A Comparative Metamodel Based Shape Optimization Study for Maximizing Thrust of a Helicopter Rotor Blade Under a Torque Constraint

The solution of Reynolds-Averaged Navier-Stokes (RANS) equations is crucial for accurately predicting the aerodynamic loads on helicopter rotor blades. In particular, the computational process required for blade shape optimization, involving numerous RANS solutions, is highly time-consuming. To reduce this computational cost, a recently adopted approach is the use of metamodels, such as machine learning methods. A well-established metamodel is expected to successfully replicate CFD solutions. In this study, different machine learning techniques were employed as metamodels and evaluated based on a series of CFD solutions. The machine learning models aimed to capture the functional relationship between the generated thrust and torque and the twist distribution along the rotor blade. The smooth twist variation was modelled using a 3-knot cubic spline, with five parameters serving as inputs for the spline definition. The optimal twist distribution was determined concerning a reference helicopter rotor blade, the Caradonna-Tung rotor blade. The optimization scenarios were defined to maximize thrust force while maintaining the baseline torque value. The optimal cases were identified using the Quadratic Response Surface Method, Support Vector Regression, and Artificial Neural Network Regression. As a result of this study, a significant increase in the thrust force generated by the helicopter rotor blade was observed.