一种基于Bi-LSTM和MLP融合的滚动轴承剩余寿命预测方法

A rolling bearing RUL prediction method based on a novel fusion model of Bi-LSTM and multilayer perceptron

目前长短期记忆网络(long short-term memory,LSTM)是滚动轴承剩余使用寿命(remaining useful life,RUL)预测的常用方法,但其训练过程中收敛速度慢、波动性剧烈、预测误差大的问题,严重影响其在实际生产中的应用.针对该问题,提出一种双向长短期记忆网络(Bi-directional LSTM,Bi-LSTM)与多层感知机(multilayer perceptron,MLP)融合的预测方法.首先,采用堆叠去噪自编码器(stacked denoising auto encoder,SDAE)对预处理后的滚动轴承振动信号进行特征提取;其次构建Bi-LSTM网络与多层感知机融合的多种预测模型,并通过实验获取较优模型;最后使用较优模型对其剩余寿命进行预测.实验结果表明,相对LSTM,Bi-LSTM,以及LSTM融合MLP等常用模型,采用文中提出的方法,模型在滚动轴承剩余寿命训练过程中,波动性更低、收敛速度更快,同时预测误差也得到明显降低.

For mechanical equipment that rotates around an axis,rolling bearings play an extremely important role.In order to do a good job in the maintenance and management of rolling bearings,it is necessary not only to understand the failure of rolling bearings,but also to find the cause of the failure in a relatively short time.If the remaining useful life(RUL)of the rolling bearing can be predicted,measures can be made in advance,which greatly reduces the time for troubleshooting.The remaining useful life of rolling bearings is of great significance to ensure the safe and stable operation of rotating machinery.Timely replacement of bearings can not only avoid major economic losses,but also avoid casualties caused by mechanical equipment failure.At the same time,the cost of bearing replacement and the time for troubleshooting can be effectively reduced,reducing economic losses.Therefore,the research on the remaining useful life of the bearing has significance and value for the maintenance and normal operation of mechanical equipment.In the process of predicting the remaining useful life of rolling bearings,Bi-directional long short-term memory(Bi-LSTM)is widely used,however,the slow convergence speed and large volatility is a common issue,leading a large prediction error.In this paper,a prediction method that combines a Bi-LSTM network and a multilayer perceptron(MLP)is proposed.First,a stacked denoising auto encoder(SDAE)is used to extract the unsupervised deep information feature from the time domain vibration signal of the rolling bearing;second,a fusion model is proposed to combine a Bi-LSTM network with a MLP,and an optimized fusion model structure is determined based on extensive experiments.Finally,the optimal model is used predict RUL of rolling bearings.The experimental results show that the volatility is effectively reduced,the convergence speed of the model is accelerated,and the error is significantly reduced in the process of predicting the life of the rolling bearing.