军事知识服务平台
基于图像驱动机器学习方法的微型扑翼升力预测
摘要 本文融合图像识别、有限元分析和机器学习方法,提出了一种用于生物启发扑翼微型飞行器(FWMAVs)的升力预测方法.FWMAVs具有灵活机动的优势,其升力受气动力载荷下复杂结构变形的影响.与刚性翼模型不同,该方法显式引入了昆虫仿生翼固有柔性及肋-膜耦合特性.本文构建了基于U-Net的分割模型,从膜翅目昆虫翅膀图像中自动提取翼型轮廓和翅脉分布,为刚度分析提供了精确的形态学特征.采用了改进的MobileNetV2网络结构,从结构图像中估算弯曲和扭转刚度,实现了高效的刚度预测.在此基础上,提出了考虑刚度的升力预测模型,可反映变形引起的扑动角和攻角变化.实验验证结果表明,该模型相较于传统刚性翼模型可降低69.72%的预测误差,平均预测精度达到89.65%.本文为可变形翼的升力预测提供了一种可扩展且具有普适性的途径,为轻量化、高性能FWMAVs设计与优化提供了指导.
Abstract:
This paper presents an integrated framework for lift prediction in bio-inspired flapping-wing micro air vehicles(FWMAVs)by synergistically combining image recognition,finite element analysis,and machine learning.FWMAVs offer advantages in agility and maneuverability,yet their lift generation remains limited due to complex structural deformation under aerodynamic loads.Unlike rigid-wing models that neglect the intrinsic flexibility and rib-membrane coupling of insect-inspired wings,our method explicitly incorporates these structural characteristics.A U-Net-based segmentation model is developed to automatically extract wing contours and rib distributions from Hymenoptera wing images,providing accurate morphological features for stiffness analysis.Furthermore,a modified MobileNetV2 architecture is employed to directly estimate bending and torsional stiffness from structural images,enabling efficient stiffness prediction.A stiffness-aware lift prediction model is subsequently proposed,accounting for deformation-induced variations in flapping and angle of attack.Experimental validation demonstrates that the proposed framework reduces prediction error by 69.72%compared to conventional rigid-wing models,achieving an average accuracy of 89.65%.This study provides a scalable and generalizable approach for lift prediction in deformable wings and offers practical insights for the design and optimization of lightweight,high-performance FWMAVs.
刊 名 力学学报(英文版) ISTICEISCI
年,卷(期): 2026, 42(7)
机标分类号: V221.7 U661.1 TP391
在线出版日期: 2026年8月18日