Development and Application of Transition Prediction Techniques in an Unstructured CFD Code

Abstract

For some time computational fluid dynamics (CFD) based numerical simulations are an essential component in the industrial design process of aircraft. For many aircraft configurations it is possible to obtain highly accurate and reliable results using current CFD methods if the simulations are carried out for design point applications. The ever increasing capabilities of high-performance computing (HPC) systems have led to the vision of the ‘digital aircraft’ which can be flown in the computer while it is carrying out unsteady maneuvers. The keyword ‘flying the equations’ is a strongly condensed wording of the idea to execute a highly coupled simulation that, at the same time, incorporates the effects of flight mechanics, the structural deformation of the aircraft and the flow physics, the latter via high-fidelity CFD simulations in a time-accurate manner. First steps towards a highly multi-disciplinary simulation system being the prerequisite for such a coupled simulation are currently done in current research and development projects, such as the DLR project Digital-X [1], in order to support aircraft design and analysis based on a much higher number of numerical simulation results than today. In so doing, the future development and testing of completely new configurations based on highly accurate simulation data within the full flight envelope of an aircraft shall be made possible. At present, it seems that the successful execution of industrial-like Reynolds-averaged Navier-Stokes (RANS) computations of large full-aircraft configurations of highest geometrical complexity, the incorporation of more and more geometrical details and, as a result, the corresponding grid densities and point numbers can be achieved some day from a technical point of view if only the computational resources in terms of memory and processor cores of HPC clusters are large enough. Grid generation techniques and tools are either available today or under development so that appropriate computational grids can be generated exploiting, for example, sliding meshes, overlapping (chimera) grids, or the incorporation of large hexahedral portions within an non-hexahedral remainder of a purely unstructured grid composed of arbitrary cell types. Even the high demands made by the large-scale unsteady effects of maneuvers or the low-frequency unsteadiness that can exist at the borders of the flight envelope can be satisfied, partially by the ongoing development and improvement of numerical algorithms for time-accurate computations or by hybrid parallelization strategies combining classical domain decomposition with multi-threaded processing of the data on each domain [2]. Thus, accurate time-dependent flow solutions for very large aircraft configurations seem to be within reach. The predominant majority of simulations for design point applications is done for steady flows and based on standard RANS turbulence models. Although a number of Reynolds stress models (RSM) [3-7], that are considered to represent the highest level of RANS modeling for practical use, are available, in most CFD simulations one-equation or two-equation eddy viscosity models (EVM) are used in fully-turbulent computations. A major obstacle that hinders RANS-based CFD to yield the desired accuracy of results at the borders of the flight envelope for most cases is the physical models. Many of the crucial physical phenomena in transport aircraft flows in these flight regimes are characterized by strong non-linearities as, for example, flow separation and reattachment, shock/boundary-layer interaction, free vortices, wakes, and free shear layers. For some of them the correct boundary-layer representation in the numerical flow solution is of highest importance. In order to catch these phenomena correctly two modeling areas are of highest importance: turbulence models and laminar-turbulent transition models [8], and the interaction between the two. While for turbulence models one can be skeptical if their predictive capabilities can be improved and, at the same time, a reasonable trade-off between computational effort and accuracy of results can be achieved for situations at the borders of the flight envelope the situation is different for laminar-turbulent transition models.