基于EEMD结合二次小波包降噪的齿轮箱故障诊断

Gearbox Fault Diagnosis Based on EEMD Combined with Secondary Wavelet Packet Noise Reduction

针对齿轮箱振动信号信噪比低、故障识别精确度不高等问题,提出了聚合经验模态分解(EEMD)结合小波包二次降噪的故障诊断方法.首先,对采集到的原始信号进行小波包降噪并重构;再对第一次降噪后的信号进行EEMD分解,得到一系列的固有模态函数(IMF);并计算分解得到的每个IMF与第一次降噪后信号的相关系数,从而确定二次降噪的IMF有效集;然后,通过选择不同消失矩的db系小波,对筛选出的IMF进行二次降噪;最后,将二次降噪之后的IMF进行重构,提取特征向量输入到BP神经网络,识别齿轮箱的故障类型和位置.测试结果表明,此二次降噪方法用于齿轮箱故障诊断,识别准确率更高,在神经网络训练和测试中耗时更短.

Aiming at the low signal-to-noise ratio and the low accuracy of faults identification for rolling gearboxs vibration signal, a new method which combined Ensemble Empirical Mode Decomposition （EE- MD） with wavelet packet secondary noise reduction was proposed. Firstly, it used wavelet packet for noise reduction of the original signal, and then reconstructed the signal. Through the Ensemble Empiri- cal Mode Decomposition method, a series of intrinsicmode functions （IMF） were obtained. The correla- tion coefficients between the IMF and the reconstructed signal were computed to determine which IMFs were the effective components to be reprocessed. Then, daubechies wavelets of different vanishing mo- ments were selected for secondary noise reduction. Finally,after the secondary noise reduction the IMFs were reconstructed, and the characteristics were extracted from the reconstructed signal. Put the charac- teristics into the BP neural network to identify the type and location of the gearbox faults. The test result showed that this secondary noise reduction method for gearbox fault diagnosis took less time in neural network training and testing, and had higher recognition accuracy.