多通道电流信号深度特征融合的开关磁阻电机故障诊断研究

Fault diagnosis of switched reluctance motor based on deep feature fusion of multi-channel current signals

提出一种基于卷积神经网络(convolution neural networks,CNN)与Transformer结合的新型深度学习框架1D-Uniformer(one dimensional unified transformer),解决开关磁阻电机高阻接触故障和相间短路故障的分类识别问题。搭建开关磁阻电机故障诊断实验平台,在开关磁阻电机定子绕组上设置高阻接触故障和相间短路故障,并通过非侵入式方法采集电机的三相电流信号;引入动态位置嵌入、多头关系聚合器和前馈网络对传统CNN进行改进,得到1D-Uniformer以充分提取高阻接触故障和相间短路故障的特征。实验结果表明:该模型在高阻接触故障和相间短路故障诊断方面均具有很好的分类效果,在18种故障状态下识别精度能达到100%,在不同的噪声强度下仍然具有较高的鲁棒性。

This paper proposes a new deep learning framework One Dimensional Unified transformer(1D-Uniformer)based on the integration of convolution neural networks(CNN)with Transformer to address the classification and identification of high resistance connection faults and interphase short circuit faults in switched reluctance motors.First,an experimental platform for fault diagnosis of switched reluctance motors is built to set up high resistance connection faults and interphase short circuit faults on the stator windings of switched reluctance motors,and the three-phase current signals of the motors are collected by a non-intrusive method.Then,the dynamic position embedding,the multicollinear relationship aggregator,and feed-forward networks are introduced to improve the traditional CNN,and the 1D-Uniformer is obtained to sufficiently extract the features of the high resistance connection faults and interphase short circuit faults.Our experimental results indicate the model achieves impressive classification in both high resistance connection faults and interphase short circuit faults diagnosis and reaches an accuracy of 100%in the recognition of 18 types of faults.Additionally,it achieves an excellent robustness under different noise conditions.