基于多尺度卷积神经网络的轴承剩余寿命预测

Research on Remaining Life Prediction of Bearing Based on Multi-scale Convolution Neural Network

针对滚动轴承剩余寿命难预测的情况,在分析了轴承原始信号特征提取困难的基础上,提出了基于多尺度卷积神经网络的轴承剩余寿命预测方法。该方法将原始振动加速度信号作为输入,依次经过浅层特征提取模块、深层特征提取模块、数据融合模块和输出模块这4部分进行处理,最后输出预测的剩余寿命。同时提出了一种新型的改进均方误差作为网络的损失函数,取得了较好的效果。通过对轴承寿命预测实验的测试数据进行预测分析,该方法能够有效的预测轴承的剩余寿命。

It is difficult to predict the remaining life of rolling bearings,because the original signal characteristics of rolling bearings are not obvious,a method for predicting the remaining life of bearings based on multi-scale convolutional neural networks has been proposed.This method takes the original vibration acceleration signal as input,and then processes it through four parts:shallow feature extraction module,deep feature extraction module,data fusion module and output module,finally outputs the predicted remaining life.At the same time,an improved mean square error was proposed as a loss function,which achieved good results.By predicting and analyzing the test data of the bearing life prediction experiment,this method can effectively predict the remaining life of the bearing.