基于神经网络的感应电机故障诊断技术研究

【作者】 周慧；

【导师】 陈强；

【作者基本信息】 江西理工大学 ， 控制理论及其控制工程， 2008， 硕士

【摘要】 人工神经网络已越来越多地被用于故障诊断领域。论文以感应电机为研究对象,研究将人工神经网络应用于感应电机故障诊断中的一系列问题。在电机正常、转子断条、匝间短路、复合故障(转子断条和匝间短路情况下)四种不同工作状态下,对电机定子电流和振动信号进行采样,从采集的信号中,提取振动幅值信号的均方根值作为特征参数之一;应用“小波包-能量”法提取电流信号中的有效特征参数,将时域参数和频域参数相结合作为人工神经网络的输入矢量,对网络进行训练和测试。传统的神经网络算法应用于故障诊断时,具有易陷入局部极小值,收敛速度较慢等缺点。而生物免疫系统的一些诱人特性为人们开发新的智能算法提供了新的途径。本文将免疫进化学习算法和人工神经网络相结合设计了一个免疫神经网络。利用免疫进化学习算法对神经网络的隐层中心进行聚类,得到隐层中心值后,利用四种故障模式下的特征样本对网络进行训练和测试。利用小波网络精度高,学习速度快的特点,本文将小波理论与神经网络技术相结合。用小波函数代替传统神经网络中的激活函数,推导了小波神经网络算法,并对小波神经网络进行了加动量项的改进,改进后的算法能够有效地加快神经网络的收敛速度,避免陷入局部极小值。对以上算法进行了大量的仿真研究,仿真结果表明:利用时域方法和小波包-能量法提取特征参数是有效的;使用免疫进化学习方法与RBF神经网络相结合,训练后的网络可以对感应电机故障作出较为准确的诊断。仿真结果还表明:改进的小波网络算法用于电机故障诊断是有效的。将小波网络与BP网络仿真结果进行比较,结果表明该算法提高了处理速度,精度较高。更多还原

【Abstract】 Artificial neural networks are increasingly used in the areas of fault diagnosis.In this thesis, the research comes from the neural network,regarding induction motor as a target,and the research will use artificial neural network to solve series of induction motor fault diagnosis problems.Under the four models of composite fault (broken bar and short-circuit fault)、short-circuit fault、broken bar、normal operation.the vibration and current signals are sampled, From the collection of signals,pick up the amplitude of vibration from the motor,use rms value as one of the characteristic parameters and use"wavelet packet - energy" extract useful parameters of current signal. Combine the time domain parameters and the frequency parameters to be the input vectors of the artificial neural network.then train and test the network.Traditional neural network algorithms are easy to fall into the local minimum、slow convergence when in fault diagnosis.And the biological immune system has some attractive properties,they provide an important inspiration for people to develop new intelligent algorithm. Therefore,based on the advantages of immune algorithm,the paper combined immune evolutionary learning algorithms and artificial neural networks.use arithmetic immune neural network algorithm to cluster the hidden layer, after get the center of hidden layer, the further use of immune neural network to train four motor fault modes under teachers.Because wavelet neural network have the following traits:high precision,learning rate fast etc.so in this thesis, we combined wavelet theory and neural network,using wavelet function instead of the activation function of traditional neural networks,put forward wavelet neural network,and improve it.the improved algorithm can effectively accelerate the convergence rate of neural networks, avoid the local minimum.Series of simulating tests are conducted by algorithm in this theis,the results indicate that the effective of extracting characteristic parameters by using time domain wavelet packet - energy.combined immune evolutionary learning algorithm and RBF,after the train.the network have accurate fault diagnosis of induction motor breakdown. In addition,the results indicate that the effective of using improved wavelet network algorithm for fault motor diagnosis,compared wavelet network with BP Network,the results show that the algorithm increased the processing speed and have high accuracy.更多还原