摘要
针对轴承等机械部件的退化状态识别问题,提出一种多维退化特征的退化状态GG聚类方法。首先分析谱熵参数在复杂性表征以及运算速度方面的优势,提出基于谱熵的性能退化特征。考虑退化状态在时间尺度的连续性,将时间参数映射到指数函数中,形成更符合性能退化规律的“弯曲时间参数”,并与谱熵、有效值构成性能退化过程的三维特征向量。最后,采用GG模糊聚类方法对性能退化状态进行阶段划分,识别不同的退化状态。在分类系数和平均模糊熵的基础上,提出并采用序列离散度评估聚类的时间聚集度。采用来自IMS轴承实验中心的全寿命试验数据进行实例分析,结果表明:提出的三维特征向量既能反映性能退化趋势,又能体现同一状态在时间尺度上的连续性,能够较好地识别轴承性能退化过程的不同阶段。
Aiming at the degradation condition recognition issue of mechanical components such as bearings,a degradation condition Gath-Geva(GG)fuzzy clustering method using multi-dimensional degradation features is proposed in this paper.Firstly,the advantages of spectral entropy parameter in complexity character and computational speed are analyzed,and performance degradation character based on spectral entropy is proposed.Considering the continuity of the degradation condition on time scale,the time parameter is mapped into exponential function to form a“bending time parameter”which is more in line with the performance degradation law,and spectral entropy and root mean square(RMS)value constitute the three-dimensional feature vector of the performance degradation process.Finally,GG fuzzy clustering method is used to classify the performance degradation condition and identify different degradation conditions.Based on the classification coefficient and the average fuzzy entropy,the time dispersion degree of clustering is proposed and evaluated by sequence dispersion.The example analysis is carried out by using the life test data from IMS(intelligent maintenance system)bearing test center.The results show that the proposed three-dimensional feature vector is able to reflect the performance degradation trend and the continuity of the same condition on time scale,and different degradation conditions can be recognized.
作者
王微
胡雄
王冰
孙德建
WANG Wei;HU Xiong;WANG Bing;SUN Dejian(Logistic Engineering College,Shanghai Maritime University,Shanghai 201306,China)
出处
《中国工程机械学报》
北大核心
2020年第2期95-100,共6页
Chinese Journal of Construction Machinery
基金
国家高技术研究发展计划(863计划)资助项目(2013AA041106)。
关键词
谱熵
GG模糊聚类
滚动轴承
弯曲时间
特征提取
spectrum entropy
GG fuzzy clustering
rolling bearing
curved time
feature extraction
