Application of the multiblock method in computational aerodynamics

[摘要] The main challenge in computational aerodynamics is to provide practical, credible, cost and schedule effective methods for routine design application and for full integration of these methods into the design cycle. Although advances in physical modelling and solution algorithms are continuing requirements of the aerospace industry, other more practical difficulties also impede the full realisation of the potential of existing methods. The contribution of this thesis is to examine and tackle several of these issues and to evaluate computational aerodynamics as a tool for engineering design and scientific enquiry. An advanced computational aerodynamics method is evaluated as an engineering tool for axisymmetric forebody and base flow problems. First the adaption of an existing two-dimensional flow solver to axisymmetric flow is described, then specific test cases are considered. The motivation for creating an axisymmetric flow solver is the considerable performance improvements compared to a fully three-dimensional method. The accuracy and robustness of the method are very good for forebody problems. For base flow problems accuracy and robustness are less satisfactory, although the performance of other prediction methods is also poorer for this more demanding problem. For both problem types the speed of the flow solver, the required computing resource and the time and effort necessary for pre- and post-processing are all satisfactory for routine calculation in an engineering environment. Shock reflection hysteresis and plume structure in a low density, axisymmetric highly underexpanded air jet is examined using a Navier-Stokes flow solver. This type of jet is found in a number of applications e.g. rocket exhausts and fuel injectors. The plume structure is complex, involving the interaction of several flow features, making this a demanding problem. Two types of shock reflection appear to occur in the plume, regular and Mach, depending on the jet pressure ratio. The existence of a dual solution domain where either type may occur has been predicted, in agreement with experiment where the same phenomenon has been observed for a nitrogen jet. There is a hysteresis in the shock reflection type; the reflection type observed in the dual solution domain depends on the time history of the plume development. A quasi-steady approach is employed in order to calculate the entire hysteresis loop. The results of the computational study are used to examine the structure of the plume, and are compared with experimental data where possible. Some flow features not initially recognised from experiment have been identified, notably curvature of the Mach disc, recirculation behind the Mach disc and the 'regular' reflection having Mach reflection characteristics. Included in the study is a review of the two dimensional shock reflection hysteresis problem to establish a theoretical background. The value of CFD as a tool for scientific investigation is clearly demonstrated by this study. The need for automation of the multiblock grid generation process is discussed. A new approach to automatically process a multiblock topology in order to prepare it for the grid generation process is described. The method is based on a cost function which attempts to model the objectives of the skilled grid generation software user who at present performs the task of block positioning and shaping in an interactive manner. A number of test cases are examined. It is also suggested that an existing unstructured mesh generation method could be adopted as an initial topology generation tool. Further work towards creating a fully automatic grid generation tool and extension into three dimensions are discussed. The parallel execution of an aerodynamic simulation code on a non-dedicated, heterogeneous cluster of workstations is examined. This type of facility is commonly available to CFD developers and users in academia, industry and government laboratories and is attractive in terms of cost for CFD simulations. However, practical considerations appear at present to discourage widespread adoption of this technology. The main obstacles to achieving an efficient, robust parallel CFD capability in a demanding multi-user environment are investigated. A static load-balancing method is described which takes account of varying processor speeds. A dynamic re-allocation method to account for varying processor loads has been developed. Use of proprietary software has facilitated the implementation of the method.

[发布机构] University:University of Glasgow

[关键词] Aerospace engineering