[ai] physics problems

A flat plate in uniaxial tension with a circular hole. Stress concentrates at the notch, so the peak and the field around the hole are well understood and easy to check. We use surrogates to predict that stress response when geometry and load change: hole size, thickness, plate dimensions, hole position, and applied tension. Same physics problem across design spaces. Miss the hot spot and you notice fast.

A fixed airfoil profile in steady freestream flow, with angle of attack as the design knob. Surrogates predict lift and drag for performance, plus the surface pressure coefficient (Cp) field that loads the wing. Mesh and Reynolds number stay fixed, so the hard part is getting forces and Cp right as AoA changes. The smallest step into external aero before a full wing or vehicle.

A full vehicle in frontal impact, built from the open OpenRadioss Neon crash model and the Rescale AI Physics tutorial protocol. Surrogates predict how the body displaces over time as bumper, oil-pan, and pillar thicknesses change. Mesh resolution can differ by dataset. Transient crash response, not a static part study. Miss the fold pattern and the field score falls fast.

A road-car external-aero problem on open DrivAerML geometries: morph length, width, ride height, and related shape knobs, then predict drag and lift plus surface pressure coefficient (Cp) and wall shear on the native mesh. Ground truth is upstream CFD, not a re-run here. Full-scale car aero. Miss the wake or underbody loading and the field metrics show it.

A high-lift aircraft external-aero problem on open HiLiftAeroML geometries: change angle of attack plus flap and slat settings, then predict drag, lift, and pitching moment plus surface pressure coefficient (Cp) and wall shear on the aircraft skin. Ground truth is upstream CFD, not a re-run here. Landing and takeoff aero. Miss separation or stall and the lift curve and field metrics show it.