关键词: 飞机结冰, 图像预测, 气动特性, Transformer, 卷积神经网络

Abstract: To address the problem of limited accuracy in single-task learning for airfoil icing prediction and the difficulty in simultane-ously obtaining icing morphology and aerodynamic characteristics, a multi-task learning based airfoil icing prediction framework called MTAF (Multi-Task Airfoil Former) is proposed. The framework employs an encoder-decoder architecture that extracts deep representations of airfoil geometry and flight conditions through a shared feature encoder, realizes feature fusion and task interaction via a task correlation module, and accomplishes icing prediction and aerodynamic coefficient prediction respectively through dedicated prediction heads. The performance of Transformer and ConvNeXt backbone net-works within this framework is comparatively studied. Based on training and testing with 625 NACA 2412 airfoil samples, results demonstrate that the Transformer architecture achieves 99.79% pixel accuracy and 0.9906 coefficient of determination in icing prediction tasks, significantly outperforming ConvNeXt's 99.50% and 0.9772. For aerodynamic coefficient predic-tion, Transformer achieves coefficient of determination values of 0.9754 for lift coefficient and 0.9874 for drag coefficient, which are comparable to ConvNeXt's 0.9733 and 0.9901 respectively, with each architecture showing distinct advantages. In addition, generalization was evaluated using 100 external cases excluded from training. The results show that the proposed MTAF framework delivers high prediction accuracy and robust generalization across a wide range of icing types and flight conditions at both pixel and geometric levels,providing a high-precision prediction method for airfoil icing engineering ap-plications.

Key words: aircraft icing, image prediction, aerodynamic characteristics, Transformer, convolutional neural network