基于混合式特征选择的滚动轴承故障诊断方法

Rolling bearing fault diagnosis based on hybrid feature selection

为降低滚动轴承故障特征集的维数,提升诊断精度,提出一种混合式特征选择方法。该方法由两个阶段构成,首先通过费舍尔分值法对原始特征集进行预排序,根据特征的费舍尔得分按照降序排序,利用得分曲线的拐点确定预选子集的范围,去除原始特征集中的无关特征;然后将遗传算法嵌入Wrapper阶段中,利用分类器的识别精度作为评价标准,从预选子集中去除冗余特征,确定最优子集。通过实验证明,该方法可以有效地用于滚动轴承不同故障类型和不同故障程度的诊断,最优子集在仅保留了关键特征的同时,识别精度得到提升。

A hybrid feature selection method is proposed to reduce the dimensionality of the rolling bearing fault feature sets and improve the fault diagnostic accuracy.The implementation of the method consists of two stages.In the former stage,the original feature sets are pre⁃ranked by the Fisher score(FS)method,the features are sorted in descending order according to their FSs,the inflection point of the score curve is used to determine the range of the pre⁃selected subsets,and the irrelevant features are removed from the original feature sets.In the later stage,the genetic algorithm(GA)is embedded into the stage of Wrapper,and the recognition accuracy of the classifier is used as the evaluation criterion to remove redundant features from the pre⁃selected subset features,so as to determine the optimal subset.The experiment proves that the proposed method can be effectively used for the diagnosis of different fault types and different fault degrees of rolling bearings,and the recognition accuracy of the optimal subset is improved while only the key features are retained.