Ph.D Thesis

Ph.D StudentGluzman Igal
SubjectDisturbance Identification in Boundary Layer Flow via
Blind Source Separation
DepartmentDepartment of Aerospace Engineering
Supervisors PROF. Jacob Cohen
PROF. Yaakov Oshman
Full Thesis textFull thesis text - English Version


A novel approach is presented for identifying disturbance sources in wall-bounded shear flows, which can prove useful for active control of boundary layer transition from laminar to turbulent flow. The idea underlying this research consists of considering the flow state, as measured in sensors, as a mixture of sources, and using Blind Source Separation (BSS) techniques to recover the separate sources and their unknown mixing process.

First, a BSS method is introduced that is based on the Independent Component Analysis (ICA) technique. Linear stability theory (LST) is used to model the measured mixtures of sources acquired by sensors placed in the boundary layer. A physics-based criterion is derived for proper sensor placement in order to separate Tollmien-Schlichting (TS) wave disturbances. This criterion is based on assuming a priori knowledge of the TS wave-length. The criterion is verified via numerical simulations of wall-bounded shear flows and by an experimental study involving flow over a flat plate.

Second, to alleviate the limitations of the ICA-based source identification technique, a BSS method based on the Degenerate Unmixing Estimation Technique (DUET) is introduced. This method can be used to identify any (a priori unknown) number of sources by using the data acquired by only two sensors. The DUET method is adapted and used to identify disturbance sources in measured mixtures comprising TS waves and wave-packets. The power of the new method is demonstrated numerically and experimentally.

Third, the DUET-based BSS method is exploited to determine the propagation velocity vector of each of the identified sources by using sensor signals from (at least) three different locations in the flow field. The viability of the method is demonstrated via numerical simulations and wind-tunnel experiments.

Finally, the problem of measuring the velocity signal by nonlinear sensors is addressed. The BSS-based identification methods presented in this work rely on the capability to accurately measure velocity signal mixtures using nonlinear flow sensors, such as hot-wire flow velocity probes. This gives rise to the problem of fast and accurate sensor calibration. To address this issue, a Gaussianization-based statistical calibration technique is presented for estimating the nonlinear calibration curve of hot-wire probes. The method uses as input a measured sequence of voltage samples, corresponding to different unknown flow velocities in the desired operational range, and (only) two measured voltages along with their known (calibrated) flow velocities. The novel calibration method is validated against standard calibration methods using data acquired by hot-wire probes in wind-tunnel experiments.