摘要
针对轴承剩余使用寿命(RUL)预测中的不确定性量化问题,综合考虑数据不确定性与模型不确定性,提出了一种基于卷积神经网络(CNN)的RUL区间预测方法。首先,对轴承输入数据进行了预处理,并提取了轴承振动信号的时域特征,选用有强趋势性的参数作为模型输入,接着设计了一个在输出层放置正态分布的CNN模型,将其用于点预测及数据不确定性的捕捉;然后,采用集成方法对模型不确定性进行了量化,输出了区间预测结果;最后,采用PHM2012轴承退化公开数据集对基于CNN的区间预测方法的有效性进行了验证,并将结果与采用贝叶斯神经网络(BNN)所得结果进行了比较。实验结果表明:在轴承RUL预测的应用中,基于CNN的区间预测方法的区间覆盖率(PICP)最高,其值比BNN高出了63.9%,点预测结果的均方根误差(RMSE)值最小,其值为0.1997。研究结果表明:基于CNN的区间预测方法可确保点预测估计的准确性,同时,在描述预测不确定性方面具有更大的优越性和实际意义。
Aiming at the problem of uncertainty quantification of remaining useful life(RUL)prediction of bearings,integrating both data uncertainty and model uncertainty,an interval prediction method on RUL of bearings based on convolutional neural network(CNN)was proposed.Firstly,the input data was pre-processed to extract the time domain features of the vibration signal,and the parameters with the strong trend were selected as the input of the model.Then,a CNN model with a normal distribution placed in the output layer was designed for the point prediction and data uncertainty capture.After that,an ensemble method was used to quantify the model uncertainty and output the interval prediction results.Finally,the effectiveness of the proposed method was validated using the published PHM2012 bearing degradation dataset,and the results were prepared with those obtained by Bayesian neural network(BNN).The experiment result indicates that,in the application of RUL prediction of bearings,the proposed method has the highest prediction interval coverage probability(PICP),63.9%higher than that of BNN,while it holds the smallest root mean squared error(RMSE)of the point prediction with 0.1997.The results indicate that,the proposed interval prediction method on RUL of bearings based on CNN has greater advantages in describing the uncertainty of prediction while ensuring the accuracy of point prediction estimation,which has more significance in practice.
作者
周明珠
张艺宝
吴双
孔丽军
王梓齐
ZHOU Ming-zhu;ZHANG Yi-bao;WU Shuang;KONG Li-jun;WANG Zi-qi(Inner Mongolia Huomei Hongjun Aluminum Power Co.,Ltd.,Tongliao 029200,China;Hunan Zhongrong Huizhi Information Technology Co.,Ltd.,Changsha 410221,China;School of Control Science and Engineering,Zhejiang University,Hangzhou 310027,China;Huzhou Institute of Zhejiang University,Huzhou 313002,China)
出处
《机电工程》
CAS
北大核心
2023年第8期1225-1230,共6页
Journal of Mechanical & Electrical Engineering
基金
国家自然科学基金面上项目(61973269)。
关键词
滚动轴承
剩余使用寿命
区间预测
不确定性量化
卷积神经网络
区间覆盖率
rolling bearing
remaining useful life(RUL)
interval prediction
uncertainty quantification
convolutional neural network(CNN)
prediction interval coverage probability(PICP)
