基于自编码器和门限循环单元神经网络的滚动轴承退化趋势预测

Degradation trend prediction of rolling bearing based on auto-encoder and GRU neural network

针对现有滚动轴承性能退化趋势预测方法存在退化指标选取困难、预测精度较低的问题,提出基于自编码器和门限循环单元神经网络的滚动轴承退化趋势预测方法。首先,构建轴承振动信号混合域高维特征集,采用指标综合评价值初步筛选敏感性高、趋势性好的性能退化指标;然后,利用自编码器融合高维特征集,消除混合域特征之间的冗余信息;在此基础上,将融合后的特征输入门限循环单元(gated recurrent unit,GRU)神经网络模型以完成滚动轴承退化趋势预测。试验结果表明,所提方法能获得更加准确的滚动轴承退化趋势预测结果。

Aiming at problems of degradation index selection being difficult and prediction accuracy being low in the rolling bearing performance degradation trend prediction method,a prediction method for rolling bearing degradation trend based on auto-encoder and GRU neural network was proposed.Firstly,a mixed field high-dimensional feature set of bearing vibration signals was constructed,and the index comprehensive evaluation value was used to preliminarily screen performance degradation indexes,and gain those with higher sensitivity and better trend.Then,the auto-encoder was used to fuse the high-dimensional feature set,and eliminate redundant information among mixed field features.Finally,fused features were input into the GRU neural network model to complete rolling bearing degradation trend prediction.Test results showed that the proposed method can obtain more accurate prediction results for rolling bearing performance degradation trend.