Date of Award
Doctor of Philosophy (PhD)
Mechanical and Aerospace Engineering
Mark N. Glauser
Jet Flows, Neural Networks, Turbulence
The desire for aircraft to fly higher, farther, and faster has led to the use of more complex nozzle geometries over the last several decades. These nozzles often take advantage of multiple high-velocity streams issuing from non-axisymmetric exit areas and can be optimized for airframe integration and stealth performance. The associated turbulent jet physics and the aeroacoustic phenomena have been studied thoroughly over the past 80 years; however, the complex non-linear interactions between multiple merging canonical flows are not well understood. In this study, we seek to gain a better understanding of the flow physics in the near-field of a particular complex supersonic jet flow by using a combination of fundamental experiments, data fusion, and novel estimation techniques based on modern machine learning techniques.
The experiments described herein focus on a Multi-Aperture Rectangular Single Expansion Ramp Nozzle (MARS ) and are a direct extension of the research performed by Magstadt , Berry , and Kan . These experiments characterize the near-acoustic pressure field of this configuration in several operating conditions. This pressure field is probed using six Kulite transducers arranged in several different configurations and sampled simultaneously at 100 kHz.
In the nominal operating configuration, the pressure data exhibit a strong peak at StDs=0.3 when scaled using the thickness of the internal splitter plate. In cases of differing shear between the main and third streams, this peak shifts to different Strouhal Numbers based on this change in shear. In cases of high third-stream nozzle pressure ratio (NPR3), the frequency of the dominant pressure fluctuations shifts to StDs =0.39. This suggests that the asymmetric von Kármán vortex street observed in other studies has been stabilized under these conditions. This lends credence to recent observations made by Stack  re- garding the structure of the wake behind the splitter plate. Differing shear conditions were also found to impact the directionality of the far-field noise in the sideline plane. At the
nominal third-stream operating condition (NPR3 = 1.89) an increase in acoustic emission is observed centered around the 60◦ from the jet axis. This trend in the directionality is suppressed when the third stream is subsonic, or highly under-expanded. The low-frequency components (<10,000 Hz) of the pressure fluctuations just outside the jet and in the far field remain largely unaffected by the third-stream nozzle pressure ratio, suggesting that they are tied to the dynamics associated with the bulk flow.
Because the timescales in the flow issuing from the MARS nozzle are very short, collecting time-resolved full-field velocity measurements is not feasible in the current configuration. To circumvent this, a neural network-based formulation of the POD/LSE Modified Complementary Technique is proposed and studied. The method combines the data reduction capabilities of Proper Orthogonal Decomposition (POD), with a machine learning based formulation of Linear Stochastic Estimation (LSE). The baseline method is tested using data from the mixing layer of a subsonic axisymmetric jet and the pressure field around it. To quantify its performance, the machine learning based estimation method is compared to a traditional LSE model. In relation to the MARS configuration, the functional relationship between the near-field pressure measurements and the time-dependent coefficients of the POD modes is modeled using an Artificial Neural Network (ANN ) trained on a fused data set. This data set combines experimentally measured POD eigenfunctions and high-fidelity Large Eddy Simulation (LES) data. Using the resulting model and only the pressure data measured during the near-field experiments, we estimate a low-dimensional representation of the time-resolved velocity field at several locations. The estimated fluctuating velocity fields produce an average vorticity field that is qualitatively similar to that measured by Magstadt . In addition, an ANN is used to model the relationship between basic nozzle geometric properties and operating parameters, and overall sound pressure level (OASPL) measured in the far-field. It is observed that the model makes accurate predictions (± 4 dB) for a variety of nozzle configurations not included in the training data.
Tenney, Andrew Steven, "Modern Methods in Machine Learning as Applied to the Study of a Complex Supersonic Jet Flow" (2019). Dissertations - ALL. 1023.