摘要
智能故障诊断因为其高效、准确的特点,近些年来受到越来越多的关注。然而,在实际的工业应用中,工作负载是会发生变化的,并且健康状态下的数据远多于故障数据,这就导致了故障诊断准确率的降低。这里提出了一种改进的深度残差网络的模型来解决这个问题。首先,通过小波变换将振动信号转换为时频图片。其次,采用空间变换网络和注意力机制以提高分类的准确性。然后,使用Sigmoid Linear Unit(Silu)激活函数代替Rectified Linear Unit(Relu)激活函数。最后,使用类平衡损失函数解决数据类型不均衡的问题。实验通过设置多个不平衡数据集,并结合变工况条件,对改进后的ResNet模型进行验证。实验结果表明,这里所提出的方法结果优于其他方法,且具有良好的鲁棒性和泛化能力。
Intelligent fault diagnosis has received increasing attention in recent years because of its efficiency and accuracy.How-ever,in real industrial applications,the workload is subject to change and there is much more data in healthy states than in faulty ones,which leads to a reduction in fault diagnosis accuracy.An improved model of deep residual networks is proposed to solve this problem.Firstly,the vibration signal is converted into time-frequency images by wavelet transform.Secondly,a spatial transfor-mation network and an attention mechanism are used to improve the classification accuracy.Then,the Silu activation function is used instead of the relu activation function.Finally,the class balance loss function is used to solve the problem of uneven data types.The improved ResNet model is verified by setting multiple unbalances and combining with variable working conditions.Ex-perimental results show that the proposed method outperforms other methods,and has good robustness and generalization ability.
作者
田嘉野
梁朋飞
袁晓明
张立杰
TIAN Jiaye;LIANG Pengfei;YUAN Xiaoming;ZHANG Lijie(Yanshan University,College of Mechanical Engineering,Heavy Machinery Fluid Power Transmission and Control in Hebei Province,Hebei Qinhuangdao 066004,China;Agricultural University of Hebei,College of Mechanical and Electrical Engineering,Hebei Baoding 071001,China)
出处
《机械设计与制造》
北大核心
2025年第6期332-337,共6页
Machinery Design & Manufacture
基金
河北省自然科学基金(E2022203022)。
关键词
深度残差网络
不平衡数据
变工况
故障诊断
Deep Residual Network
Unbalanced Data
Variable Working Conditions
Fault Diagnosis
