基于MSCAE的滚动轴承健康指标提取及其应用  

Extraction of rolling bearing health index based on MSCAE and its application

作　　者：叶子健 张强[1] 邵思羽 杨新宇[1] 牛天林[1] 赵玉伟 焦晓璇 YE Zijian;ZHANG Qiang;SHAO Siyu;YANG Xinyu;NIU Tianlin;ZHAO Yuwei;JIAO Xiaoxuan(Air Defense & Antimissile School, Air Force Engineering University, Xi’an 710051, China;School of Aeronautical Engineering, Air Force Engineering University, Xi’an 710051, China)

机构地区：[1]空军工程大学防空反导学院,西安710051 [2]空军工程大学航空工程学院,西安710051

出　　处：《兵器装备工程学报》2022年第6期239-247,301,共10页Journal of Ordnance Equipment Engineering

基　　金：国家自然科学基金资助项目(52106191);陕西省自然科学基础研究计划(2020JQ-475)。

摘　　要：提出了一个端到端的健康指标提取模型,称为多尺度卷积自编码器(MSCAE)。该模型由3个卷积核大小不同的卷积自编码器并行组成,能够充分地利用振动信号的全局信息和局部信息,并且所提取的健康指标不再需要依据专家经验设置失效阈值。通过PHM2012挑战赛数据集进行验证,结果表明,相比于单尺度卷积自编码器,所提出的MSCAE提取的健康指标更能反映滚动轴承的真实退化趋势,与自编码器和卷积神经网络等传统模型进行对比分析,验证了所提出方法的优越性。同时,使用GRU神经网络的剩余使用寿命预测实验证明了使用MSCAE提取的健康指标具有很高的预测精度。The construction of health indicator is a key step in the data-driven remaining useful life prediction methods.However,the existing methods for health indicator extraction often need to artificially select proper failure threshold,and do not make full use of the global and local information of the vibration data.In order to solve the above problems,this paper proposed an end-to-end health index extraction model,called multi-scale convolutional auto encoder(MSCAE).This model was composed of three convolutional auto encoders with different sizes of convolution kernels in parallel,which can make full use of the global and local information of the vibration data,and the extraction of health indicator do not need to set the failure threshold in advance based on expert experience.The experimental results on the PHM2012 challenge data set show that compared with the single-scale convolutional autoencoder,the health index extracted by the proposed MSCAE can better reflect the true degradation trend of rolling bearings.In addition,the proposed framework was compared with traditional models such as auto encoder and convolutional neural network to further verify the superiority.The remaining useful life prediction experiment using the GRU neural network also proved that the health indicator extracted by the MSCAE have high prediction accuracy.

关 键 词：MSCAE 滚动轴承 剩余使用寿命预测 健康指标 

分 类 号：TH165.3[机械工程—机械制造及自动化] TN911.7[电子电信—通信与信息系统] TH133.3[电子电信—信息与通信工程]