Intelligent Boundary Conditions for Blade Film Cooling CFD using Machine Learning

The turbine blades within a gas turbine engine require cooling in order to avoid damage from the high temperatures generated by the combustion process. This is usually achieved by routing cooler gas through the blade interior, and out through small holes in its surface. Engineering of the cooling holes often uses CFD simulations with simplified geometry to help accelerate the design process. However, removing the blade interior and the hole channels from the mesh, and replacing them with inlets on the blade surface, can reduce the fidelity of the results because secondary flow patterns develop within the channels that have a significant effect on the cooling gas as it emerges. It is proposed that ‘Intelligent’ boundary conditions be used at the hole surface inlet, which can reproduce these secondary flow effects. A simple Machine-Learning Network was created using TensorFlow and trained using a set of results derived from detailed CFD simulations of the complete geometry. The geometry (specifically the drilling angle and length), together with the non-dimensional flow conditions of Blowing Ratio and Freestream Mach number, was varied across the set within realistic ranges. The trained network was then integrated into Ansys Fluent using a UDF. Within the subsequent simplified CFD, the input characteristics were fed into the Network model. The boundary conditions of the hole inlet were then dynamically updated to give an appropriate non-uniform distribution of massflow and flow direction. A conservative mapping was implemented between the Network solution and the inlet mesh to reduce the sensitivity to that mesh. This allows the engineer to smoothly transition between coarse early-stage simulations, and the more detailed latter-stage simulations of the design cycle, with increasingly accurate and appropriate hole conditions. Comparisons showed good agreement between the original detailed geometry CFD and the simplified solution with this novel boundary condition.