Session: 7.6 - Experimental methods for multiphase flows

Paper Number: 157939

157939 - Imaging Techniques for Determining Entrained Droplet Size Distributions From a Paraffin Slab Under a Heated, High Shear Environment 

Abstract: 

Liquefying hybrid rocket fuels have gained interest due to having higher regression rates than classical gasifying fuels.  This phenomenon was correlated to mass entrainment of the liquid melt layer in the 90s, causing the motor to have fuel injector behavior.  While studies have been done on melt layer instabilities, there have been no diagnostics for measuring the droplet distributions for a melt layer under a high shear environment.  Additionally, other work has stated liquefying fuels can be in a critical state, depending on the motor conditions, showing there is a limited understanding in the behavior of these fuels.  This study uses an optical chamber with heated air at low pressure to melt wax and create a high shear environment for mass entrainment to occur.  Air fluxes ranged from 42.8 – 123 kg/m²s with temperatures of 93.2 – 190.8 °C.  Shadow photography is used to capture droplets using an Edgertronic SC1 high speed camera.  Due to the optical chamber environment, it is necessary to process noise out of the images due to wax deposits on the windows.  Additionally, since droplet sizes have not been measured in this environment before, it is important to define the best automated method for determining the droplet size distributions for this experimental setup.  Therefore, four image processing methods for determining droplet size distributions within a melting slab experiment are presented: hand-drawing (``truth"), moving image subtraction with a size filter (MISSF), a fast radial transform (RT), and a U-Net convolution neural network (CNN).  This is confirmed by the droplet distributions following similar trends for spray nozzles in the literature where the droplet size decreases with increasing gas flux with the greatest sensitivity being shown in the D<sub>0.9</sub> value.  Standard normal distributions of the droplets are shown to trend towards a single distribution where only the trailing edge changes due to the maximum droplet size dependence on the oxidizer flux.  Additionally, it is shown that the U-net is superior to the other methods considered for tracking the size distributions. The associated error for the U-net is 2.86 % error from the mean of the ``truth".  For the standard deviation of the size distribution, the RT and U-Net methods are similar in error with 8.81 % error vs 9.01 % error, respectively, where MISSF deviated more with 11.0 % error. Likewise, for the span, the errors are 6.23 %, 5.74 %, and 15.4 % for the U-net, RT, and MISSF, respectively. 

Presenting Author: Elektra Katz Ismael SUNY Buffalo

Presenting Author Biography: Elektra Katz Ismael is a PhD student at the University at Buffalo whose work involves designing and managing experiments for the Combustion and Energy Transport (CET) lab. Her research interests are in solid fuels, multiphase flow, materials processing methods, and combustion.