Data Mining Vortex Cores Concurrent with Computational Fluid Dynamics Simulations

Abstract

Computational fluid dynamics (CFD) simulations present a variety of data mining challenges. At the forefront, CFD computations can require weeks of computation on expensive high performance clusters, delaying investigation of results until a fully converged solution is obtained. Also, advanced modeling can create large data sets that risk concealing rather than revealing useful flow information. Twentyfirst Century Systems, Inc. and Brigham Young University have been collaborating on a concurrent agent-enabled feature extraction project designed to provide intelligent feedback to researchers while a CFD simulation is executing. This approach can extract flow features while a simulation is running and then project their expected probability for a complete simulation. This article gives a detailed outline of our approach and then shows the results of our implemented approach on two sample CFD data sets. The results show vortex core features can be successfully extracted while a simulation is running and provide information as much as 50 % earlier than waiting for complete simulation convergence.