基于FBG传感器和卷积神经网络的复合材料结构载荷识别研究

Load Identification of Composite Structural Based on FBG Sensor and Convolutional Neural Network

复合材料因其优异的性能被广泛应用于飞行器结构(如飞机机翼)中,对作用在复合材料上的载荷进行识别为飞行器的结构设计和可靠性分析提供了重要的保证,在结构的工程应用中具有很高的价值。本工作研究了一种基于光纤布拉格光栅(FBG)传感器和卷积神经网络(CNN)的多点复杂载荷识别方法。在复合材料悬臂梁上布设FBG传感器,利用FBG实际测量得到的应变数据,首先通过支持向量机(SVM)算法对施加载荷的个数进行识别。进一步根据测点排列顺序,再将应变数据转化成矩形图片,经过归一化处理后输入CNN中,实现同时施加多个载荷时载荷的定位和定量,并与反向传播神经网络(BPNN)和梯度提升决策树(GBDT)的预测结果对比。SVM模型识别准确率为99.584%,CNN模型对两点载荷施加位置预测的平均绝对误差(MAE)分别为0.6379 mm和0.5762 mm。结果表明,基于SVM和CNN的多点载荷识别方法是一种有效的方法,可以稳定、准确地识别载荷个数、载荷施加的位置和大小,为飞行器飞行载荷测量提供新的解决方案。

Composite materials have been widely utilized in aircraft structures such as aircraft wings because of excellent performance.Therefore,recognizing the loads acting on composite materials has a significant influence on the structure design and reliability analysis of aircraft.In this work,a method based on fiber Bragg grating(FBG)sensor and convolutional neural network(CNN)was utilized to recognize the complex multi-point load.The experiment was carried out on a composite cantilever beam and the strain data was collected by sensors placed on the structure.Firstly,the strain data was input into the support vector machine(SVM)algorithm to identify the number of applied loads.Furthermore,the strain data was transformed into a rectangular picture according to the arrangement of the measuring points.After the normalization process,the picture was input into the CNN model to realize the positioning and quantification of the multi-point load.Finally,the results of CNN model were compared with the results of back-propagation neural network(BPNN)and gradient boosting decision tree(GBDT),respectively.In the experiment,the recognition accuracy of the SVM model was 99.584%.For the position of the two-point load,the mean absolute error(MAE)of the CNN model is 0.6379 mm and 0.5762 mm.The results indicated that the load identification method based on SVM and CNN is an effective method to identify the number of loads,the position and size of the loads applied.Additionally,the method is capable of providing a new solution for aircraft flight loads measurement.