基于人工神经元的电机断条故障诊断方法

The Fault Diagnosis of Cage Motor Broken Bars Based on ANN

笼型电动机的转子常发生断条故障,由于电气三相不平衡,在定子电流中感应出(1±2s)f1谐波分量,这为检测电机转子故障提供了方便。但是,因为转差率s很小,故障的谐波信号分量与基波分量很接近,而且由于基波分量很大,常常掩盖故障信号。基于自适应噪声干扰对消原理,提出了一种用人工神经元进行电机断条故障信号提取的方法。该方法将基波部分作为要消除的部分,采用与基波电流同源的电压经过神经元的计算,将电机定子的基波电流滤除大部分,从而可以明显地发现电机断条故障的信号。由于很难获得合适的神经元学习率及动量系数,本文提出了使用改进的遗传算法对神经元学习率及动量系数进行择优的算法。实验中采用了Y100-L4-2.2KW型电机,在人为使转子产生断条故障情况下,使用霍尔电流传感器与电压传感器分别测量了故障电机的电流和电压,采取FFT对电流进行计算,实验结果验证了本文提出方法的正确性。

The fault of broken bars in cage motors often happens. The （1±2s）f1 harmonic current of stator is induced by 3-phase electric imbalance. It is convenient for diagnosing fault of motor＇s rotor, because s is very small, the harmonic current is close to basal current and basic current is very big, harmonic current is often covered by basic current. Based on adaptive noise canceling principle, an approach of detecting motor＇s rotor fault is proposed by artificial neuron （AN）. In the method, basic current is to be canceled. The voltage with same fountain of basic current is computed by AN to cancel great part of basic current. So the fault is easy to be found. However, the learning rate and momentum coefficient of AN are difficulty to find. In the paper, improved GA is proposed to select the best learning rate and momentum coefficient of AN. Y100-L4-2.2kW motor, the Hall current sensor and the Hall voltage sensor are adopted in the experiment. The current and voltage of fault motor are measured under artificial broken bars. Current is computed by FFT. The experiment result validates the effectiveness of the proposed approach.