基于神经网络的柔性翼肋变形自感知方法研究  

Research on Self-sensing Method of Flexible Rib Deformation Based on Neural Network

作　　者：刘丁嘉 杨睿[1] Liu Dingjia;Yang Rui(School of Mechanical Engineering,Dalian University of Technology,Dalian,Liaoning 116024,China)

机构地区：[1]大连理工大学机械工程学院,辽宁大连116024

出　　处：《机电工程技术》2023年第1期129-132,共4页Mechanical & Electrical Engineering Technology

摘　　要：为了准确反映出柔性翼肋在飞行过程中的变形情况,提出了一种基于神经网络技术的柔性变体翼肋主动变形自感知方法。参照BP神经网络模型搭建神经网络结构,然后选取Leaky ReLU作为神经元激活函数。通过有限元软件仿真得到柔性翼肋上选定点处的应力数据和整个柔性翼肋的位移数据,然后将得到的数据划分为训练集、验证集和测试集。选取神经网络在训练过程中需要的Adam学习算法和drop out过拟合抑制方法,然后用训练集上的数据对已搭建好的神经网络进行训练。训练完成后的神经网络经过验证集数据的验证,不存在明显的过拟合现象,可以用于柔性翼肋变形的自感知。利用测试集上的数据评价所提出的方法的自感知精度,该方法能够在使用较少应变传感器的情况下达到足够的自感知效果。In order to accurately reflect the deformation of the flexible wing rib during flight, a self-sensing method of active deformation of flexible wing rib based on neural network technology was proposed. The neural network structure was built with reference to the BP neural network model, and then Leaky ReLU was selected as the neuron activation function. The stress data at selected points on the flexible rib and the displacement data of the entire flexible rib were obtained through finite element software simulation, and then the obtained data were divided into training set, validation set and test set. The Adam learning algorithm and drop out overfitting suppression method required in the training process of the neural network were selected, and then the data on the training set to train the built neural network was used. After training, the neural network was verified by the validation set data, and there was no obvious over-fitting phenomenon, which could be used for the self-perception of the deformation of the flexible rib. The data on the test set is used to evaluate the self-sensing accuracy of the proposed method, which can achieve sufficient self-sensing effect with less strain sensors.

关 键 词：神经网络 变体飞机 柔性翼肋 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]