摘要
传统的轴承剩余使用寿命预测方法大多是对原始振动信号进行时域特征、频域特征以及时频域特征的提取,创建轴承的健康指标来建立模型,实现剩余寿命预测.为了简化轴承剩余寿命预测方法及提高预测的准确度,提出一种只保留传统卷积神经网络里的卷积层,且把卷积层改为窄卷积的降维方法.首先,将窄卷积层神经网络对原始输入信号进行特征学习,构建健康指标;其次,采用Adam优化损失函数及加权平均方法对网络输出结果进行降噪处理,得到健康指标,进而根据健康指标反向计算且平滑后得到剩余使用寿命;最后,通过滚动轴承全寿命试验数据仿真证明该方法能够准确预测轴承剩余使用寿命,且与传统卷积神经网络的预测结果进行对比,该方法的寿命百分比误差均值为7.33%,传统卷积神经网络的寿命百分比误差均值为61.65%,该方法的平均误差降低了88.11%,验证了其有效性.
The traditional bearing remaining service life prediction methods mostly extract the original vibration signals in time domain,frequency domain,and time-frequency domain characteristics,and create bearing health indicators to establish a model to achieve remaining life prediction.In order to simplify the prediction method of bearing remaining life and improve the accuracy of prediction,this paper proposes a dimensionality reduction method that only retains the convolutional layer in the traditional convolutional neural network and changes the convolutional layer to narrow convolution.Firstly,the proposed narrow convolutional layer neural network performs feature learning on the original input signal to construct health indicators.Secondly,Adam optimization loss function and weighted average method are used to denoise the network output results to obtain health indicators,and then according to the health indicators the remaining service life is obtained after reverse calculation and smoothing.Finally,the simulation of the full life test data of rolling bearings shows that this method can accurately predict the remaining service life of the bearing,and compared with the prediction results of traditional convolutional neural networks,the average life percentage error of this method is 7.33%,while the average lifetime percentage error of the traditional convolutional neural network is 61.65%.The average error of this method has been reduced by 88.11%,which verifies its effectiveness.
作者
高淑芝
褚智伟
GAO Shuzhi;CHU Zhiwei(Shenyang University of Chemical Technology,Equipment Reliability Research Institute,Shenyang 110142,China;Shenyang University of Chemical Technology,School of Information Engineering,Shenyang 110142,China)
出处
《沈阳化工大学学报》
CAS
2023年第2期151-158,共8页
Journal of Shenyang University of Chemical Technology
基金
辽宁省自然科学基金项目(20170540725)
辽宁省高端人才建设项目—辽宁省特聘教授([2018]3533)。
关键词
窄卷积层
神经网络
轴承
加权平均法
剩余寿命预测
narrow convolutional layer
neural network
bearing
weighted average method
remaining life prediction
