【作者】 王旭红；

【导师】 何怡刚；

【作者基本信息】 湖南大学 ， 电气工程， 2012， 博士

【摘要】 定子绕组匝间短路故障是异步电机最常见的故障之一，该故障由作用在定子绕组上的综合应力共同作用产生，如热应力、电应力、机械应力和环境应力等。匝间短路故障产生的大电流和过热将导致更为严重的相间、匝间或对地故障，对绕组和铁芯将产生灾难性的损害。因此，快速而准确地诊断电机定子绕组匝间短路故障十分重要。本文主要目的是研究定子绕组匝间短路的诊断方法，主要内容如下：根据电机的电磁原理，从正常电机的模型出发，通过坐标变换，推导出含定子绕组匝间故障的异步电机模型，通过状态变量表示其动态方程并进行数值仿真。提出基于改进故障模型的匝间故障诊断方法，采用故障模型计算出电机电流，将其与实测电流相比较而得到电流计算误差，将该误差分解为两部分，一部分用于检测匝间故障，另一部分用于不同运行状态下修正电机模型的角速度参数。无论故障发生在哪相，该方法都能快速诊断早期故障。提出基于扩展Park变换和模糊神经网络的匝间故障诊断方法。电机定子匝间故障的严重程度，受到负载、供电电压平衡度的影响，这些因素之间呈现出一种模糊关系。模糊神经网络具有结构简单、收敛快速、易于实施的特性，能实现电机故障征兆与故障模式之间的非线性映射。定子绕组发生匝间故障时，其电流的Park矢量模将发生变化，用电机定子匝间故障的Park矢量模的两倍基频分量作为故障特征，与电机负载、电压不平衡度一起构建模糊神经网络模型，进行电机定子匝间故障的诊断。提出一种基于递归小波神经网络的异步电机定子绕组匝间故障诊断方法。定子匝间故障从开始发生到严重短路，是一个缓慢变化的过程，及早对匝间短路进行监测，有利于确定电机容错运行的时间，以便合理、经济地进行停机检修。该方法采用两个对角递归小波神经网络分别用于诊断匝间故障，一个用于估算故障严重度，另一个用于确定短路故障匝数。为了克服BP算法收敛慢、易陷入局部极小值的缺陷，通过LM学习算法提高神经网络的收敛速度。为诊断并确定匝间短路故障所在相，提出基于故障模型和BP神经网络的异步电机定子绕组匝间短路故障定位方法。以三相电流和电压的相位移作为故障特征，通过神经网络判断故障发生相，并采用遗传算法优化神经网络参数以提高其动态处理能力。本文提出的异步电机定子绕组匝间短路故障诊断方法均经过了仿真及试验验证，具有一定的工程实用价值。更多还原

【Abstract】 Induction motors play an important role as prime movers in manufacturing,process industry, agricultural production and transportation. With the increase inproduction capabilities of modern manufacturing systems, capacity of a single motorkeeps increasing and the load also becomes more complicated. Unexpected downtimedue to machinery failures has become more costly than before. The faults of inductionmotors may not only cause the interruption of product operation but also increasemaintenance costs, decrease product quality and affect the safety of operators. Statorwindings short circuit is one of the most common faults in electric machines. Thistype of fault is caused by the combination of various stresses acting on the stator, suchas thermal, electrical, mechanical, and environmental stresses. Such fault produceshigh currents and winding overheating resulting in severe phase-to-phase, turn-to-turnor turn-to-ground faults. It may produce a catastrophic damage in the windings or inthe motor core. For this reason, rapid and accurate detection of incipient faultsbetween turns during motor operation is very important. Therefore, the objective ofthis thesis is to study the on-line turn fault detection methods. The main contents andachievements of the thesis are as follows:According to the electromagnetic theory and the healthy model, a transientmodel for an induction machine with stator winding turn faults is derived usingreference frame transformation theory. The dynamic equations is presented instate-space form, which is suitable for digital simulation. A model-based strategy forstator inter-turn short circuit detection on induction motors is proposed. The currentestimation error is used as a state observer for the incipient detection of the inter-turnfault. By measuring only stator voltages and currents,the current for the observer canbe calculated. The current estimation error between the measurement and calculationis decomposed into two parts: one part is used for the faulty severity, the other is usedin an adaptive scheme for speed estimation. The proposed technique is able to rapidlydetect incipient faults, independently of the phase in which the fault occurs. And theobserver includes an adaptive scheme for rotor-speed estimation, avoiding the use of aspeed sensor.Stator inter-turn short circuit fault in the induction motors is affected by theloads and unbalanced supply voltages. The relationship between them is usually uncertain. Fuzzy neural network has the ability to achieve nonlinear dynamicmappings with simple structure, rapidly convergence and easily implementation,therefore it can be used to detect stator winding turn fault in induction motors atvarious conditions. When inter-turn short circuit occurs in stator winding, the Parkmodule will change. In order to detect inter-turns accurately, a method based onextended-Park transform and fuzzy neural network is presented. By spectrum analysis,the ratio between the magnitude of2f1component and the Park module can serves asfaulty characteristic.Then a model which take load and phase voltage unbalance intoconsideration is constructed based on fuzzy-neural network for detection of statorinter-turn short circuit fault.The development of stator inter-turn fault is a dynamic and slow processfrom the incipience to severe. Recurrent wavelet neural network based on-line statorwinding turn fault detection approach for induction motors has been done. In theapproach, two recurrent wavelet neural networks are employed to detect inter-turnfault, one is used to estimate the fault severity, the other is used to determine the exactnumber of fault turns. In order to overcome the lacks of BP algorithm such as slowconvergence, non-stability of convergence and local minimum problem In the courseof training, Levenberg-Marquardt (LM) algorithm is introduced to make therecurrent wavelet neural network network converging more quickly.In order to detect and locate an inter-turn short circuit fault on the statorwindings of induction motors, an approach based on faulty model and BP neuralnetwork is presented. The diagnostic process is achieved through monitoringsimultaneously the values of the three-phase shifts between the line current and thephase voltage of the machine. Genetic algorithm is adopted to optimize the BPNNparameters for improving dynamical processing ability.All the proposed models and methods are verified through simulations andexperiments. The results demonstrate great engineering value.更多还原