Department of Engineering
Project Overview Iterative RANS-based CFD design is approaching its practical limits. While high-fidelity simulation remains essential, the repeated geometry-CFD-evaluation loop that dominates turbomachinery design is increasingly the primary bottleneck. Within the next decade, this workflow will be largely replaced by data-driven inverse design models, enabled by the growing availability of high-quality simulation data. This PhD aims to fundamentally change the turbomachinery design process by replacing iterative RANS-based CFD loops with regression-driven inverse design. Instead of repeatedly modifying geometry and re-running CFD, the research will develop models that map aerodynamic requirements directly to 3D blade geometries and associated flow fields. The project builds on a proven 2D inverse design framework, already adopted by major industrial partners including Rolls-Royce, and extends it to full 3D turbomachinery configurations. By leveraging inverse design, the approach deliberately stays within physically sensible design spaces avoiding the need to learn every pathological flow scenario and making machine learning both efficient and reliable. The ultimate goal is to retain the fidelity and robustness of RANS-based design while reducing design iteration times by over 90%.