Evolutionary optimization of scalar transport in cylinder arrays on multiGPU/multicore architectures
Diego Rossinelli, Laurent Chatagny and Petros Koumoutsakos,
Abstract. We optimize a scalar transported by the flow field past an array of cylinders as a model of feeding behavior in swimming schools. We use an evolutionary optimization approach to identify cylinder arrangements that maximize the scalar field in the vicinity of twelve cylinders at ReD = 100. The large number of evaluations required by the Evolutionary Algorithms are compensated by GPU/multicore implementation of particle based flow solvers. We use an FTLE analysis of the flow field to quantify optimal transport and discuss the physical mechanisms associated with optimal configurations.