【目的】针对永磁同步电机(PMSM)系统开路故障诊断方法中电流检测法在利用归一化电流平均值进行开路故障检测和定位时需要根据经验设置故障诊断阈值的问题，本文提出了一种基于电流矢量分析的自适应诊断阈值方法。【方法】通过Park归一化电流平均值对绝缘栅双极晶体管(IGBT)开路故障进行诊断。首先，根据归一化三相定子电流的平均值得到了故障检测变量，利用归一化三相定子电流的绝对平均值建立故障诊断自适应阈值，从而确定故障诊断准则，实现故障的检测和定位。然后，介绍了冠豪猪优化(CPO)算法和最小二乘支持向量机(LSSVM)，并以LSSVM作为基础故障分类模型，把故障检测变量作为故障分类模型的故障特征向量，利用CPO算法对LSSVM分类模型进行优化。最后，基于Matlab/Simulink搭建IGBT开路故障诊断的仿真模型，并利用优化后的故障分类模型对本文所研究的四类故障进行分类预测。【结果】仿真结果表明，在IGBT发生开路故障后由Park矢量模值归一化的三相定子电流的绝对平均值所建立的故障诊断自适应阈值会随着不同开路故障类型中故障IGBT的位置自适应地发生变化，经CPO算法优化后的LSSVM分类模型的故障分类准确率达到了99.21%。【结论】本文所提方法不仅能够弥补电流检测法的不足，还具有较高的故障分类准确率，在IGBT开路故障分类方面具有明显的优势且能够获得最佳故障诊断性能。

[Objective] To address the problem that the current detection method in open-circuit fault diagnosis for permanent magnet synchronous motor (PMSM) system needs to set the fault diagnosis threshold empirically when using the normalized current average value to detect and locate the open-circuit fault, this paper proposes an adaptive diagnostic threshold method based on the current vector analysis. [Method] The open-circuit fault of insulated gate bipolar transistor (IGBT) was diagnosed by Park normalized current average. The fault detection variables were obtained from the average values of normalized three-phase stator currents. An adaptive diagnostic threshold was then established using the absolute average value of these currents, which determined the fault diagnosis criterion and enabled fault detection and localization. The Crested Porcupine optimization (CPO) algorithm and the least squares support vector machine (LSSVM) were introduced. The LSSVM served as the basic fault classification model, with fault detection variables used as its fault feature vectors. The CPO algorithm was utilized to optimize the LSSVM classification model. A simulation model of IGBT open-circuit fault diagnosis was built using Matlab/Simulink, and the optimized fault classification model was used to classify and predict the four types of faults studied in this paper. [Results] The simulation results showed that the adaptive diagnostic threshold established using the absolute average values of three-phase stator currents normalized by Park vector modulus after the IGBT open-circuit fault occurred would change adaptively with the location of the fault IGBT in different types of open-circuit faults. The fault classification accuracy of the LSSVM classification model optimized by the CPO algorithm reached 99.21%. [Conclusion] The method proposed in this paper can not only address the shortcomings of the current detection method, but also achieve high fault classification accuracy. It offers significant advantages in IGBT open-circuit fault classification and provides the best fault diagnosis performance.

安徽省教育厅基金重大项目(022AH040110);国家自然科学基金青年项目(51905003)