Kriging模型在滚动轴承剩余使用寿命预测中的应用  

Application of Kriging Model in Residual Useful Life Prediction of Rolling Bearings

作　　者：刘吉文 秦东晨[1] 袁峰[1] 陈江义[1] LIU Jiwen;QIN Dongchen;YUAN Feng;CHEN Jiangyi(School of Mechanical and Power Engineering,Zhengzhou University,Zhengzhou 450001,China)

机构地区：[1]郑州大学机械与动力工程学院,郑州450001

出　　处：《轴承》2025年第3期104-110,共7页Bearing

基　　金：国家重点研发计划资助项目(2018YFB2000501-01)。

摘　　要：针对常用优化算法对滚动轴承剩余使用寿命(RUL)预测模型进行超参数优化时易陷入局部最优的问题,提出了一种基于Kriging代理模型和长短期记忆网络(LSTM)的滚动轴承剩余使用寿命预测模型。首先,改进小波阈值函数对轴承原始振动信号进行降噪处理;其次,通过自适应融合方法构建轴承健康指标(HI)曲线并作为预测模型的输入;然后,搭建Kriging代理模型,以寿命预测结果的均方根误差(RMSE)值为优化目标,LSTM模型隐藏层单元数和Dropout层丢弃率为优化变量对LSTM模型寻优得到最优参数组合;最后,用超参数优化后的LSTM模型进行滚动轴承的RUL预测。基于西安交通大学轴承数据集,与传统LSTM、反向传播(BP)神经网络和多层感知机(MLP)的预测结果进行了对比,结果表明所提模型的预测曲线能更好地贴近轴承真实退化趋势,预测结果更加接近轴承真实寿命,验证了该模型的有效性。In order to solve the problem that the remaining useful life(RUL) prediction model of rolling bearings is prone to local optimum when hyperparameters are optimized by common optimization algorithms,a RUL prediction model of the bearings is proposed based on Kriging surrogate model and long short-term memory(LSTM).Firstly,the wavelet threshold function is improved to denoise the original vibration signal of the bearings.Secondly,the bearing health indicator(HI) curve is constructed by adaptive fusion method and used as input of prediction model.Then,a Kriging surrogate model is built,the root mean square error(RMSE) value of life prediction results is taken as optimization objective,and the number of hidden layer units and Dropout layer dropout rate are taken as optimization variables to optimize the LSTM model and obtain the optimal parameter combination.Finally,the LSTM model optimized with hyperparameters is used for RUL prediction of the bearings.Based on bearing data set from Xi'an Jiaotong University,the prediction results of the proposed model are compared with those of traditional LSTM,back propagation(BP) neural network and multilayer perceptron(MLP).The results show that the prediction curve of the proposed model is closer to real degradation trend of the bearings,and the prediction results are closer to real life of the bearings,verifying the effectiveness of the proposed model.

关 键 词：滚动轴承 剩余使用寿命 寿命预测 长短期记忆网络 Kriging代理模型 

分 类 号：TH133.33[机械工程—机械制造及自动化]