基于卷积神经网络的轴承剩余寿命预测方法
Prediction Method for Bearing Remaining Useful Life Based on Convolutional Neural Networks
张浩 1赵军 1王鹿 1张银龙 2程思宇2
作者信息
- 1. 南京地铁建设有限责任公司,江苏 南京 211806
- 2. 中铁第四勘察设计院集团有限公司,湖北 武汉 430063
- 折叠
摘要
为提高自动扶梯轴承剩余使用寿命(RUL)预测模型的预测精度和泛化能力,提出一种基于卷积神经网络(CNN)的轴承RUL预测方法.首先基于3σ准则对原始数据进行去噪,通过快速傅里叶变换获得其频率特征,其次将不同于传统时间序列数据划分方法的分层抽样应用于数据划分,并构造一个由三个卷积层和两个全连通层组成的深度卷积神经网络DCNN模型,最后利用NASA IMS数据集对预处理方法、DCNN模型精度和泛化能力进行评估,证明了该方法的优越性.
Abstract
To improve the prediction accuracy and generalization ability of the Remaining Useful Life(RUL)prediction model for escalator bearings,a bearing RUL prediction method based on Convolutional Neural Network(CNN)is proposed.Firstly,it denoises the original data based on 3σ criterion,obtains its frequency characteristics through fast Fourier transformation.Secondly,it applies layered sampling different from traditional time series data partitioning methods to data partitioning,and constructs a Deep Convolutional Neural Network(DCNN)model consisting of three convolutional layers and two fully connected layers.Finally,the NASA IMS dataset is used to evaluate the preprocessing method,DCNN model accuracy,and generalization ability,proving the superiority of this method.
关键词
剩余寿命预测/3σ准则/分层抽样/DCNN/泛化能力Key words
RUL prediction/3σ criterion/layered sampling/DCNN/generalization ability引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据