基于ARMA和遗传算法优化的BP神经网络电动机断条故障诊断

Fault Diagnosis of BP Neural Network Based on ARMA and Genetic Algorithm Optimization

为判断鼠笼式三相异步电动机转子断条故障情况,提出了一种利用定子电流信号,基于ARMA和遗传算法优化的BP神经网络的诊断方法。首先,使用改进的ARMA算法对电动机的定子电流波形进行拟合,将自回归系数模型系数提取出来,作为表征电动机故障的特征向量,并分为训练集和测试集。然后利用遗传算法优化BP神经网络的初始阈值和权值,以避免BP神经网络陷入局部极值点的问题。再用训练集对BP神经网络进行训练,用训练好的神经网络对测试集进行判断。实验结果显示,ARMA模型可较好地对三相异步电动机定子电流波形进行拟合,BP神经网络可较为准确地判断特征向量表征的故障情况,此方法具有较好的诊断结果。

In order to recognize the broken rotor bar fault of squirrel cage asynchronous motors, a method of diagnosis was proposed using stator current signal, based on ARMA model and BP neural network optimize by genetic algorithm. Firstly, fit stator current signal waveform with ARMA algorithm, and use parameters of auto- regressive model in ARMA as eigenvectors representing motor fault, and divide these eigenvectors into training set and testing set. Optimize original threshold and weight in BP neural network to avoid the problem that the network was trapped into local extremum. Train the neural network with training set and do recognition with the testing set. The experiment shows that, ARMA model has the ability to fit stator current signal waveform well, and the BP neural network has the ability to accurately recognize faults represented in eigenvectors. This method has a high diagnosis accuracy.