摘要
为快速准确识别轴承的运行状态,提出了一种基于多维缩放和随机森林的轴承故障诊断方法。该方法采用函数型数据分析,得到轴承振动信号自相关函数的拟合系数,构造故障特征集;使用网格搜索法优化随机森林参数,得到特征重要性排序;然后使用多维缩放方法对特征选择后的故障特征集进行降维;最后采用随机森林对降维后的故障特征进行诊断识别。为验证所提方法的有效性,开展了正常、内圈故障、外圈故障、滚子故障状态下的轴承振动实验,结果表明,函数型数据分析的特征提取方式能有效表征不同状态轴承振动信号的不同特征,与t分布随机邻域嵌入和主分量分析方法相比,多维缩放方法具有更高的类间距和类内距的比值,且优势明显,各类状态的诊断准确率均高达100%,较使用原始特征集的随机森林平均准确率提高了5%。
For quick and accurate recognition of bearing operating condition, a fault diagnosis method based on multiple dimensional scaling and random forest is proposed. Functional data analysis is adopted to obtain the fitting coefficients of the auto-correlation function of bearing vibration signal, and a fault feature set is built. Then grid search method is employed to optimize the random forest parameters to obtain feature importance ranking, and multiple dimensional scaling is used to reduce dimension of the fault feature set selected according to the feature importance ranking. The dimension-reduced features are diagnosed and identified by the random forest. To verify the effectiveness of the proposed method, bearing vibration experiments under normal, inner ring fault, outer ring fault and roller fault conditions are conducted. The data analysis results indicate that the feature extraction method of functional data analysis can effectively characterize the different features of vibration signals of bearings under different conditions. Compared with t-distributed stochastic neighbor embedding and principal component analysis, the multiple dimensional scaling method achieves higher intra-class distance to inter-class distance ratio, and has more obvious advantages. The classification accuracy of the proposed method in all conditions is up to 100%, which is 5% higher than the average accuracy of random forest using original feature set.
作者
张西宁
张雯雯
周融通
余迪
ZHANG Xining;ZHANG Wenwen;ZHOU Rongtong;YU Di(State Key Laboratory for Manufacturing Systems Engineering, Xi’an Jiaotong University, Xi’an 710049, China)
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2019年第8期1-7,共7页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金资助项目(51275379)
国家自然科学基金创新研究群体资助项目(51421004)
关键词
函数型数据分析
多维缩放
随机森林
轴承故障诊断
functional data analysis
multiple dimensional scaling
random forest
bearing fault diagnosis
