中文摘要：

振动信号是反应齿轮故障的显著信号之一，将测得的振动信号进行小波分析变换到时频域，对其高频成分加以提取并进行Hilbert包络功率谱分析，发现通过识别边频带成分来进行齿轮故障诊断的方法并不理想．对测得的振动信号进行小波包分析变换到时频域，对故障频率变化明显的频段进行重构，对重构信号进行谱分析并提取特征能量．建立BP神经网络，以提取的特征能量作为网络输入量，进行故障识别，实验分析结果表明该方法取得了较好的实验效果．

英文摘要：

The vibration signal is one of the significant signals that reflects the gear fault. In this paper the measured vibration signal is transformed to the time and frequency domain by the wavelet analysis, the high frequency components of the gear vibration signal is extracted and analyzed by Hilbert envelope power spectrum. It is proved that this is not ideal to diagnose the gear fault by identifying the sidebands ingredient. The wavelet packet analysis is used to transform the measured vibration signal to the time and frequency domain, in order to reconstruct the frequency band, analyze the reconstructed signal and extract the energy feature, establish the BP neural network, use the characteristic energy as a network input, and identify the fault. The results show that the method gets better experimental results.