基于残差卷积网络的多传感器融合永磁同步电机故障诊断

Fault diagnosis of multi-sensor fusion permanent magnet synchronous motor based on residual convolutional neural network

作为工业生产与日常生活的常见设备,永磁同步电机的故障诊断研究具有十分重要的意义。以永磁同步电机的匝间短路、退磁、轴承故障为诊断目标,提出一种新型的多传感器特征融合网络(MSFFN),结合多传感器融合技术与卷积神经网络实现永磁同步电机的可靠故障诊断。网络采用2个带有残差模块的卷积神经网络,对输入的电流信号与振动信号并行提取隐藏特征,并设计一种中间特征融合模块(IFFM)有效融合电流和振动的各层隐藏特征,IFFM基于注意力机制对网络中的电流特征与振动特征进行筛选,自适应关注不同信号的内在相关特征,以实现更好的诊断效果。搭建了故障样机测试平台进行数据采集与实验验证,实验结果表明,提出方法具有更高的诊断准确率,同时在叠加了强噪声的条件下,具备更强的抗干扰能力。

Permanent magnet synchronous motor(PMSM)is a widely used equipment in industrial production and daily life,research on PMSM fault diagnosis is of great significance.Aiming at the diagnosis of inter-turn short circuit,demagnetization and bearing fault of PMSM,a new type of multi-sensor feature fusion network was proposed,which combines multi-sensor fusion technology and convolutional neural network to achieve reliable fault diagnosis.In the network,two convolutional neural networks were used with residual blocks to extract features from current and vibration,and an intermediate feature fusion module(IFFM)was proposed to fuse the multilayer features of current and vibration.IFFM is based on attention mechanism to screen the current and vibration features in the network,adaptively focusing on the intrinsic correlation features of different signals,in order to achieve better diagnostic performance.A motor fault test platform was built for data acquisition and experimental verification.The experiments show that compared with other methods,the proposed method exhibits a higher diagnostic accuracy and demonstrates stronger robustness against interference particularly under conditions with strong noise.