Evaluation of gas- and particle-phase separation methods for velocity measurements in turbulent multiphase flows

Two previously established and commonly used methods for separating gas and dispersed particle phases in images are evaluated and directly compared for the first time. Median filtering and subtraction (MFS; Kiger and Pan in J Fluids Eng 122:811, 2000) and two-parameter filtering (TPF; Khalitov and Longmire in Exp Fluids 32:252, 2002) are assessed using both synthetic and experimental data over a broad range of operating conditions to identify limitations in the applicability of the phase-separation methods in deriving instantaneous velocity fields, statistics, and derivative quantities such as vorticity. Both methods are applied to synthetic data containing a controlled number of objects (small tracers and large particles) with defined sizes and evaluated to determine how the methods alter the targeted phases in terms of objects lost and the addition of erroneous artifacts. Subsequently, PIV software is used to determine the effects of the phase separation-induced errors on the tracer images in terms of the derived gas-phase velocities. Both methods return results with sub-pixel PIV errors, with MFS slightly more accurate than TPF. The methods also are evaluated with experimental data from a turbulent multiphase jet. Both MFS and TPF reproduce the bulk (mean) gas-phase velocity, but fail to accurately estimate smaller-scale motions, producing mean error in the velocity fields as high as 25–30%. The errors are observed to be largely independent of large particle concentration indicating that the largest source of error is the direct alteration of the tracer images through the phase separation process. Significant errors are noted in instantaneous velocity fields and gradient quantities such as vorticity, with the MFS results being particularly unreliable. While TPF was shown to be more accurate, the processing time for TPF is several thousand times longer than MFS for the experimental images, limiting its applicability. The results indicate that low-order statistics may be recovered using the two phase-separation approaches, but resolving instantaneous turbulent motion of the continuum phase and fluid-particle interactions may not be possible with the phase-separation approaches.