A wing near the ground gets its own simulation set
770 airflow simulations aim to test AI shortcuts for racing aerodynamics, though the download link is not yet live.
Picture made for ABUREAU.
Engineers use AI models as fast stand-ins for slow airflow simulations. They need hard test sets to judge them. A paper from 3 November describes one for a wing working close to the ground, a case missing from other wing datasets.
The set holds 770 simulations of 194 wing shapes, with files for the surface, the full airflow and a table of forces. The paper reports about 38,300 core hours of computing to make it, which shows why the data itself has value. Six AI designs are benchmarked, and they lose accuracy outside the range they were trained on.
The paper is open about gaps. Six runs failed and were not repeated, only flat ride heights are covered, and one proprietary profile is left out. The dataset link is marked to be confirmed, and the dataset's licence is not stated, so nobody should assume reuse terms.
For people who make data, this is a clear case. Costly simulation output becomes more useful when it comes with a benchmark, a stated envelope and clear terms. Without a live link and licence, a buyer cannot yet use it.
