基于一维卷积神经网络的圆柱滚子轴承保持架故障诊断  

Fault diagnosis of cylindrical roller bearing cage based on 1D convolution neural network

作　　者：郑一珍 牛蔺楷 熊晓燕 祁宏伟 马晓雄 ZHENG Yizhen;NIU Linkai;XIONG Xiaoyan;QI Hongwei;MA Xiaoxiong(School of Mechanical and Transportation Engineering,Taiyuan University of Technology,Taiyuan 030024,China;Key Laboratory of New Sensors and Intelligent Control Ministry of Education,Taiyuan University of Technology,Taiyuan 030024,China)

机构地区：[1]太原理工大学机械与运载工程学院,太原030024 [2]太原理工大学新型传感器与智能控制教育部重点实验室,太原030024

出　　处：《振动与冲击》2021年第19期230-238,285,共10页Journal of Vibration and Shock

基　　金：国家自然科学基金(51705351);山西省研究生创新项目(2020sy546)。

摘　　要：为解决滚动轴承保持架故障振动信号存在的不稳定性、无冲击特性和故障特征难以获取问题,研究提出基于“端到端”识别的适应性卷积神经网络故障诊断模型。将不同保持架故障状态下的振动信号按一定比例采用有重叠样本分割进行数据增强,并对样本实施分段标准化预处理构建训练和测试集合;利用卷积神经网络实现对振动信号的自适应特征提取和特征降维;在输出端利用全局平均池化替换经典构架中使用的全连接运算,以减少训练模型参数和过程运算量,避免发生过拟合,最终经Softmax分类输出诊断结果。试验结果表明算法能够达到99%以上的故障识别率,且在不同负载和转速下均保持良好的泛化性能和鲁棒性,可有效应用于轴承保持架故障诊断任务。Here,to solve existing problems of instability,non-impact characteristics and difficult to obtain fault characteristics in rolling bearing cage fault vibration signals,an adaptive convolution neural network fault diagnosis model based on"end-to-end"recognition was proposed.Firstly,vibration signals under different cage fault conditions were segmented by overlapping samples according to a certain proportion for data enhancement,and samples were preprocessed by subsection standardization to construct training and test sets.Then,the convolution neural network was used to realize adaptive feature extraction and feature dimensionality reduction of vibration signals.Finally,the global average pooling was used to replace the full connection operation used in the classical framework at the output end,reduce amounts of training model parameters and process calculation,and avoid over-fitting.The diagnosis results were output through Softmax classification.The test results showed that the algorithm can reach more than 99%fault recognition rate,maintain good generalization performance and robustness under different loads and rotating speeds,and be effectively applied in fault diagnosis of bearing cage.

关 键 词：保持架故障诊断 故障损伤程度 卷积神经网络 振动信号 故障诊断 

分 类 号：TH113.1[机械工程—机械设计及理论] TP391.4[自动化与计算机技术—计算机应用技术]