基于无监督深度模型迁移的滚动轴承寿命预测方法

Rolling Bearing Life Prediction Based on Unsupervised Deep Model Transfer

针对实际中某种工况滚动轴承带标签振动数据获取困难,健康指标难以构建及寿命预测误差大的问题,提出一种基于无监督深度模型迁移的滚动轴承剩余使用寿命(Remaining useful life,RUL)预测方法.该方法首先对滚动轴承全寿命周期振动数据提取均方根(Root mean square,RMS)特征,并引入新的自下而上(Bottom-up,BUP)时间序列分割算法将特征序列分割为正常期、退化期和衰退期3种状态;对振动信号经快速傅里叶(Fast Fourier transform,FFT)变换后的幅值序列进行状态信息标记,并将其输入到新增卷积层的全卷积神经网络(Full convolutional neural network,FCN)中,提取深层特征,得到预训练模型;提出将预训练模型的梯度作为一种“特征”与传统预训练模型特征一起参与目标域网络训练过程,从而得到状态识别模型;利用状态概率估计法结合状态识别模型建立滚动轴承寿命预测模型.实验验证所提方法无需构建健康指标,可实现无监督条件下不同工况滚动轴承剩余寿命预测,并获得较好的效果.

In order to solve the problems such as difficulty in acquiring labeled vibration data of rolling bearings under certain working condition in practice,difficulty in constructing health indicators and large error in life prediction of rolling bearings,a method of remaining useful life(RUL)prediction of rolling bearings is proposed based on unsupervised deep model transfer.Firstly,the root mean square(RMS)features of the vibration data of the full life cycle of the rolling bearings are extracted,and a new bottom-up(BUP)time series segmentation algorithm is introduced to divide the feature sequence into three states:Normal period,degradation period and recession period.Mark the state information of the amplitude sequence of the vibration signal after the fast Fourier transform(FFT),and input it into the fully convolutional neural network(FCN)of the newly added convolutional layer to extract deep features,and the pre-trained model can be obtained.The gradient of the pre-trained model is proposed and used as a“feature”to participate in the target domain network training process together with the traditional pretrained model features,and the state identification model is obtained.Using state probability estimation method combined with state identification model,life prediction model of rolling bearing can be established.Experiments verify that,without establishing health indicators,the proposed method can realize remaining useful life prediction of rolling bearings for different working conditions under unsupervised conditions,and achieve better results.