定子绕组匝间短路是影响永磁牵引电机安全稳定运行的主要故障之一,受运行工况、供电与电机本体不平衡的影响,现有方法难以实现永磁牵引电机匝间短路在线精准评估,这成为永磁电机推广应用迫切需要解决的关键技术难题.因此,文章提出一种基于多特征融合的深度高斯过程永磁牵引电机匝间短路分级评估方法:首先通过建立永磁牵引电机匝间短路故障模型,提取电流不平衡、电流三次谐波与dq电流的二次谐波特征;然后采用一种双随机变分推断深度高斯过程(Doubly Stochastic Deep Gaussian Processes,DSDGP)方法对提取特征进行融合训练建模,实现永磁牵引电机匝间短路劣化状态在线分级评估;最后通过永磁电机匝间短路试验与现场案例进行算法验证.结果表明,文章所提方法在多特征融合条件下的评估准确率达到95%以上,相较于支持向量机(support vector machine,SVM)和反向传播神经网络(back-propagation neural,BPN)等分类方法,具有准确率高,适用于变工况、小样本的工程实际应用环境等优点,解决了永磁牵引电机匝间短路早期故障检测及故障严重程度评估的行业难题.
Grading evaluation method for inter-turn short circuit of permanent magnet traction motor based on deep Gaussian processes
The inter-turn short circuit of stator winding is one of the main faults affecting the safe and stable operation of permanent magnet traction motors.Affected by operating conditions,power supply and motor body imbalance,existing methods are difficult to achieve online accurate evaluation of inter-turn short circuit of permanent magnet traction motors.This has become a key technical prob-lem that urgently needs to be solved for the widespread application of permanent magnet motors.Therefore,this paper proposed a grad-ing evaluation method for inter-turn short circuit in permanent magnet traction motors based on multi-feature fusion of deep Gaussian processes.Firstly,by establishing a fault model for inter-turn short circuit in permanent magnet traction motors,features such as current unbalance,third harmonic currents and second harmonic feature of dq currents were extracted.Then,a doubly stochastic deep Gaussian processes(DSDGP)method was adopted to fuse and train the extracted features,achieving online grading evaluation of inter-turn short circuit degradation in permanent magnet traction motors.Finally,the algorithm was validated through permanent magnet motor inter-turn short circuit tests and field cases.The results show that the proposed method achieves an evaluation accuracy of over 95%under the con-dition of multi-feature fusion.Compared with classification methods such as support vector machine(SVM)and back-oropagation neural network(BPN),it exhibits high accuracy and suitability for engineering practical environments with variable operating conditions and small samples,addressing industry challenges in early fault detection and severity evaluation of inter-turn short circuits in permanent magnet traction motors.
permanent magnet traction motormulti-feature fusioninter-turn short circuitgraded evaluationdeep Gaussian processes
NETL
NSTL
万方数据
