基于有限样本的永磁同步电机退磁故障诊断方法

Demagnetization Fault Diagnosis Method for a Permanent Magnet Synchronous Motor Based on Limited Samples

针对永磁同步电机(PMSM)因样本数据稀少、可用性低、特征弱化和结构复杂等因素引发的退磁识别问题,提出一种融合稀疏自编码与最小二乘生成式对抗网络的退磁故障诊断方法。该方法首先采集PMSM的电磁转矩和磁动势分布数据构成有限样本集合,其次采用最小二乘生成式对抗网络对样本在保持特征分布一致的条件下进行标签化扩张,最后运用稀疏自编码网络和Soft max分类器对样本进行训练和分类,实现退磁故障的诊断与识别。在模型训练和故障识别过程中,一方面合理设计深度网络隐层节点、训练算法以及层数等影响学习效率的参数;另一方面训练优化网络并测试验证网络的优劣以提高故障诊断性能。经过多次试验,最终可实现PMSM退磁故障的有效诊断。

Aiming at the demagnetization identification problem of permanent magnet synchronous motor due to the sparse sample data,low availability,weak feature and complex structure,this paper proposes a demagnetization fault diagnosis method combining sparse self-encoding and least squares generative countermeasure network.This method first collects the electromagnetic torque and magnetomotive force distribution data of the permanent magnet synchro-nous motor to form a limited sample set.Secondly,the least squares generative confrontation network is used to label and expand the sample while maintaining the same feature distribution,and finally use sparse The self-encoding network and Soft max classifier train and classify the samples to realize the diagnosis and identification of demagnetization faults.In the process of model training and fault identification,on the one hand,the parameters that affect learning efficiency such as the hidden nodes of the deep network,the training algorithm and the number of layers are rationally designed;on the other hand,the optimized network is trained and tested and verified to improve the fault diagnosis performance.Af-ter many tests,the effective diagnosis of permanent magnet synchronous motor demagnetization fault was finally real-ized.