MK
M.M. Kanala
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
1 records found
1
Turbulence is a ubiquitous phenomenon in nature, characterized by complex and unpredictable fluid motion that is challenging to simulate. Turbulence modeling seeks to develop mathematical frameworks capable of predicting turbulent flow behavior in various systems. Common approaches include Reynolds-Averaged Navier-Stokes (RANS), Large Eddy Simulation (LES), and Direct Numerical Simulation (DNS). Among these, LES offers high-fidelity predictions by resolving most of the energetic scales of turbulence, making it particularly suitable for accurate flow analysis. However, LES faces limitations in capturing near-wall phenomena, which require extremely fine meshes to resolve adequately.
Recent advances in deep learning, a branch of machine learning, have been applied to overcome some shortcomings of traditional LES models, especially in predicting subgrid-scale (SGS) terms. This thesis investigates the extension of existing multilayer perceptron (MLP) models from prior studies to 3D turbulent channel flow. The focus is on comparing network hyperparameters, feature sets, and training data to determine the most compact and efficient model capable of accurately reproducing SGS closure terms. Building on previous research at TU Delft on LES-VMM, this study emphasizes developing an effective network with a simple architecture that maintains high predictive accuracy while reducing computational complexity. ...
Recent advances in deep learning, a branch of machine learning, have been applied to overcome some shortcomings of traditional LES models, especially in predicting subgrid-scale (SGS) terms. This thesis investigates the extension of existing multilayer perceptron (MLP) models from prior studies to 3D turbulent channel flow. The focus is on comparing network hyperparameters, feature sets, and training data to determine the most compact and efficient model capable of accurately reproducing SGS closure terms. Building on previous research at TU Delft on LES-VMM, this study emphasizes developing an effective network with a simple architecture that maintains high predictive accuracy while reducing computational complexity. ...
Turbulence is a ubiquitous phenomenon in nature, characterized by complex and unpredictable fluid motion that is challenging to simulate. Turbulence modeling seeks to develop mathematical frameworks capable of predicting turbulent flow behavior in various systems. Common approaches include Reynolds-Averaged Navier-Stokes (RANS), Large Eddy Simulation (LES), and Direct Numerical Simulation (DNS). Among these, LES offers high-fidelity predictions by resolving most of the energetic scales of turbulence, making it particularly suitable for accurate flow analysis. However, LES faces limitations in capturing near-wall phenomena, which require extremely fine meshes to resolve adequately.
Recent advances in deep learning, a branch of machine learning, have been applied to overcome some shortcomings of traditional LES models, especially in predicting subgrid-scale (SGS) terms. This thesis investigates the extension of existing multilayer perceptron (MLP) models from prior studies to 3D turbulent channel flow. The focus is on comparing network hyperparameters, feature sets, and training data to determine the most compact and efficient model capable of accurately reproducing SGS closure terms. Building on previous research at TU Delft on LES-VMM, this study emphasizes developing an effective network with a simple architecture that maintains high predictive accuracy while reducing computational complexity.
Recent advances in deep learning, a branch of machine learning, have been applied to overcome some shortcomings of traditional LES models, especially in predicting subgrid-scale (SGS) terms. This thesis investigates the extension of existing multilayer perceptron (MLP) models from prior studies to 3D turbulent channel flow. The focus is on comparing network hyperparameters, feature sets, and training data to determine the most compact and efficient model capable of accurately reproducing SGS closure terms. Building on previous research at TU Delft on LES-VMM, this study emphasizes developing an effective network with a simple architecture that maintains high predictive accuracy while reducing computational complexity.