摘要
基于传统方法预测轴承剩余使用寿命(Residual Useful Life,RUL),步骤繁多,成本昂贵,且模型不具泛化性.现有的基于深度学习(Deep Learning,DL)的预测方法,由于数据量过大,经常导致模型出现过拟合现象,从而使模型精度不高.为了克服以上缺点,提出一种基于MRSDAE-SOM结合HGRU的滚动轴承RUL预测方法.首先,使用无监督式网络流形正则化堆栈去噪自编码器(Manifold Regularization Stack Denoising Auto Encoder,MRSDAE)结合自组织映射(Self-Or-ganizing Mapping,SOM)构建轴承健康因子(Health Indicator,HI).然后,通过分层门控循环单元(Hierarchical Gated Re-current Unit,HGRU)网络建立预测模型,HGRU网络通过加入多尺度层和密集层,使其具有捕获时序特征且集成不同时间尺度注意力信息的能力.最后,通过实验验证表明,相比于其他基于数据驱动的方法,所提方法构建健康因子使用无监督方式,高效快捷且便于应用;预测模型泛化能力好,并有效防止了过拟合现象,实现了更高的预测精度.
Abstract
Predicting the Residual Useful Life(RUL)of bearings based on traditional methods has many steps,is expensive,and the model is not generalizable.And the existing deep learning-based prediction methods often lead to overfitting of the model due to the excessive amount of data,which results in poor model accuracy.In order to overcome the above drawbacks,a rolling bearing RUL prediction method based on Manifold Regularization Stack Denoising Auto Encoder(MRSDAE)combined with Self-Organi-zing Mapping(SOM)combined with HGRU was proposed.Firstly,a bearing Health Indicator(HI)was constructed using an un-supervised network MRSDAE-SOM.Then,a prediction model was built through a Hierarchical Gated Recurrent Unit(HGRU)network.The HGRU network has the ability to capture temporal features and integrate attention information from differ-ent time scales by adding multi-scale and dense layers.Finally,experimental validation shows that compared with other data-driv-en methods,the proposed method constructs health indicator in an unsupervised way,which is efficient and easy to apply.The model generalizes well and effectively prevents overfitting,and achieves higher prediction accuracy.
维普
万方数据