摘要
针对固体发动机烧蚀率的预示,现有传统建模方法存在复杂度高、计算需求大、试验数据少、样本不平衡等问题,提出了一种基于深度卷积神经网络和数据增强的固体发动机烧蚀率预示方法。将传感器数据处理为长度相同、特征相近的序列数据,并使用自适应高斯噪声和随机漂移这2种数据增强方法扩充数据样本,再将扩充后的试验样本和伪样本作为深度卷积神经网络的输入进行训练,将训练得到的模型与传统方法计算得到的烧蚀率预示值进行对比。结果表明,该方法下烧蚀率预示值误差低至0.013 5 m/s,预示精度可达95%。
In order to solve the problem of solid motor ablating rate identification,the existing methods have many shortcomings,such as high complexity,large calculation demand,few test data and sample imbalance.A method of ablating rate indentification based on deep convolutional neural network and data augmentation is proposed.Firstly,the sensor data is processed into sequence data with the same length and similar characteristics.Secondly,two data augmentation methods are applied to expand the data set.Then the augmented data set that is taken as the train set is inputted for the deep convolutional neural network model.Finally,the results of model with the traditional methods are compared with the ablating rate identification tasks.The results show that the mean absolute error of ablating rate using the data augmentation method proposed is as low as 0.0135 m/s.The prediction accuracy can reach 95%.
作者
杨慧欣
项子健
李响
滕英元
YANG Huixin;XIANG Zijian;LI Xiang;TENG Yingyuan(College of Aerospace Engineering,Shenyang Aerospace University,Shenyang 110136,China;Key Laboratory of Education Ministry for Modern Design&Rotor-Bearing System,Xi'an Jiaotong University,Xi'an 710049,China)
出处
《测控技术》
2023年第8期64-70,80,共8页
Measurement & Control Technology
基金
辽宁省科技厅“揭榜挂帅”科技项目(220122127)。
关键词
固体发动机
烧蚀率预示
深度卷积神经网络
数据增强
solid motor
ablating rate identification
deep convolutional neural network
data augmentation
