Robust genetic algorithm for high-accuracy particle position estimation in three-dimensional particle image velocimetry applications

Abstract

A method for three-dimensional particle position estimation employed in particle image velocimetry applications is presented. The method includes the application of a robust optimization process, which involves the use of a genetic algorithm. The algorithm derives particle position by pattern matching theoretical to experimental images, using the concept of image peak signal-to-noise ratio as the objective error measure for this comparison. To produce sufficiently accurate theoretical images comparable to experimental images for positioning purposes, it is found that a Lorenz-Mie treatment of the seeding scattering field was required, which also took into consideration the incident wavefront. The use of a genetic algorithm for positioning proved to be more accurate and faster than a Nelder-Mead algorithm combined with neural nets used previously. This method has also been shown to be an effective means of isolating contaminant particles in velocimetry images, which can sub- stantially increase the overall error. We discuss some aspects of the theory regarding this method, illustrate the ideas with a simple experi- mental image, as well as detail our implementation of a pattern-matching approach combined with a genetic algorithm for positioning purposes. © 2003 Society of Photo-Optical Instrumentation Engineers. (DOI: 10.1117/1.1533038) Subject terms: diffraction pattern; genetic algorithms; particle position estimation; tunneling velocimetry; particle image velocimetry.