基于邻域属性重要度的齿轮箱故障特征优选方法

为有效降低齿轮箱故障特征的维数并提高诊断效率，提出了基于邻域属性重要度的齿轮箱故障特征优选方法，并结合支持向量机和BP神经网络对诊断的准确率和时间进行对比分析。以齿轮箱中不同裂纹齿轮为对象，选取能够表征齿轮箱故障状态的时域、频域和基于希尔伯特变换的36个特征。构造基于邻域模型的前向贪心数值属性约简算法进行特征优选，提取属性重要度较大的9个特征组成最优特征子集，数据压缩量达到75%，并输入支持向量机分类器中进行分类识别，用BP神经网络分类器进行结果的比较分析。结果表明，采用基于邻域属性重要度的齿轮箱故障特征优选方法，既可以在降低特征维数的情况下有效地表征齿轮箱的运行状态，又可以提高诊断的精确度和诊断效率。

Optimal selection method of gearbox fault feature based on neighborhood attribute importance

In order to reduce the dimension of gearbox fault feature and improve the efficiency of fault diagnosis，the optimal fault feature selection method of gearbox was proposed based on neighborhood attribute importance，and the support vector machine and BP neural network were used to analyze the diagnosis accuracy and time.The 36 features of the gears with different cracks in gearbox were selected based on the time-domain，frequency-domain and hilbert transform，which can be used to characterize the fault state of the gearbox.A forward -greedy numerical attribute reduction algorithm based on neighborhood model was established to select the optimal features.The nine features with higher attribute importance were selected as the optimal feature set and input into the support vector machine classifier for identification，where the data compression amount reach to 75%.The BP neural network classifier was used for a comparative analysis of the results.Results show that using the presented feature optimal selection method can reduce the feature dimension，characterize the gearbox running status effectively，and improve the diagnosis accuracy and efficiency.