数据缺失下SGAIN融合TCN预测滚动轴承剩余寿命
SGAIN fusion TCN for predicting residual life of rolling bearings with missing data
刘静涛 1邱明 2李军星 1刘志卫 1高锐1
作者信息
- 1. 河南科技大学机电工程学院,河南洛阳 471003
- 2. 河南科技大学机电工程学院,河南洛阳 471003;机械装备先进制造河南省协同创新中心,河南洛阳 471003
- 折叠
摘要
由于网络传输故障和传感器漏读会引起数据缺失问题.为了在数据缺失条件下能够较准确地预测滚动轴承使用寿命,论文给出了一种将精简生成对抗插补网络(SGAIN)与时间卷积网络(TCN)相融合的剩余寿命预测(RUL)方法.首先,通过SGAIN算法学习缺失数据集的分布规律,掌握已有数据和缺失数据的关联,对缺失数据进行插补填充.其次,使用TCN网络建立轴承寿命预测模型,运用插补完成的数据集实现数据缺失下滚动轴承的剩余寿命预测.最后,借助于公开数据集将SGAIN插补方法与其他插补方法进行对比,揭示了SGAIN插补方法的优越性.同时,选择20%缺失率下的轴承缺失数据做出预测,插补后寿命预测结果的得分达到了0.722 2,与缺失未插补数据的预测结果的得分0.542 5 相比提高了0.179 7,接近原始数据寿命预测结果的得分0.755 2.这说明了SGAIN融合TCN的滚动轴承剩余寿命预测方法是有效的.
Abstract
As the net-work transmission faults and sensor miss-reading will cause the problem of data missing,to predict the rolling bearing service life more accurately under the condition of data missing,the paper gives a prediction method of residual life(RUL)that integrates the slim generative adversarial interpolation net-work(SGAIN)with the temporal convolutional network(TCN).Firstly,SGAIN is used to learn the distribution pattern of the missing dataset,to grasp the association between the existing data and the missing data so as to interpolate the missing data.Secondly,a rolling bearing residual life prediction model is established using TCN network.The interpolated completed dataset was used to achieve the residual life prediction of rolling bearings with missing data.Finally,the SGAIN interpolation method was compared with other interpolation methods using publicly available datasets,which the superiority of the SGAIN interpolation method was revealed.At the same time,the missing bearing data at 20%missing rate were selected to make the prediction of residual life.The results show that the interpolated data life prediction score reaches 0.722 2,which is 0.179 7 higher than the score of 0.542 5 for the missing uninterpolated data,and is closer to the original data life prediction score of 0.755 2.This shows that the rolling bearing residual life prediction method based on SGAIN integrated with TCN is effective.
关键词
滚动轴承/数据缺失/精简对抗生成插补网络/时间卷积网络/寿命预测Key words
rolling bearings/missing data/SGAIN/TCN/residual life prediction引用本文复制引用
基金项目
国家重点研发计划(2020YFB2007303)
国家自然科学基金(52005159)
河南省青年托举人才项目(2023HYTP050)
出版年
2024
NETL
NSTL
万方数据