Inverse Design Method using Classification and Regression Networks for Internal Structure of Rotor Blade Cross-section

In this study, a deep learning-based surrogate model for inverse design of composite rotor blade cross-sections is developed. The surrogate model classifies the section shape type and predicts detailed structural geometry of the cross-section by considering target cross-sectional properties during the early stages of the design process as the input of the network. A graph neural network(GNN) is utilized to classify the types of cross-sectional shape, such as spar types and the presence of structural components within the cross-section. Based on the classification results, a multi-layer perceptron(MLP) is employed to predict the positions and thicknesses of the structural components. The surrogate model is trained and validated using analysis results from the cross-sectional analysis program VABS(Variational Asymptotic Beam Sectional analysis). Moreover, the resulting classification and prediction performance is evaluated using the cross-section of main rotor blade for the multi-purpose unmanned helicopter(MPUH) developed by the Korea Aerospace Research Institute(KARI).