郭旭东,宋浏阳,王华庆,徐福健,董作一.基于改进CNN-LSTM的剩余使用寿命预测方法[J].测控技术,2021,40(5):21-26
基于改进CNN-LSTM的剩余使用寿命预测方法
Remaining Useful Life Prediction Method Based on Improved CNN-LSTM
  
DOI:10.19708/j.ckjs.2021.05.004
中文关键词:  剩余使用寿命(RUL)预测  长短时记忆(LSTM)网络  航空发动机  深度学习
英文关键词:remaining useful life (RUL) prediction  long short-term memory network (LSTM)  aeroengine  deep learning
基金项目:国家科技重大专项(2017-I-0006-0007)
作者单位
郭旭东 北京化工大学 机电工程学院 
宋浏阳 北京化工大学 机电工程学院 
王华庆 北京化工大学 机电工程学院 
徐福健 中国航发湖南动力机械研究所 
董作一 中化化肥有限公司 
摘要点击次数: 44
全文下载次数: 15
中文摘要:
      机电设备的寿命预测是状态维修中的一项重要任务。剩余使用寿命(RUL)预测不仅可以有效地防止机械装备发生突发性故障,而且可以最大限度地利用装备的工作能力、减少维修成本。为了更好地预测多工况条件下的设备RUL,提出一种基于卷积神经网络(CNN)联合长短时记忆(LSTM)网络的寿命预测模型。通过变窗口取样获得不同长度的时间序列,基于深度学习方法来发现传感器时序信号与RUL之间的隐藏关系,在训练过程中引入带有热重启的随机梯度下降(SGDR)学习率设定策略,通过感官融合层将子网络的输出特征融合并导入到逻辑回归分类器获得RUL。最后,基于发动机退化仿真数据集进行了有效性验证,表明所提方法在预测精度方面具有明显优势。
英文摘要:
      The life prediction of electromechanical equipment is an important task in condition maintenance.The remaining useful life(RUL) prediction can not only effectively prevent sudden failures of mechanical equipment,but also maximize the use of the equipments working capacity and reduce maintenance costs.In order to better predict the RUL of the equipment under multiple operating conditions,a life prediction model based on convolution neural network(CNN) and long short-term memory network(LSTM) is proposed.Time series of different lengths are obtained through variable window sampling.Based on deep learning,the hidden relationship between the sensor timing signal and the RUL is discovered,and the stochastic gradient descent with warm restarts (SGDR) learning rate setting strategy is introduced during the training process,through the sensory fusion layer,the output features of the sub-network are fused and imported into the logistic regression classifier to obtain the RUL.Finally,the validity is verified based on the engine degradation simulation data set.
查看全文  查看/发表评论  下载PDF阅读器
关闭