compared with conventional zero-initialized solvers and converges to solutions at any prescribed tolerance. Plasolver thus provides an efficient, accurate, and discretization-invariant computational framework for nonlinear, path-dependent elastoplastic problems.
Yizheng Wang, M. Eshaghi, Hua-dong Zhang et al.· 0 citations
This study focuses on inertial fusion burn via a novel hybrid surrogate model. The proposed hybrid model delivers computational speedup of several orders of magnitude compared with conventional radiation-hydrodynamic simulations, while maintaining high predictive fidelity. Traditional Fourier Neural Operators (FNOs) typically suffer from numerical divergence after 13–16 timesteps; in contrast, the developed FNO-CA model exhibits superior stability and robustness throughout a 50-timestep simulation window. Although FNO and FNO-CA show comparable one-window prediction accuracy, the advantage of FNO-CA becomes evident in long-term rollout, where it better preserves the temporal evolution of key burn quantities. When applied to parameter-space exploration, the model precisely reproduces the highly nonlinear “ignition cliff” phenomenon and reveals the sensitive dependence of fusion gain on initial thermodynamic and boundary conditions. These results demonstrate that the FNO-CA surrogate serves as a powerful, reliable, and efficient tool for the design, optimization, and real-time control of inertial fusion ignition systems.
Sihan Shen, Mingquan Feng, Y. Lei et al.· Plasma Science and Technolog...· 0 citations
Data-driven turbulence closures are usually calibrated by inverse methods that embed a CFD solver in the loop, tying the model to a particular discretization and requiring every iterate to yield a convergent solve. We instead train the closure inside a physics-informed neural network (PINN): the Reynolds-averaged Navier-Stokes residual is imposed by automatic differentiation, so the inverse problem is mesh-free, differentiable, and solver-agnostic. Because no forward solve runs during training, only the final closure need be solver-stable, arbitrary neural closures are admitted without an adjoint, and the iterative cost of adjoint or ensemble methods vanishes; each hypothesis trains in minutes on a single GPU, so the framework rapidly screens closure forms. We develop four closures: three model the Reynolds stress on a realizable tensor basis -- a local map, a non-local model transporting the turbulent kinetic energy and recovering the out-of-plane normal stress, and the same with a learned length scale l -- and a fourth models the Reynolds force F = -\nabla \cdot \tau directly, free of the realizability constraint. All four are trained across six two-dimensional bluff-body wakes at Re = 10^4 and deployed frozen in a standard finite-element solver, stabilized by input-gradient smoothing and a Lipschitz constraint. Under a strict leave-one-shape-out (LOSO) protocol, all four improve substantially on a steady SST k-omega baseline. The learned-length-scale closure is most accurate on the stress fields, while the force model generalizes best on the mean velocity and drag (LOSO drag error ~8.5%). The closures also train efficiently on Particle Image Velocimetry data, enabling geometries intractable for DNS.
Zhen Zhang, T. Kaufer, Louise Ronglan et al.· 0 citations