基于RBM-CNN模型的滚动轴承剩余使用寿命预测  

Remaining Useful Life Prediction of Rolling Bearings Based on RBM-CNN Model

作　　者：张永超[1] 杨海昆 刘嵩寿 赵帅 陈庆光[1] ZHANG Yongchao;YANG Haikun;LIU Songshou;ZHAO Shuai;CHEN Qingguang(College of Mechanical and Electronic Engineering,Shandong University of Science and Technology,Qingdao 266590,China)

机构地区：[1]山东科技大学机械电子工程学院,山东青岛266590

出　　处：《轴承》2025年第5期96-101,共6页Bearing

基　　金：山东省自然科学基金资助项目(ZR2021ME242)。

摘　　要：针对滚动轴承剩余使用寿命预测时存在特征提取困难及预测准确性较差的问题,提出一种基于受限玻尔兹曼机(RBM)与卷积神经网络(CNN)的滚动轴承剩余使用寿命预测模型。首先,采用快速傅里叶变换对轴承原始振动信号进行频域变换构建幅值谱;其次,通过RBM挖掘幅值谱中的深度全局特征;然后,通过建立早期故障阈值点划分退化阶段;最后,利用深度CNN对轴承剩余使用寿命进行预测。使用辛辛那提大学轴承数据集对所提方法进行验证,并与其他深度学习方法进行对比,结果表明RBM-CNN模型的均方误差(MSE)、均方根误差(RMSE)、平均绝对误差(MAE)最小,预测准确度最高,达到90.05%,验证了RBM-CNN模型在滚动轴承剩余使用寿命预测中的优越性。A remaining useful life prediction model for rolling bearings based on Restricted Boltzmann Machine(RBM)and Convolutional Neural Networks(CNN)is proposed to address the difficulties in feature extraction and poor prediction accuracy in remaining useful life prediction of rolling bearings.Firstly,the fast Fourier transform is used to perform frequency domain transformation on original vibration signal of the bearings,and the amplitude spectrum is constructed.Secondly,the deep global features in amplitude spectrum are mined by RBM.Then,the degradation stages are divided by establishing early fault threshold points.Finally,the deep CNN is used to predict the remaining useful life of the bearings.The proposed method is validated using bearing dataset from University of Cincinnati and compared with other deep learning methods.The results show that the RBM-CNN model has the smallest Mean Squared Error(MSE),Root Mean Squared Error(RMSE)and Mean Absolute Error(MAE),with the highest prediction accuracy of 90.05%,demonstrating the superiority of RBM-CNN model in predicting the remaining useful life of the bearings.

关 键 词：滚动轴承 使用寿命 寿命预测 玻尔兹曼机 卷积神经网络 

分 类 号：TH133.33[机械工程—机械制造及自动化]