OR1ON-EIRA: Physics-Informed Neural Networks for Navier-Stokes Discovery — GPU+Pi5 Distributed Stack

Key Points

• The central aim is to evaluate OR1ON-EIRA's effectiveness in solving Navier-Stokes equations using advanced computational techniques.
• Employed Physics-Informed Neural Networks (PINNs) to solve 2D incompressible Navier-Stokes equations.
• Utilized a distributed stack comprising RTX 3060 GPU and Raspberry Pi 5.
• Conducted an honest assessment of the capabilities, comparing GPU versus CPU performance.
• Achieved continuity RMS of 0.0002 in Cavity Flow, significantly better than the threshold.
• Discovered Reynolds number with 0% error using inverse PINN.
• Demonstrated a speedup of 23 times with GPU compared to CPU.
• Produced R² of 0.9994 for SINDy Lorenz model.