基于深度置信网络的行星齿轮箱振动特征提取  

Vibration feature extraction of planetary gearbox based on deep belief networks

作　　者：李忠刚 何林锋 Li Zhonggang;He Linfeng(College of Mechanical and Electrical Engineering,Beijing Information Science & Technology University,Beijing 100192,China)

机构地区：[1]北京信息科技大学机电工程学院

出　　处：《电子测量与仪器学报》2019年第9期199-205,共7页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(51275052)资助项目

摘　　要：针对行星齿轮箱在复杂工况下提取反映机械故障振动特征困难的问题,研究具备数据降维与特征提取能力的深度置信网络(DBN)提取机械故障振动特征。通过分析不同梯度优化算法在DBN网络中的表现,应用振动数据集对优化算法进行验证,选择最优优化算法,并分析训练迭代次数、隐含层节点数、网络层数等DBN网络参数变化对于振动数据特征提取能力的影响,提出了相关参数优化设计方法。实验分析证明,优化设计后的深度置信网络能够有效提取行星齿轮箱振动特征,有利于其故障状态分类识别。Aiming at the difficulty of extracting vibration features of planetary gearbox under complex working conditions, a deep belief network ( DBN) with the ability of dimensionality reduction and feature extraction was studied to extract vibration features of mechanical faults. The performance of different gradient optimization algorithms in DBN was analyzed in this paper. The vibration data sets were used to validate the optimization algorithm and selected the optimal optimization algorithm. The influence of training iteration times, the number of hidden layer nodes and the number of network layers on the feature extraction ability of vibration data was analyzed, and the optimal design method of relevant parameters was proposed. The experimental analysis proves that the optimized DBN can effectively extract the vibration features of planetary gearbox and it is beneficial to the classification and recognition of its fault state.

关 键 词：行星齿轮箱 振动 特征提取 深度置信网络 

分 类 号：TH17[机械工程—机械制造及自动化]