基于多路稀疏自编码的轴承状态动态监测

Bearing condition dynamic monitoring based on Multi-way sparse autocoder

机械系统的运行是一个时变的过程,为了更好的监测系统的健康状况,通常在设备系统的关键部位加装各种传感器,由此产生大量的数据,传统的单一数据或者人为经验指导均无法快速有效的提取其中的状态信息,排除冗余成分的影响,实现对设备运行状态实时有效的判断。为了有效利用设备上的多路传感器信息,并融合这些信息提取描述系统运行状态的有效成分,实现对机械系统的在线监测。提出利用稀疏自编码深度学习模型对各个传感器采集到的数据进行融合,并结合平方预测误差SPE(Square Prediction Error)指标描述设备运行状态,轴承仿真及轴承故障实验证明,采用稀疏自编码与平方预测误差相结合的模型能够有效的监测轴承故障,并对故障部位进行准确定位。

Mechanical systems operation is a time-varying process. In order to better monitor the health condition of a system, various sensors were installed at key positions of the equipment system, there fore, large amounts of data were generated. Traditional single data and human experience were both unable to quickly and efficiently extract the state information, the effects of redundant components were excluded and effective judgments for the operational status of the system at real-time was realized. In order to make use of multi-sensor information of the device, the information was fused for extracting effective components to realize on-line monitoring of mechanical system, a sparse autocoder deep learning model was proposed to fuse sensors’data. The square prediction error （ SPE） index was combined to describe the equipment running status. Bearing simulation and bearing fault tests showed, that the sparse autocoder deep learning model can effectively monitor bearing failure and identify fault locations.