2018/04/24 by Arvind Mohan, Mohan, Arvind T., Datta V. Gaitonde +1 · 1 citation
Engineering · Environmental Science · Physics and Astronomy · #Computational Physics (physics.comp-ph) #Energy Load and Power Forecasting #FOS: Physical sciences #Fluid Dynamics (physics.flu-dyn) #Hydrological Forecasting Using AI #Model Reduction and Neural Networks
paper · pdf · doi:10.48550/arxiv.1804.09269
openalex publication_date 2018/04/24 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Reduced Order Modeling (ROM) for engineering applications has been a major\nresearch focus in the past few decades due to the unprecedented physical\ninsight into turbulence offered by high-fidelity CFD. The primary goal of a ROM\nis to model the key physics/features of a flow-field without computing the full\nNavier-Stokes (NS) equations. This is accomplished by projecting the\nhigh-dimensional dynamics to a low-dimensional subspace, typically utilizing\ndimensionality reduction techniques like Proper Orthogonal Decomposition (POD),\ncoupled with Galerkin projection. In this work, we demonstrate a deep learning\nbased approach to build a ROM using the POD basis of canonical DNS datasets,\nfor turbulent flow control applications. We find that a type of Recurrent\nNeural Network, the Long Short Term Memory (LSTM) which has been primarily\nutilized for problems like speech modeling and language translation, shows\nattractive potential in modeling temporal dynamics of turbulence. Additionally,\nwe introduce the Hurst Exponent as a tool to study LSTM behavior for\nnon-stationary data, and uncover useful characteristics that may aid ROM\ndevelopment for a variety of applications.\n