Optimisation techniques applied to the design of gas turbine blades cooling systems
This study is part of an AVIO project concerning the development of High Pressure Turbine blades with advanced cooling systems. Due to the high gas temperatures entering the turbine of the most recent aero-engines (in general up to 2000 K at the turbine inlet), a very efficient cooling system is required in order to maintain the metal temperatures below the allowable limits. One of the most critical area, from a thermal point of view, is the tip region of the unshrouded rotor blades. As the tip region is characterized by a very complex 3D flow field, it is very difficult to optimize the cooling system using the standard design methodologies, also considering the other blade tip requirements (to minimize the hot leakage air from pressure to suction side, which has a negative impact on turbine aerodynamic efficiency). That is the reason why AVIO decided to introduce a stochastic optimization approach in tip cooling design process. As a consequence of the geometrical complexity of the problem and of the high computational time, the use of the Response Surface Method is almost compulsory if a 3-D fluid-dynamic optimization has to be approached. This way, after a preliminary series of CFD analyses and after the estimation of Neural Nets, the 3-D CFD model can be substituted by a series of mathematical functions and the computational time is considerably reduced.
The CFD model reproduces the entire blade-to-blade channel of an axial un-shrouded turbine rotor stage. A series of cooling holes are explicitly modeled on the tip surface and their distribution is defined by a series of parameters. In order to describe different configurations ICEM-TETRA was used to generate a tetrahedral mesh (with 3 prismatic layers adjacent to the walls) in the tip region. The remaining part of the blade is fixed and was discretized with a hexahedral mesh. The two fluid domains, with two non-matching mesh, were connected in CFX5 using the general grid interface technique (GGI). After the integration of the CFD model in the modeFRONTIER process flow, an initial series of 24 CFD analyses was planned using the SOBOLmethod. The Neural Nets were then coupled to a Multi Objective Genetic Algorithm (MOGA) and the first optimization step was performed with 3 generations of virtual designs. A second set of Neural Nets was estimated using 30 CFD points and the same procedure was followed. Before the last step the problem was reduced to a single objective optimization using the Multi Criteria Decision Making method (MCDM). The last optimization phase was carried out using the SIMPLEX algorithm coupled to the last set of Neural Nets (estimated with 39 CFD points). The comparison of the optimal and the standard configurations gives evidence that the cooling mass-flow can be considerably reduced (-20%), keeping the same wall heat flux on the tip surface.