Feature Extraction from High Reynolds Number High-Swirl Flows

Presenter:       Hanna Ek

Committee:    Prof. Tim Lieuwen (advisor), Prof. Jerry Seitzman, Prof. Adam Steinberg

When:             Friday May 17th, from 1:00 – 2:00 pm

Where:            MK 317

Abstract

The understanding of high Reynolds number, high-swirl flows is important due to its widespread use in combustion applications. While these flows have certain key features, including an annular jet, inner and outer recirculation zones, and shear layer disturbances, the topological details depend on a number of factors related to both geometry and operating conditions. Additionally, the flow features interact and couple with flame dynamics and acoustics, resulting in complex spatio-temporal dynamics occurring over a wide range of scales. Due to the level of complexity, we currently do not have a complete understanding of these types of flows. The extensive research within this field, together with recent improvements in computational and experimental combustion, have however led to an increased availability of high-fidelity data sets. Data reduction and analysis therefore become a key aspect in improving our understanding of highly turbulent, swirl-stabilized flows. Since unsteady features do not appear in a time-averaged representation of the flow, but governs many important physical processes, it is necessary to further develop tools targeting spatio-temporal dynamics.

Extraction and analysis of unsteady flow features can be accomplished using either physics-based techniques, data-driven methods, or a combination of the two. Physics-inspired vortex tracking and data-driven modal decompositions, including proper orthogonal decomposition (POD), dynamic mode decomposition (DMD) and some of their extensions, forms the foundation of my work. The understanding of coherent shear layer disturbances is a key aspect for the understanding of many unsteady combustion processes, and therefore motivates the vortex tracking. Additionally, POD and DMD are currently two of the most commonly used modal decompositions techniques within fluid dynamics and combustion. POD has gained its popularity by allowing us to extract spatially coherent, high-energy flow features, while DMD is a purely spectral technique which targets coherent structures through their temporal periodicity. Despite their widespread use, both methods have their limitations, and interpreting the results is often challenging.

The goal of the proposed work is to develop a set of data reduction and analysis tools which together can provide a better, cohesive understanding of the physics governing highly turbulent, swirl-stabilized, reacting flows. To accomplish this, I will use POD, DMD and some of their extensions, while leveraging our current physical understanding. I will also try to incorporate ideas commonly used for data reduction and analysis of high-dimensional data in other fields, such as for example clustering techniques, kernels, geometric distances and/or networks. The tools will target extraction and analysis of unsteady flow structures, and their fuel and flame interactions. They must also allow for a direct comparison of experimental and computational data, supporting the validation process of simulations based on unsteady metrics. Simultaneous, high-speed stereoscopic particle image velocimetry (sPIV), planar laser induced fluorescence (PLIF) measurements of fuel and OH from a liquid-fueled, swirl stabilized dump combustor operated at elevated pressures will be used to demonstrate the capabilities and limitations of these tools.