中文摘要：

在轴承故障诊断中，故障信号的提取是一个关键问题。实际测得的轴承振动信号一般是非平稳和非高斯分布的信号，信噪比很低，微弱的故障信息往往完全淹没在噪声中，信号特征的提取非常困难。信号的高阶累积量对加性高斯噪声和对称非高斯噪声不敏感，应用在轴承的故障诊断中，可以有效地分离信号与噪声，提高信噪比，增强故障信息。对轴承在不同状态下的振动信号进行对比分析，提取了不同状态下轴承振动信号的功率谱与高阶累量谱（双谱），建立了用于故障诊断的双谱特征向量，并利用BP神经网络进行了故障诊断。分析结果表明，从高阶累积量提取的特征与功率谱相比，对故障特征比较敏感，容易实现智能诊断中的数字特征提取，可有效地区分轴承的故障。

英文摘要：

In the fault diagnosis of bearing, the extracting of fault signal is a key problem. The practical testing vibration signal of bearings is unsteady or non-Gauss distribution. It has low signal-to-noise. Usually the faint fault information is entirely in the noise, the characteristic extracting is very difficult. Higher-order cumulant spectrum is not sensitive to the adding Gauss noise and symmetry non-Gauss noise. It can separate signal and noise effectively by utilizing higher-order cumulant spectrum for fault di- agnosis of bearing. Also it can improve signal-to-noise and enhance fault information. The vibration signal of bearings in different states is contrasted and analyzed, synchronously the higher-order cumulant spectrum （bispectrum） of bearing vibration signal in different states is picked up. The analysis result is shown that the characteristic extracting from the higher-order cumulant spectrum is more sensitive to the fault of bearing than the characteristic from power spectrum. It can separate faults of bearing effectively, and carry out intelligent fault diagnosis.