基于卷积神经网络的机械轴承故障智能识别  

Intelligent Identification of Mechanical Bearing Faults Based on Convolutional Neural Networks

作　　者：王路 殷鸿鑫 刘鸿旭 WANG Lu;YIN Hongxin;LIU Hongxu(China Energy Railway Equipment Company Limited Suning Branch,Cangzhou 062350,China;Tianjin Hveic Technologies Co.,Ltd.,Tianjin 301799,China)

机构地区：[1]国能铁路装备有限责任公司肃宁车辆维修分公司,沧州062350 [2]天津哈威克科技有限公司,天津301799

出　　处：《自动化与仪表》2023年第11期71-75,共5页Automation & Instrumentation

摘　　要：机械轴承振动信号噪声和复杂特征均会对故障识别造成不利影响,导致故障识别精度下降,所以研究基于卷积神经网络的机械轴承故障智能识别方法。利用小波包分析法对轴承振动信号展开多层次分析处理,结合二进制变换法实现信号重构。利用卷积神经网络中的卷积层对重构后信号展开局部卷积后,通过池化运算完成信号特征提取。利用基于SVDD的信号分类函数判定该信号样本是否为故障信号,实现机械轴承故障智能识别。实验表明,所提方法对于轴承故障的识别效率较高,在面对多类型轴承故障情况下能够做到高效精确识别。The vibration signal noise and complex characteristics of mechanical bearings have a negative impact on fault detection,leading to a decrease in fault detection accuracy.Therefore,a research is conducted on intelligent fault recognition methods for mechanical bearings based on convolutional neural networks.The wavelet packet analysis method is used to perform multi-level analysis and processing of the bearing vibration signal,and signal reconstruction is achieved with the binary transform method.The convolutional layer in the convolutional neural network is used to perform local convolution on the reconstructed signal,and signal feature extraction is completed through pooling operation.The signal classification function based on SVDD is used to determine whether the signal sample is a fault signal,achieving intelligent fault recognition of mechanical bearings.Experiments show that the proposed method has high efficiency in identifying bearing faults and can achieve efficient and accurate recognition in the case of multiple types of bearing faults with good performance.

关 键 词：卷积神经网络 机械轴承 故障智能识别 小波包分析法 二进制变换 池化运算 SVDD 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TH133[自动化与计算机技术—控制科学与工程]