Recent advances in unstructured grids and solution algorithms, coupled with current parallel computing hardware and software, have enabled unsteady three-dimensional flow simulations for complex geometric configurations at much higher resolution and fidelity than was feasible in the past. Even in this environment, however, full stage simulations in turbomachinery applications can require upwards of∼ 109 grid points for adequate resolution of full-wheel domains. The aim of this project is to reduce the overall point count for rotor/stator (turbomachinery stage) simulations through the use of axisymmetric boundary conditions and demonstrate a method for rotor/stator coupling in the context of multielement unstructured meshes. Through the reduction in point count, time to solution can be drastically reduced, and/or the fidelity of the simulation can be increased. In this work, it is shown that stage performance mappings become feasible in a reasonable amount of computational time.

Parallel capability for unstructured solvers has become a key staple in their usability; at the current time, unless the solver code is written for parallel usage, it has very limited applicability toward any real-world problems. As such, any approach toward implementing domain coupling and/or axisymmetric boundary conditions must be general enough to readily parallelize and scale to tens of thousands of processors. In this work, the approach for both rotor/stator domain coupling and axisymmetric boundary conditions is such that the subdomain decomposition (and thus, the load balancing) can proceed without modification. For handling bodies in relative motion, a number of procedures have been popularized within the CFD community. The Chimera approach has been used for over thirty years in structured grid CFD, 2 and has, much more recently, been extended for use in unstructured meshes, 3, 4 most popularly through the use of technology available via SUGGAR5 and DIRTlib. 6 The overhead in performing the required cutting and interpolation is quite substantial, however, and such approaches still suffer from scalability problems; typically, the package performing the cutting is not parallel at all, which introduces a serious limitation in the problem sizes that can be reasonably handled via this approach. Also, remeshing7, 8, 9 is a reasonably popular approach; holes are cut in the grid, and the grid is locally remeshed in order to rejoin the moving body with the surrounding grid. Given a limited amount of relative motion, mesh movement via a linear or torsional spring analogy is also useful. A paper comparing overset methods and unstructured remeshing is McClung. 10