Automatic Design of Turbomachinery Blading Using GPU Accelerated Adjoint Compressible Flow Analysis
This work presents the development of an Automatic Design Optimization tool, with the declared objective that it be actually practical in the context of aerodynamic design of turbomachinery components. For that, the requirements are: that it solves a realistic design problem fulfilling stringent quality criteria, that the results can be readily integrated in daily workflow, that the turnaround times are faster than conventional human driven designs, and that is robust enough that is does not
... that is does not need human intervention once the procedure is initiated. The starting point has been the existence of a set of validated design tools used routinely in the usual human driven process, comprising geometry generation, flow analysis, and solution postprocessing tools, developed at the Tecnology & Methods department at Industria de TurboPropulsores S.A. Initial conceptual studies and development of an adjoint flow solver (integral part of a sensitivity calculation methodology) were performed by Fernando Gisbert in his doctoral thesis . During the course of this thesis, these design tools have been interfaced in a seamless manner to build a fully automatic chain for airfoil geometry definition and evaluation in terms of thermodynamic efficiency and manufacturability. The result is that the output of this chain can be used by an external optimization algorithm to propose a high performance geometry, without more human input than that of the specification of the design problem. Regarding this issue, routine industrial design often involves an number of informal or implicit criteria. An effort has been done to bring these to light so that they can be translated to algorithmic language. Critical stages of the geometry generation and analysis have been accelerated by the use of general purpose GPU computing, achieving very low turnaround times. For that, the 7 relevant computer science knowledge has been developed and is presented. Results of different design exercises carried out at different stages of development are provided, illustrating the improvements in speed and capabilities of the growing environment. At its current state, turbomachinery components with a quality comparable to that of a human design with strict requirements can be generated in a fraction of the time. Por último, gracias a los miembros del tribunal por aceptar a valorar este trabajo.