基于深度学习多样性特征提取与信息融合的行星齿轮箱故障诊断方法  

Planetary Gearbox Fault Diagnosis Based on Multiple Feature Extraction and Information Fusion Combined with Deep Learning

作　　者：金棋 王友仁[1] 王俊[1] JIN Qi;WANG Youren;WANG Jun(College of Automation Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing,211106)

机构地区：[1]南京航空航天大学自动化学院,南京211106

出　　处：《中国机械工程》2019年第2期196-204,共9页China Mechanical Engineering

基　　金：国家自然科学基金资助项目(61371041);航空科学基金资助项目(2013ZD52055);国家商用飞机制造工程技术研究中心创新基金资助项目(SAMC14-JS-15-051)

摘　　要：针对行星齿轮箱振动信号噪声干扰大、单一分类器泛化能力不强的问题,提出了一种基于深度学习多样性特征提取与信息融合的行星齿轮箱故障诊断方法。利用多目标优化算法优化多个堆栈去噪自动编码器(SDAE)以获得多个性能优异的SDAE,并提取多样性的故障特征;采用多响应线性回归模型集成多样性故障特征实现信息融合,得到多目标集成堆栈去噪自动编码器(MO-ESDAE),最后将其应用于行星齿轮箱故障诊断。实验结果表明:该方法能有效提高故障诊断精度与稳定性,具有较强的泛化能力。According to the heavy noises of vibration signals and the difficulty of incipient fault diagnosis for planetary gearboxes using single classifier,a method of planetary gearbox fault diagnosis was proposed based on multiple feature extraction and information fusion combined with deep learning.The multiple excellent stacked denoising autoencoders(SDAEs)were acquired based on multi-objective evolutionary algorithm.Then,multi-response linear regression model was employed to integrate multiple SDAEs for building multi-obiective ensemble stacked denoising autoencoders(MO-ESDAEs),which was used to diagnose faults of planetary gearboxes.The experimental results show that the proposed method may enhance the fault diagnosis accuracy and stability.

关 键 词：行星齿轮箱故障诊断 深度神经网络 多样性特征提取 多目标进化算法 

分 类 号：TH17[机械工程—机械制造及自动化]