Michael Hohmann

Institute for Artificial Intelligence, Helmut-Schmidt-University, Hamburg, Germany, D PKW benchmark dataset is provided, comprising thousands of parametrically generated geometries, dense CFD simulations per design, and multiple exported representations such as watertight meshes, point clouds, low-dimensional descriptors, and discharge labels. • Extended nomenclature and reproducible generation An extended PKW parameterization is provided that covers trapezoidal plan views and sidewall effects, together with feasibility constraints and a fully reproducible geometry-generation pipeline. • Benchmark tasks and baselines Standardized regression tasks for predicting cDc_{D} coefficients and reconstructing rating curves, including out-of-distribution splits that reflect real design scenarios are defined. Strong baselines using parametric ML and geometric deep learning are reported. The WeirNet dataset (CC BY-NC 4.0) and the full training and evaluation code is available at, Helmut-Schmidt-University