A Novel Parameterization Framework and Neural Network-Based Surrogate Modeling for Rotorcraft Fuselage Aerodynamic Prediction.

Designing rotorcraft fuselages in the early stages requires a simple, flexible, and simulationready shape representation that supports detailed analysis and is essential for performing design optimization studies. This work introduces a novel parameterization framework based on four fundamental 1D design curves—camber , height , width , and superellipse exponent . Relevant curves are transformed into Class-Shape Transformation (CST) space for smooth, controllable shaping. Breakpoints along these curves are automatically identified using a genetic algorithm to minimize fitting error, and segments are reconstructed using piecewise cubic Hermite interpolation, enabling local control. Principal Component Analysis (PCA) is applied to the geometric dataset to compress the high-dimensional design space, allowing efficient exploration and structured sampling of feasible fuselage shapes. An automatic quad-dominant surface mesh generator is developed to convert the parametric model into CFD-ready meshes. These are used for aerodynamic analysis and to train neural network–based surrogate models, which leverage the reduced-dimensional representation to provide rapid and accurate aerodynamic predictions for design evaluation and optimization.