Litcius/Paper detail

Turbulence Modeling Through Deep Learning: An In-Depth Study of Wasserstein GANs

Wajdi Alghamdi, S. Mayakannan, G A Sivasankar, Jagendra Singh, B. Ravi Naik, Ch. Venkata Krishna Reddy

202351 citationsDOI

Abstract

This research study compares the accuracy of different techniques based on deep learning (DL) for predicting turbulent flows. Different types of Generative Adversarial Networks (GANs) are examined in terms of their applicability to the study and simulation of turbulence. Next, we select Wasserstein Gans (WGANs) to produce localized disturbances. Network features including the learning rate and loss function are examined as they pertain to the performance of the WGANs during training on turbulent data gleaned from high-resolution Direct Numerical Simulations (DNS). DNS input data and the generated turbulent structures are proven to agree qualitatively well. The projected turbulent fields are evaluated quantitatively and statistically.

Topics & Concepts

TurbulenceGenerative adversarial networkGenerative grammarComputer scienceDeep learningResolution (logic)Adversarial systemFunction (biology)Direct numerical simulationHigh resolutionAlgorithmArtificial intelligenceStatistical physicsMechanicsPhysicsGeologyReynolds numberRemote sensingBiologyEvolutionary biologyFluid Dynamics and Turbulent FlowsModel Reduction and Neural NetworksAerodynamics and Acoustics in Jet Flows
Turbulence Modeling Through Deep Learning: An In-Depth Study of Wasserstein GANs | Litcius