基于MRSDAE-KPCA结合Bi-LST的滚动轴承剩余使用寿命预测
Residual Useful Life Prediction of Rolling Bearings Based on MRSDAE-KPCA Combined with Bi-LSTM
古莹奎 1陈家芳 1石昌武1
作者信息
- 1. 江西理工大学 机电工程学院,江西 赣州 341000
- 折叠
摘要
针对现有滚动轴承剩余使用寿命预测方法在提取数据特征时没有充分考虑数据的内部分布,且在构建健康因子时还需要专家经验进行人工提取等问题,提出一种基于流形正则化堆栈去噪自编码器、核主成分分析并结合双向长短时记忆网络的滚动轴承剩余使用寿命预测方法.首先采用无监督的堆栈去噪自编码器网络对原始振动数据进行深层特征提取,并使用核主成分分析法进一步降维,以提高健康因子的指标稳定性;然后在堆栈去噪自编码器中加入流形正则化,最大程度保留编码器隐藏层内部的数据分布结构,提高模型提取数据特征的有效性.最后使用双向长短时记忆网络预测轴承的剩余使用寿命,并采用AdaMax优化算法对网络模型的超参数进行自适应寻优.分析结果表明,提出的滚动轴承剩余使用寿命预测方法具有更高的精度.
Abstract
Aiming at the problem that the existing rolling bearing residual useful life prediction methods do not fully consider the internal distribution of the data when extracting the data features,and the artificial extraction of expert experience is also required when constructing the health factors,a rolling bearing residual useful life prediction method based on manifold regularization stack denoising autoencoder(MRSDA),kernel principal component analysis(KPCA)and combined with bi-directional long short-term memory(Bi-LSTM)network is proposed.Firstly,the unsupervised stack denoising autoencoder network is used to extract the deep features of the original vibration data,and the KPCA is used to further reduce the dimension to improve the index stability of the health factors.Then,manifold regularization is added to the stack denoising autoencoder to maximize the data distribution structure within the coder's hidden layer and improve the effectiveness of the model in extracting data features.Finally,the Bi-LSTM network is used to predict the residual useful life of the bearing,and the AdaMax optimization algorithm is used to adaptively optimize the super parameters of the network model.The analysis results show that the proposed method for predicting the residual useful life of rolling bearings has higher accuracy.
关键词
故障诊断/滚动轴承/剩余使用寿命预测/健康因子/流形正则化堆栈去噪自编码器/双向长短时记忆网络Key words
fault diagnosis/rolling bearing/residual useful life prediction/health factors/manifold regularization stack denoising autoencoder/bi-directional long short-term memory network引用本文复制引用
基金项目
国家自然科学基金(61963018)
江西省自然科学基金重点项目(20212ACB202004)
出版年
2024
国家科技期刊平台
NSTL
万方数据