电工技术学报2025，Vol.40Issue(4)：1105-1116,12.DOI:10.19595/j.cnki.1000-6753.tces.240067
基于多路漏磁信号阵列的无刷直流电机匝间短路故障诊断
Turn-to-Turn Short Circuit Fault Diagnosis of Brushless DC Motor Based on Multiple Magnetic Flux Leakage Signal Arrays
摘要
Abstract
The traditional fault detection method cannot detect or locate inter-turn short-circuit faults.Therefore,this paper develops a new detection method by combining magnetic flux leakage and backpropagation neural networks. The variation of the leakage magnetic field originates from the stator current and radiates to the outside of the motor.Therefore,it provides information about the state of the stator winding.If a turn-to-turn short circuit fault occurs in the motor,the amplitude of the leakage flux(Å st)emanating from outside the motor changes and can be detected by the flow sensor.A backpropagation neural network is an error correction algorithm that uses multi-layer feedback based on the errors between input and output signals.Hall sensors collect magnetic flux leakage signals,backpropagation neural networks analyze and extract features from the data,and the fault location and situation can be obtained. There are four types of motor states:health,A-phase fault,B-phase fault,and C-phase fault,and the degree of fault is also divided into four types:μ=1/9,2/9,3/9,and 4/9.In the simulation,the inter-turn short circuit is modeled as a turning slot in the faulty phase.Regarding μ=1/9 and 2/9 fault modeling,the Fourier transform is applied to three-phase current,and current sensors obtain amplitude change maps of three-phase current in the time and frequency domains.Then,the Hall sensor obtains the time-domain and frequency-domain waveforms of the magnetic flux leakage signals at three positions.It is found that the frequency spectrum of the magnetic flux leakage signal contains the motor state information and can be used as input for the backpropagation neural network. Magnetic flux leakage and three-phase current signals are collected from three different positions under four different testing conditions,with the health and fault states of phase A(2/9)for analysis.Fourier analysis is performed on magnetic flux leakage signals,which are used as inputs for backpropagation neural networks to evaluate the degree and location of faults.The experiment shows that(1)the amplitude of leakage flux and current signal regularly increases with the increase of motor speed;(2)When a turn-to-turn short circuit fault occurs,the amplitude of leakage flux and current decreases with the increase of load. The training accuracy of the fault phase dataset,the four fault degree datasets(1/9-4/9),and the three different fault positions is 99.8%,98%,and 80%,respectively.The proposed method only collects the external MLF signals of the motor and then performs Fourier transform and BPNN for spectrum analysis and fault diagnosis.Therefore,it is non-intrusive and easy to implement in practice.Potentially,it has a great prospect for engineering applications as the costs of sensors and implementation are low,and the computational resources are already in place for motor drives.关键词
无刷直流电机/匝间短路/多路漏磁信号阵列/故障诊断/反向传播神经网络Key words
Brushless DC motor/inter-turn short-circuits/multiple magnetic leakage flux arrays/fault diagnosis/back propagation neural network分类
信息技术与安全科学引用本文复制引用
吴振宇,王慧,胡存刚,席浩天,曹文平..基于多路漏磁信号阵列的无刷直流电机匝间短路故障诊断[J].电工技术学报,2025,40(4):1105-1116,12.基金项目
国网安徽省电力公司科技资助项目(SGAHMA00YJJS2400635). （SGAHMA00YJJS2400635）
