Determination of Erroneous Vectors in a Velocity Field Using Genetic Algorithms

Many measurement methods of velocity vector distributions by image processing have been developed and applied to various flows. Most of them, however, have a serious problem. The estimated velocity vectors in a flow field include some erroneous ones because of mismatching a tracer image pattern to another one. Accordingly, it is necessary to determine the erroneous vectors which exist in the estimated velocity vector distribution. This paper presents a new algorithm for determining the erroneous vectors using Genetic Algorithms, which are search algorithms based on the mechanics of natural selection and natural genetics. With genetic algorithms, the first step of the optimization process is to code the estimated distribution as a binary (0, 1) matrix, where "1" and "0" should correspond to right and erroneous vectors, respectively. Second, many individuals with binary matrices, which have random and different (0, 1) combinations, are prepared as an initial population. Third, genetic operations such as "Reproduction", "Crossover", and "Mutation" , are performed to the prepared candidates and are repeated step by step according to a fitness function which evaluates similarity between neighboring velocity vectors. Finally, an optimal solution of the optimization problem is obtained from a binary (0, 1) matrix with the highest fitness after thousands of generations. The Genetic Algorithms are applied to two velocity vector distributions obtained by computer simulation and image processing respectively, which includes erroneous vectors. After thousands of generations, most of the erroneous vectors are determined and eliminated. As a result, it is proved that this new algorithms can distinguish erroneous vectors from right ones by checking over the entire measured flow field even if the density of erroneous vectors in a flow field is high.