Application of Data-Driven Modal Decomposition Techniques to the Non-Stationary Case of Scramjet Unstart
|dc.description.abstract||Scramjets are a class of high-speed aerospace propulsion devices that operate with no moving parts and perform combustion at supersonic speeds. They offer a viable option for reusable, efficient, air-breathing hypersonic propulsion and they will be of critical importance to powering the transportation and defense systems of the future. Due to the lack of turbo-machinery in scramjet flow paths, sufficient compression and velocity reduction of the flow must be achieved using the Pre-Combustion Shock Train (PCST), which is a series of shocks in the isolator portion of a scramjet. Under certain conditions, the PCST is ejected out of the front of the engine, causing the engine to unstart. The consequences of unstart are severe, with the engine no longer producing thrust, and potentially causing complete vehicle loss. This work focuses on a high fidelity, computationally generated unstart simulation, in which a pressure induced unstart event is potentially controlled by an upstream cavity. Since such high-fidelity databases are large and difficult to analyze, modal decomposition techniques are often employed. The challenge in applying these techniques for the unstart problem is that it is not stationary i.e., it is unsteady but does not oscillate about a mean position. To address this, two data-driven modal decomposition techniques, Dynamic Mode Decomposition (DMD) and multi-resolution DMD (mrDMD) are examined by applying them to the streamwise component of the unstarting velocity field. The mrDMD algorithm was first validated on the statistically stationary case of supersonic flow over a wall-mounted turret before it was applied to the unstart problem. Results from the DMD analysis of unstart are compared to those from a 4-level mrDMD analysis, which allows for hierarchical time-frequency analysis. DMD was able to capture some gross flow features of the unstart event, such as final PCST location, top and bottom wall separation, and cavity shear layer oscillations. mrDMD was able to identify all of these features as well, and additionally was able to capture the PCST and separation bubbles in their correct spatio-temporal vicinities as the isolator flow-field evolved over time. DMD, and mrDMD to a lesser extent, exhibited difficulty in capturing the travelling shock waves of the PCST in a physically meaningful manner, which is likely due to the methods being based on the Singular Value Decomposition. It was shown that despite these difficulties in capturing moving waves, mrDMD clearly resolved spatio-temporal coherent structures in the isolator that DMD completely failed to. However, mrDMD may require the snapshots to have finer spatiotemporal resolution, to allow for more than the current 4-level analysis, to truly reveal its benefits, such as detecting and controlling unstart.||en_US|
|dc.publisher||The Ohio State University||en_US|
|dc.relation.ispartofseries||The Ohio State University. Department of Mechanical and Aerospace Engineering Honors Theses; 2021||en_US|
|dc.title||Application of Data-Driven Modal Decomposition Techniques to the Non-Stationary Case of Scramjet Unstart||en_US|
|dc.description.academicmajor||Academic Major: Aerospace Engineering||en_US|
Files in this item
Items in Knowledge Bank are protected by copyright, with all rights reserved, unless otherwise indicated.