Low-complexity stochastic modeling and control of turbulent flows

Abstract

Experimental and numerical datasets are becoming increasingly available for a wide range of flow configurations and Reynolds numbers and advanced machine learning techniques are being used to extract patterns from multimodal datasets. However, unreliable measurements and data anomalies challenge the viability of the resulting data-driven models. Furthermore, data-driven techniques are agnostic to the underlying physics and are not robust to outliers that are not accounted for in the training process. This compromises the performance of data-driven models in regimes that were not contained in the available data and limits their utility for flow control. On the other hand, dynamical models of turbulent flows that are based on the Navier-Stokes (NS) equations typically have a large number of degrees of freedom and are not suitable for optimization and control design.We propose a synergistic approach for refining physics-based models using data-driven techniques. This will allow us to achieve statistical consistency with the result of nonlinear simulations using the stochasticallyforced linearized NS equations. Our models will be of low-complexity and well-suited for analysis, optimization, and control. Based on this, we propose to explore the efficacy of our data-refined physics-based models in the estimation and control of wall-bounded flows. We will address the design of spatially distributed architectures for sensing and actuation. In addition, we will use our models to capture the geometric scaling and wall-attachment of energetically dominant turbulent flow structures. This will allow us to critically assess the benefits and limitations of our models and will guide our efforts in the model-based control of high Reynolds number flows. As an integral component of our work, we will make algorithmic advances to improve the scalability and broaden the scope of the proposed modeling and control framework.

Document Details

Document Type
DoD Grant Award
Publication Date
Jul 11, 2018
Source ID
FA95501810422

Entities

People

  • Mihailo R. Jovanović

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • University of Southern California

Tags

Fields of Study

  • Physics

Readers

  • Computational Fluid Dynamics (CFD)
  • Distributed Systems and Data Platform Development
  • Fluid Mechanics and Fluid Dynamics.

Technology Areas

  • AI & ML
  • AI & ML - Machine Learning Algorithms