Abstract:Inflow turbulence generation with physical spatiotemporal characteristics and the requirement of a specific turbulence state is one of the most challenging scientific problems for engineering applications of large eddy simulation (LES)/direct numerical simulation (DNS) methods. The existing state-of-the-art methods require either precursor simulations or relatively long transitional sections plus a priori parameter tuning, which seriously limits their physical realizability and applicability in applications. In this paper, we propose a universal inflow turbulence generator for the large eddy simulation of a wide-speed compressible turbulence boundary called universal discretizing and synthesizing random flow generation (UDSRFG), which is developed based on a synthetic inflow turbulence generation method proposed by one of the authors for incompressible turbulence flows. Unlike the existing methods, the proposed method can theoretically guarantee divergence free of mass flux fluctuations, which fully eliminates unexpected instability noise sources for both low-speed incompressible and high-speed compressible flows. Large eddy simulations of compressible subsonic flat turbulent boundary layer flow and supersonic turbulent boundary layer flow are conducted to demonstrate the effectiveness and superiority of the proposed method. The synthetic inlet turbulence generated by UDSRFG rapidly develops into realistic turbulence and evolves the expected propertied of the turbulent boundary layer with a reduction in the transitional length in boundary layer flows of 50%~85%Â compared with that of the existing methods. The detailed turbulence quantity statistics, including the velocity fluctuations, mean velocity, friction coefficient, spectral distribution, and spatial correlation, indicate that the generated boundary layer satisfies all of the expected necessary properties provided in the literature and behaves more consistently for most of the validation datasets than the existing methods. Furthermore, the problem of the shock-wave/turbulent boundary-layer interaction induced by a compression ramp with a high Reynolds number is simulated to fully validate its advantages of good physical realizability and applicability in typical supersonic engineering issues.

CoNFiLD-inlet: Synthetic Turbulence Inflow Using Generative Latent Diffusion Models with Neural Fields

CoNFiLD: Conditional Neural Field Latent Diffusion Model Generating Spatiotemporal Turbulence

Conditional Neural Field Latent Diffusion Model for Generating Spatiotemporal Turbulence

A transformer-based synthetic-inflow generator for spatially developing turbulent boundary layers

Deep unsupervised learning of turbulence for inflow generation at various Reynolds numbers

Synthetic turbulent inflow generator using machine learning

Physics-guided deep learning for generating turbulent inflow conditions

A Universal Inflow Turbulence Generator for the Large Eddy Simulation of a Wide-Speed Compressible Turbulence Boundary Layer with Minimal Transitional Length

A General Inflow Turbulence Generator for Large Eddy Simulation

Inflow turbulence generation for compressible turbulent boundary layers

PiRD: Physics-informed Residual Diffusion for Flow Field Reconstruction

Inflow turbulence generation using an equivalent boundary layer model

Neural Differentiable Modeling with Diffusion-Based Super-resolution for Two-Dimensional Spatiotemporal Turbulence

Synthetic Lagrangian turbulence by generative diffusion models

Bayesian Conditional Diffusion Models for Versatile Spatiotemporal Turbulence Generation

A Denoising Diffusion Model for Fluid Field Prediction

Benchmarking Autoregressive Conditional Diffusion Models for Turbulent Flow Simulation

Deep Fluids: A Generative Network for Parameterized Fluid Simulations

Accelerating Multiphase Flow Simulations with Denoising Diffusion Model Driven Initializations

Nonlinear constitutive calculation method of rarefied flow based on deep convolution neural networks