GPU-accelerated aerodynamic shape optimisation framework for large turbine blades

Abstract

This paper presents a high-fidelity aerodynamic optimisation framework designed to decrease the cost of optimisation of large wind turbine blades. The framework is presented in the context of the IEA 15MW reference turbine, but is applicable to all large turbine geometries. Optimisation is performed using a surrogate model, built through latin hypercube sampling of the design space, with a GPU accelerated CFD code. Aerofoil parameterisation is handled through the use of singular value decomposition of the aerofoil nodes, built on a database of 1300 aerofoils. A preliminary optimisation study is performed on the surrogate model to demonstrate the capability and functionality of such a system.