Dissertation Defense: Data-Driven Analysis Methodologies for Unsteady Aerodynamics from High Fidelity Simulations

Committee Members

• Dr. Datta Gaitonde, Chair (MAE)
• Dr. Sandip Mazumder (MAE)
• Dr. Jen-Ping Chen (MAE)
• Dr. Mei Zhuang (MAE)

Abstract

In the recent years, with the advent of high performance computing high fidelity 3-D Computational Fluid Dynamics (CFD) for cases of practical interest have become feasible. In particular, approaches like Large Eddy Simulation (LES) and Direct Numerical Simulation (DNS) provide unprecedented insight into the physics of turbulent flows, making them indispensable research tools. Despite their promise, extracting information from them is proving to be a challenge due to their extremely large size. For such large datasets, conventional analysis and visualization techniques become intractable without resorting to statistical methods more suited to “big data” problems. There are several such methods which usually place emphasis on a particular aspect of the dataset. Interestingly, these techniques can be used in conjunction with each other to complement their strengths and identify key features of the phenomena in the dataset. The central goal of this dissertation is to develop strategies based on novel statistical, model reduction and signal processing techniques to derive fundamental physical insights into large CFD datasets of practical interest. To demonstrate this, three LES datasets of various unsteady flow-fields arising in Micro Air Vehicle (MAV) flight have been analyzed in depth. They are a) Static stall of a NACA 0015 airfoil with plasma control, b) Dynamic stall in a plunging SD 7003 airfoil, and c) Interaction of a stream-wise oriented vortex impinging on a rectangular wing. The strengths of recent model reduction techniques like the Dynamic Mode Decomposition (DMD), have been demonstrated in depth for these cases and its relationship with Proper Orthogonal Decomposition (POD) has been studied. Additionally, a major emphasis of this work has been on analysis of non-stationary signals arising from turbulent flows, since traditional Fast Fourier Transform (FFT) based approaches have significant limitations. One such technique that has been explored is the Empirical Mode Decomposition (EMD). Initially popularized in the earth sciences community, this work provides further evidence that EMD can be highly effective for aerodynamics problems, especially when used to aid other analysis techniques. However, despite its usefulness, EMD for generalized multivariate signals remains a challenge. As a potential solution, this work introduces Multivariate EMD (MEMD), which has hitherto been used primarily for neuroscience applications and sensor fusion. MEMD has been adapted to aerodynamics datasets where it successfully extracts intermittent features/oscillations among its multivariate signals. As a result, it is able to uncover several new dynamics of the flow which were previously elusive. Another major focus of this work is the spectral analysis of non-stationary signals. In order to circumvent the difficulties associated with user-defined parameters in wavelet analysis, the use of the algorithm of Matching Pursuits (MP) to compute spectra is proposed. MP has seen increased adoption in the neuroscience community for spectral analysis of brain EEG data, with considerable success. This work successfully demonstrates spectral analysis of turbulence data using MP to track and analyze disturbances in the flow-field. Finally, the overarching theme of this thesis has been to exhibit how various model reduction and signal processing algorithms can be leveraged to extract insight into turbulent flows of engineering interest. In addition to exploring fundamental physics of turbulence, this work also outlines best use practices and implementation details for some of the newer techniques like MEMD and MP, for potential application to other areas of fluid mechanics.