AI Tool BladeSDF Revolutionizes Turbine Blade Design with Precision
Designing turbine blades presents a tricky balance between performance and manufacturability. Engineers must create shapes that maximise efficiency while remaining practical to produce. A new generative modelling tool, called BladeSDF, now offers a solution to this long-standing challenge.
BladeSDF works by using Signed Distance Functions (SDFs) and a framework inspired by DeepSDF. Each blade design is first converted into a point cloud, with a convex-hull proxy helping to define inside and outside boundaries. This approach allows the system to calculate precise distance measurements for refining shapes.
The framework directly incorporates performance metrics into the design process. Instead of treating aerodynamics and manufacturability as separate concerns, BladeSDF explores controlled variations using interpretable parameters like taper and chord ratios. This ensures that generated designs remain both efficient and feasible to manufacture.
Training the model involved a dataset of 222 existing blade designs, each contributing 20,000 labelled SDF pairs. The system used Adam optimisation with an initial step size of 10^-3, refining both the decoder and latent codes simultaneously. A clamped reconstruction loss with a quadratic latent prior helped fine-tune the results.
At test time, the decoder stays fixed while a new code is optimised to match SDF observations. This demonstrates how well the model has learned to represent complex blade geometries. The boundary of each design’s convex hull, defined as a triangular mesh, further refines the SDF ground truth within a truncation band.
BladeSDF enables data-driven concept generation, reducing the time and effort needed for turbine blade design. By embedding performance criteria early in the process, it produces shapes that meet both aerodynamic and manufacturing demands. The framework’s ability to handle interpretable parameters could also simplify future design iterations.
