基于贝叶斯优化的GRU网络轴承剩余使用寿命预测方法  

Prediction method of RUL of bearings in GRU network based on Bayesian optimization

作　　者：孟琳书 张音旋[1] 张起 王豪 MENG Linshu;ZHANG Yinxuan;ZHANG Qi;WANG Hao(Shenyang Aircraft Design&Research Institute,The Aviation Industry Corporation of China,Shenyang 110035,China;ASMPT Technology(China)Co.,Ltd.,Chengdu 610000,China)

机构地区：[1]中国航空工业集团公司沈阳飞机设计研究所,辽宁沈阳110035 [2]先进科技(中国)有限公司,四川成都610000

出　　处：《机电工程》2024年第1期130-136,共7页Journal of Mechanical & Electrical Engineering

基　　金：国家自然科学基金资助项目(51905074);辽宁省自然科学基金资助项目(2019-KF-04-04)。

摘　　要：传统的滚动轴承剩余使用寿命预测模型存在参数优化的困难。针对这一问题,笔者提出了一种基于贝叶斯优化的GRU网络滚动轴承剩余使用寿命预测方法,并进行了实验验证,即以PHM2012数据集为例,结合贝叶斯优化算法对基于Encoder-Decoder结构的门控循环单元(GRU)预测模型的多个超参数进行了优化。首先,对包含噪声的原始数据进行了小波包处理,从滚动轴承的振动机理和故障特征出发提取了时域特征,针对该时域特征进行了优化、筛选,并将其输入到模型中的编码器部分,进一步提取了更深层次的时序特征;其次,结合注意力机制与Encoder-Decoder结构,构造了双向GRU神经网络模型,在模型的高维超参数空间中采用贝叶斯优化方法搜索超参数,得到了最优的超参数组合,并在解码器中融入了线性变换,得到了滚动轴承的寿命预测值;最后,封装了全部模型构建、训练与使用过程,建立了基于贝叶斯优化的GRU网络滚动轴承寿命预测流程,并对方法的有效性进行了对比实验验证。研究结果表明:采用基于贝叶斯优化的GRU网络可以有效预测滚动轴承的剩余使用寿命,相比于其他3种方法的最优结果,基于贝叶斯优化的GRU网络的平均预测得分提高了8.01%;基于贝叶斯优化的GRU网络对于真实寿命较短的轴承预测结果较为准确,而对于真实寿命较长的轴承则没有出现预测值大于真实值的情况,可以作为轴承临近失效阶段剩余使用寿命估计的参考。The traditional model for predicting the remaining service life of rolling bearings faces difficulties in parameter optimization.To address this issue,a Bayesian optimization based GRU network method for predicting the remaining service life of rolling bearings was proposed and experimentally validated.Taking the PHM2012 dataset as an example,the Bayesian optimization algorithm was used to optimize multiple hyperparameters of the Encoder Decoder based Gated Recurrent Unit(GRU)prediction model.First,the original data containing noise was subjected to wavelet packet processing.From the vibration mechanism and fault characteristics of rolling bearings,time-domain features were extracted.The extracted time-domain features were optimized and selected,and then input into the encoder part of the model to further extract deeper temporal features.Then,by integrating attention mechanism with the Encoder-Decoder structure,a bidirectional GRU neural network model was constructed.The Bayesian optimization method was employed to search in the high-dimensional hyperparameter space of the model.The optimal hyperparameter combination was gained.Linear transformation was incorporated into the decoder.The predicted remaining service life of the rolling bearings was gained.Finally,the entire process of model construction,training,and usage was encapsulated.A rolling bearing residual service life prediction workflow based on Bayesian optimization of the GRU network was established.The effectiveness of the method was validated through comparative experiments.The research results indicate that the GRU network based on Bayesian optimization can effectively predict the residual service life of rolling bearings.Comparing to the optimal results of the other three methods,the average prediction score of the GRU network based on Bayesian optimization improves by 8.01%.The GRU network based on Bayesian optimization demonstrates accurate predictions for rolling bearings with shorter life.For bearings with longer life,the predicted values do no

关 键 词：参数优化 剩余使用寿命 门控循环单元 贝叶斯优化 超参数调整 注意力机制 Encoder-Decoder结构 

分 类 号：TH133.3[机械工程—机械制造及自动化] TH17