Researchers release aircraft airflow dataset with 1,800 simulations

Researchers have released HiLiftAeroML, an open dataset of 1,800 simulations designed to help train and test models of aircraft airflow during takeoff and landing. The preprint covers 180 aircraft configurations at ten angles of attack, including conditions in which airflow separates from the wing. The authors report that reference simulations agree well with wind-tunnel measurements of lift and pressure. The release includes aircraft shapes, airflow data, validation material and fixed benchmark splits. Initial models reconstructed several test cases well, but struggled more with separated airflow at high angles and conditions outside their training data.