摘要
针对轴承故障诊断模型输入信息单一,且变负载、噪声工况下诊断精度受限的问题,提出一种多尺度卷积神经网络结合自注意力特征融合机制(SA-MCNN)的故障诊断方法。该方法首先使用不同核大小的卷积层并行提取振动信号的多尺度信息后,采用自注意力特征融合机制,为并行的多尺度特征加权融合;最后根据融合后的特征,区分轴承的健康状态。实验结果表明,与其它故障诊断模型相比,SA-MCNN模型能够根据多尺度信息有效捕捉高质量的状态特征,在跨负载工况和噪声工况下表现出强鲁棒性。
Considering that the single-source input of bearing fault diagnosis models and poor diagnostic performance under variable load and noise conditions, a new model is proposed by combining the multi-scale Convolutional Neural Network and self-attention feature fusion mechanism(SA-MCNN) for bearing fault diagnosis. In SA-MCNN, the multi-scale information of vibration signals is first extracted by using different convolution kernels in parallel. Then, the multi-scale features are weighted and fused by self-attention feature fusion mechanism. Finally, the health states of bearings are distinguished according to the fused features. The experimental results verify that compared with other fault diagnosis models, the SA-MCNN model can effectively capture high-quality state features based on multi-scale information, and SA-MCNN can stay robust against cross-load or noise conditions.
作者
黄雅静
廖爱华
丁亚琦
杨洋
师蔚
胡定玉
HUANG Yajing;LIAO Aihua;DING Yaqi;YANG Yang;SHI Wei;HU Dingyu(School of Urban Railway Transportation,Shanghai University of Engineering Science,Shanghai 201620,China;The Vehicle Branch,Shanghai Metro Maintenance Guarantee Co.,Ltd.,Shanghai 200235,China;Shanghai Engineering Research Center of Vibration and Noise Control Technologies for Rail Transit,Shanghai University of Engineering Science,Shanghai 201620,China)
出处
《智能计算机与应用》
2022年第9期37-44,共8页
Intelligent Computer and Applications
基金
国家自然科学基金(51605274)
上海市地方院校能力建设项目(20030501000)。
关键词
多尺度卷积神经网络
自注意力机制
特征融合
轴承故障诊断
multi-scale Convolutional Neural Network
self-attention mechanism
feature fusion
bearing fault diagnosis
