基于PCA和多变量极限学习机的轴承剩余寿命预测

提出了一种基于主成分分析(PCA)和多变量极限学习机(MELM)的轴承剩余寿命预测方法。该方法首先利用PCA技术融合多个表征轴承运行状态与衰退趋势的时域频域特征指标来消除特征间的冗余性和相关性;进一步在单变量极限学习机(ELM)的基础上构建多变量极限学习机模型来预测轴承剩余寿命。该方法克服了传统单变量极限学习机结构简单、信息匮乏等缺点，能有效提高轴承剩余寿命的预测精度。运用全寿命轴承振动数据对模型进行验证，结果表明，相比单独应用ELM模型或MELM模型，基于PCA和MELM剩余寿命预测方法具有更高的预测精度和稳定性。

Residual Life Predictions for Bearings Based on PCA and MELM

A predictive model of the residual life for bearings based on PCA and MELM was proposed herein. To eliminate the redundancy and relevance within features, PCA was applied to fuse multiple characteristic indexes in both of time and frequency domains which could reflect the running states and degradation trends of bearings. Further, on the basis of univariate ELM, a new MELM model was constructed to predict the remaining life of bearings. Unlike traditional univariate ELM, the proposed method overcomes the problems of the simple structure and the lack of information and can improve the prediction precision of residual life for bearings effectively. The bearing run-to-failure tests were carried out to verify the prediction model, and the experimental results demonstrate that the proposed model based on PCA and MELM has the better prediction precision and stability than that of single ELM model or MELM model.