Quantitative Evaluation and Optimization of Feeding-Back Previous Cycles Information for Computational Fluid Dynamics Periodic Convergence

Abstract

Abstract Unsteady simulations represent an expensive subset of computational fluid dynamics (CFD) simulations in terms of computational resources. Among the unsteady simulations, the time-periodic cases are of special interest in many applications such as turbomachinery. The special characteristics of these cases allow specific methodologies to improve its periodic convergence behavior, neglecting the physical transient to achieve it. The scope of this paper is focused on these unsteady periodic simulations. In this context, departing from a plain Dual Time-Step methodology, where the standard generic procedure is to converge every physical time-step from the solution of the previous one, the effect of feeding-back the previous period information to restart the convergence of the equivalent time-steps in the new period is assessed in this paper. With that aim, options to alleviate the possible penalties of the procedure are presented, and the parameter set election and boundary conditions optimized to define a complete methodology. Finally, a benchmarking of the defined methodology is performed against the standard plain Dual Time-Step technique and the much more efficient Harmonic Balance methodology. For that comparison, a basic two-dimensional (2D) case and a more realistic 3D compressor, both with harmonic torsion vibration, as well as two rotor–stator configurations, are computed and presented here. The results show that a comparable efficiency to the Harmonic Balance technique is achieved with lower memory requirements, meaning that the performance of the plain Dual Time-Step methodology is greatly improved just by changing small parts of its formulation.