Starting from the 3D inverse design method, we identify the key pillars of a Machine Learning system for turbomachinery design and optimization. We explain how we can exploit a powerful but manageable set of performance-based parameters to create a wide range of complex 3D blade shapes covering a vast design space.
Recognising that high-fidelity simulation will always be required to evaluate complex flow fields, especially when aiming for objectives across multiple operating points, the blade design space is explored and managed by a reactive optimization and search algorithm that judiciously applies high-fidelity simulation only where and when it is required.
Simulation data management is shown as a simple, integrated process, where connections between the geometry-optimizer engine and the simulation tools are managed in a couple of mouse clicks.
