基于卷积时空混合神经网络的剩余使用寿命预测  

Remaining Useful Life Prediction Based on Convolutional Spatio-Temporal Hybrid Neural Network

作　　者：刘澳龙 唐向红[1] 陆见光 王涛[2] LIU Aolong;TANG Xianghong;LU Jianguang;WANG Tao(State Key Laboratory of Public Big Data,Guizhou University,Guiyang 550025,China;Key Laboratory of Advanced Manufacturing Technology of the Ministry of Education,Guizhou University,Guiyang 550025,China;Chongqing Industrial Big Data Innovation Center Co.,Ltd.,Chongqing 400707,China)

机构地区：[1]贵州大学公共大数据国家重点实验室,贵阳550025 [2]贵州大学现代制造技术教育部重点实验室,贵阳550025 [3]重庆工业大数据创新中心有限公司,重庆400707

出　　处：《组合机床与自动化加工技术》2025年第4期1-7,共7页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：黔科合基础项目(ZK[2021]一般271/QKHJC-ZK[2021]YB271);黔科合支撑项目([2022]一般074/QKHZC[2022]YB074)。

摘　　要：针对当前剩余使用寿命(remaining useful life,RUL)预测方法侧重于捕捉数据的时间依赖,忽略多传感器间的空间关系问题,提出了一种卷积时空混合神经网络(convolutional spatio-temporal hybrid neural network,CSTHNN)用于提取多传感器时间序列数据的时空特征用于RUL预测。CSTHNN利用卷积神经网络的卷积层感知相邻特征间的空间关系并提取传感器间的空间特征。通过位置编码以记忆时间依赖信息,并使用多头自注意力机制提取时间特征。最后将提取到的时空特征进行非线性变换,映射为RUL预测结果。通过在C-MAPSS数据集上的实验对CSTHNN进行了全面的分析和验证,表明了在RUL预测上提取空间特征的重要性以及该方法优秀的性能。To address the problem that current remaining useful life(RUL)prediction methods focus on capturing the temporal dependence of the data and ignore the spatial relationship among multi-sensors,a convolutional spatio-temporal hybrid neural network(CSTHNN)is proposed for extracting spatio-temporal features of multi-sensor time series data for RUL prediction.CSTHNN utilizes the spatial relationship between perceived features and neighboring features in the convolutional layer of convolutional neural network to extract spatial features among sensors.Then the temporal dependency information is memorized by position encoding and the temporal features are extracted using a multi-head self-attention mechanism.Finally,the extracted spatio-temporal features are nonlinearly transformed and mapped into RUL prediction results.CSTHNN is comprehensively analyzed and validated through experiments on the C-MAPSS dataset,demonstrating the importance of extracting spatial features on RUL prediction as well as the excellent performance of the method.

关 键 词：剩余使用寿命 时空特征 卷积神经网络 混合神经网络 

分 类 号：TH161[机械工程—机械制造及自动化] TG68[金属学及工艺—金属切削加工及机床]