中文摘要：

利用改进的小波包对收集的信号进行特征提取，解决了小波包分解的频率混叠问题；针对故障信息中的冗余属性问题，提出了基于类差别矩阵改进属性重要度的属性约简算法，根据各条件属性在类差别矩阵中出现1的频次定义新的属性重要度，提高属性约简的效率；通过考虑条件属性与类属性间的关联性，提出了基于熵权法的属性加权朴素贝叶斯分类器算法，提高故障分类精度。通过对滚动轴承故障数据的对比分析，验证了所提组合方法在提高故障诊断正确率、快速性方面所具有的优势。

英文摘要：

An improved wavelet package was used to extract feature of collected signals and to solve the wavelet packet aliasing problem. Considering redundant attributes in fault informations, rough set attribute reduction algorithm was proposed based on class discernibility matrix and im- proved attribute significance, new attribute significance was defined according to the frequency of each condition attribute equal to 1 in the class discernibility matrix,which improved the efficiency of attrib- ute reduction. Considering the relativity among different condition attributes and class attributes, the entropy weight method-based attribute weighted naive Bayesian classifier algorithm was proposed, which improved the fault classification accuracy. By comparative analysis of roiling bearing failure da- ta, it shows that the proposed hybrid method herein has certain advantages in fault diagnosis accuracy and rapidity.