Inverse Design Study of Broad-Crested Weirs

Key Points

• This research aims to develop a differentiable optimization model for broad-crested weirs to control downstream water levels accurately.
• Utilized incompressible fluid simulations to optimize weir geometry parameters.
• Constructed a forward solver using Navier-Stokes equations and developed a loss function based on flow velocity differences.
• Applied gradient descent for iterative updates to weir shapes, confirming convergence and consistency with numerical experiments.
• The model demonstrated an MSE of 0.15092 using conjugate gradient method against traditional methods that had an MSE of 1.23497, achieving up to 87.78% improvement.
• Results with automatic differentiation showed an MSE of 0.31347, confirming the effectiveness of the proposed model.