Prediction of Bearing Remaining Useful Life With Deep Convolution Neural Network
Prediction of Bearing Remaining Useful Life With Deep Convolution Neural Network
复制标题
利用深度卷积神经网络预测轴承剩余使用寿命
DOI：
10.1109/access.2018.2804930
复制
发表时间：
2018-01-01
期刊：
影响因子：
3.9
通讯作者：
Zhang, Lin
中科院分区：
文献类型：
--
作者：
Ren, Lei;Sun, Yaqiang;Zhang, Lin
网络-物理-社会系统（CPSS）在工业物联网（IIoT）等工业应用中引起了极大的关注。作为IIoT的基础组件，轴承在IIoT的CPSS中发挥着越来越重要的作用。更好地了解轴承的工作条件和退化模式，以便更准确地预测剩余使用寿命（RUL），成为工业物联网中工业机器人的迫切需求。数据驱动的方法已经显示出良好的潜力，但预测精度仍然不令人满意。提出了一种基于深度卷积神经网络（CNN）的轴承RUL预测新方法。提出了一种新的特征提取方法，即谱主能量向量。该特征向量适合深度CNN。在预测阶段，我们提出了一种平滑方法来处理预测结果中发现的不连续性问题。据我们所知，我们是第一个提出这样一个平滑的方法，轴承RUL预测。实验表明，该方法可以显著提高轴承RUL的预测精度。
Cyber-physical-social system (CPSS) has drawn tremendous attention in industrial applications such as industrial Internet of Things (IIoT). As the fundamental component of IIoT, bearings play an increasingly important role in CPSS for IIoT. Better understanding of bearing working conditions and degradation patterns so as to more accurately predict the remaining useful life (RUL), becomes an urgent demand for industrial prognostics in IIoT. The data-driven approach has indicated good potential, but the prediction accuracy is still not satisfactory. This paper proposes a new method for the prediction of bearing RUL based on deep convolution neural network (CNN). A new feature extraction method is presented to obtain the eigenvector, named the spectrum-principal-energy-vector. The eigenvector is suitable for deep CNN. In the prediction phase, we propose a smoothing method to deal with the discontinuity problem found in the prediction results. To the best of our knowledge, we are the first to propose such a smoothing method for bearing RUL prediction. Experiments show that our method can significantly improve the prediction accuracy of bearing RUL.