一种不平衡数据下的新型轻量级轴承故障诊断模型

A New Lightweight Bearing Fault Diagnosis Model Under Unbalanced Data

在实际工程场景中,轴承故障数据远少于正常数据,不平衡数据下的轴承故障诊断方法存在参数多、模型复杂的问题。因此,提出了一种由带梯度惩罚的辅助分类瓦瑟斯坦生成对抗网络(auxiliary classier Wasserstein generative adversarial network with gradient penalty,ACWGAN-GP)和神经回路策略故障诊断(neural circuit policy-fault diagnosis,NCP-FD)网络组成的不平衡数据NCP-FD(unbalanced data NCP-FD,UDNCP-FD)模型。首先,将轴承故障信号数据转换为二维时频图,利用不平衡的训练集训练ACWGAN-GP生成高质量故障样本,并将其扩充到原始训练集中;然后,将扩充后的训练集输入到NCP-FD网络中进行学习;最后,将训练好的NCP-FD网络应用于测试集进行故障诊断。实验结果表明,所提模型与其他不平衡数据下的轴承故障诊断模型相比,参数更少,存储空间更小,故障诊断准确率更高。

In the actual engineering scenario,the bearing fault data is much less than the normal data,and the bearing fault diagnosis methods under unbalanced data has many parameters and complex models.Therefore,a unbalanced data neural circuit policy-fault diagnosis(UDNCP-FD)model,which is composed of auxiliary classier Wasserstein generative adversarial network with gradient penalty(ACWGAN-GP)and neural circuit policy-fault diagnosis(NCP-FD)network,is proposed.Firstly,the bearing fault signal data is transformed into a two-dimensional time-frequency map,and the unbalanced training set is used to train ACWGAN-GP to generate high-quality fault samples,which are expanded to the original training set.Then,the expanded training set is input into the NCP-FD network for learning.Finally,the trained NCP-FD network is applied to the test set for fault diagnosis.The experimental results show that the proposed model has fewer parameters,less storage space and higher fault diagnosis accuracy than other bearing fault diagnosis models under unbalanced data.