摘要
定量诊断技术对实现故障精确辨识与运维优化具有重要价值,针对使用传统特征的机器学习诊断方法因时域统计特征对工况敏感性强且频谱幅值特征提取困难、难以定量分析等问题,对其定量特征进行了研究。首先,基于故障机理与电机三相电流数学表达式进行了Park矢量特征提取,基于Park矢量比提取了匝间短路故障特征,基于Park矢量模平方函数低次偶阶转差特征提取了转子断条故障特征;然后,进行了异步电机典型故障建模分析,分别对健康电机、匝间短路故障电机以及转子断条故障电机进行了建模;基于精细模型结果,对该特征的可行性进行了验证;最后,搭建了异步电机故障模拟试验台并获取了实测数据,结合随机森林算法,对实测信号下两种典型故障进行了定量诊断实验。研究结果表明:与传统统计特征相比,该新特征的优势在于能够解决上述两种典型故障的故障特征频率与电流基频难以区分的问题,并能够很好地对上述典型故障进行量化表征;结合随机森林分类器,该方法的定量诊断准确率可达到95%以上。该定量特征可以作为匝间短路以及转子断条故障的定量诊断指标。
Quantitative diagnosis technology is of great value for achieving precise fault identification and maintenance optimization.Aiming at the problems of traditional machine learning diagnostic methods using conventional features,which are highly sensitive to operating conditions in time-domain statistical characteristics and have difficulties in extracting spectral amplitude features and doing quantitative analysis,the quantitative features were investigated.Firstly,based on the fault mechanism and the mathematical expression of three-phase motor currents,Park vector features were extracted:the Park vector ratio was used to extract inter-turn short-circuit fault features,and the low-order even-slip frequency components of the Park vector modulus square function were employed to extract rotor bar breakage fault features.Then,typical fault modeling analysis of induction motors was conducted,establishing models for healthy motors,inter-turn shortcircuits fault motors,and rotor bar breakage fault motors respectively.The feasibility of these features was verified based on detailed modeling results.Finally,an induction motor fault simulation test bench was built to obtain experimental data,and quantitative diagnosis experiments for two typical faults were implemented by combining the random forest algorithm with the measured signals.The research results show that compared with traditional statistical features,the new features can effectively solve the problem of difficult differentiation between fault characteristic frequencies and fundamental current frequencies for these two typical faults,and can well quantify these typical faults.Combined with the random forest(RF)classifier,the quantitative diagnostic accuracy of this method can reach over 95%,and these features can serve as quantitative diagnostic indicators for both inter-turn short-circuit and rotor bar breakage faults.
作者
刘昊航
程卫东
王天杨
LIU Haohang;CHENG Weidong;WANG Tianyang(School of Mechanical,Electronic and Control Engineering,Beijing Jiaotong University,Beijing 100044,China;Department of Mechanical Engineering,Tsinghua University,Beijing 100084,China)
出处
《机电工程》
北大核心
2026年第2期227-237,共11页
Journal of Mechanical & Electrical Engineering
基金
国家自然科学基金重大研究计划集成项目(92360308)。
关键词
异步电机
故障诊断
定子绕组
转子断条
匝间短路
随机森林算法
低次偶阶转差特征
asynchronous motor
fault diagnosis
stator winding
broken bar
interturn short circuit
random forest(RF)algorithm
loworder even slip characteristics
