摘要
针对滚动轴承工作环境复杂、常伴有较高等级噪声所致故障难以识别以及传统神经网络在小故障样本情况下识别精度不高的问题,提出一种基于去相关谱回归(Decorrelated Spectral Regression,DSR)与改进图注意力神经网络(Graph Attention Network v2,GATv2)相结合的变工况滚动轴承故障诊断方法。首先,在数据处理过程中提取多域轴承振动信号特征,丰富原始特征集;之后利用去相关谱回归方法降低特征维度,同时消除人为筛选特征引起的选择偏差;最后采用改进的图注意力神经网络进行故障诊断。实验中使用帕德博恩大学轴承数据集,与一些经典模型对比的结果表明,所提方法具有出色的抗噪能力并且在小样本情况下仍具有较高的检测精度。
Aiming at the problems that the rolling bearing fault is difficult to be identified since the complex working environment and the fault signal are often accompanied by higher level noise,and the traditional neural network methods have low recognition accuracy in the case of small samples,this paper proposes a method based on decorrelated spectral regression(DSR)combined with graph attention network v2(GATv2)for fault diagnosis of rolling bearings under variable working conditions.Firstly,multi-domain bearing vibration signal features are extracted during data processing to enrich the original feature set.Then,the DSR method is used to reduce the feature dimension,and eliminate the selection bias caused by artificial screening features.Finally,the improved graph attention network is used for fault diagnosis.The bearing dataset of Paderborn University(PU dataset)is used to test the effectiveness of the proposed method.Comparing with some classical models shows that the proposed method has excellent noise immunity and high detection accuracy in the case of small samples.
作者
李耀华
张鑫杰
LI Yaohua;ZHANG Xinjie(Transportation Science and Engineering Institute,Civil Aviation University of China,Tianjin 300300,China)
出处
《噪声与振动控制》
北大核心
2025年第5期110-116,共7页
Noise and Vibration Control
基金
国家自然科学基金资助项目(U2033209)。
关键词
故障诊断
滚动轴承
去相关谱回归
图注意力神经网络
小样本
fault diagnosis
rolling bearing
decorrelated spectral regression
graph attention neural network
small sample
