Auto-Processing Of Filtered Rayleigh Scattering Images Including Mie And Background Scattering Contributions

Abstract

Filtered Rayleigh scattering (FRS) is a non-intrusive, optical-based technique that allows for simultaneous, time-averaged, measurements of three-component velocity, static temperature, and static pressure. The method of post-processing the raw images to get these variables is a non-trivial task. The post-processing scheme starts with building a model to generate simulated spectra given known velocity, temperature, and pressure values which can be iterated upon. The model also includes Mie scattering and background scattering contributions that must be taken into account. This iteration scheme allows for the simulated spectra to be matched to the experimental spectra to back out the desired flow variables. This iteration process can be lengthy and so it is important to make spectrum generation/iteration as fast as possible. This is done through the use of support vector spectrum approximation (SVSA), which is a machine learning algorithm, as well as a multivariable minimum error solver based on the least-squares fit between the simulated data and the experimental data. To prove the methodology, simulated results were first compared at different signal-to-noise ratios to determine the expected uncertainty at each SNR level. It was shown that to achieve an error of less than 1\% in all variables (velocity, static temperature, Mie intensity ratio, and background intensity ratio), the SNR must be near 40dB. This kind of SNR level is generally not expected in an experiment. Therefore, a representative experiment was conducted which included Mie and background scattering contributions. It was shown that manually processed data was capable of achieving velocity results that were within 2\% of the probe data. No other variables were achieved. The auto-processing scheme was able to achieve an error of approximately 2.5\% in velocity when compared to probe data. It was also capable of providing static temperature, Mie intensity ratio, and background intensity ratio values. The manual iteration results took several days to achieve while the auto-processing took about an hour. It is clear that this methodology is capable of automatically processing images from experiments with Mie and background scattering contributions while saving time.