基于孤立森林-控制图与多特征联合的滚动轴承异常状态检测  

Abnormal Condition Detection of Rolling Bearings Based on Isolated Forest-Control Chart and Multi Feature Joint

作　　者：何伟挺 米俊芃 侯耀春 赵奂芃 黄文君[2] HE Weiting;MI Junpeng;HOU Yaochun;ZHAO Huanpeng;HUANG Wenjun(Zhejiang Supcon Technology Co.,Ltd.,Hangzhou 310053,China;Zhejiang University,Hangzhou 310027,China)

机构地区：[1]浙江中控技术股份有限公司,杭州310053 [2]浙江大学,杭州310027

出　　处：《轴承》2023年第5期90-98,共9页Bearing

基　　金：浙江省科技攻关计划资助项目(2022C01047)。

摘　　要：为保证滚动轴承异常状态检测的及时性并降低虚警率,提出了一种基于孤立森林的多指标改进CUSUM异常检测方法(IF-CUSUM)。首先,对轴承原始振动序列进行小波去噪;然后,基于孤立森林算法去除振动序列的离群点;最后,结合RMS和峭度等指标设计阈值分配方案,给出异常状态检测判定依据。通过60组不同振动演化规律、脉冲强度和噪声等级的滚动轴承故障仿真数据以及FEMTO-ST轴承数据集进行算法验证,结果表明,与基于单一RMS指标、峭度指标的异常检测算法相比,IF-CUSUM算法能较早地检测出滚动轴承运行过程中的异常点,虚警率较低,整体性能和鲁棒性较好。In order to ensure timeliness and reduce false alarm rate in abnormal condition detection of rolling bearings,an improved multi-index CUSUM anomaly detection method based on isolated forest(IF-CUSUM)is proposed.Firstly,the original vibration sequence of the bearings is denoised by wavelet.Then,the outliers of vibration sequence are removed based on isolated forest algorithm.Finally,a threshold assignment scheme is designed by combining the indexes such as RMS and kurtosis,and the judgment basis of abnormal condition detection is given.The algorithm is verified by 60 sets of rolling bearing fault simulation data with different vibration evolution laws,pulse intensities and noise levels and FEMTO-ST bearing data set.The results show that compared with anomaly detection algorithm based on a single RMS index and kurtosis index,the IF-CUSUM algorithm is able to detect the abnormal points during operation of rolling bearings earlier,having low false alarm rate,good overall performance and robustness.

关 键 词：滚动轴承 状态监测 异常检测 孤立森林 有效值 峭度 

分 类 号：TH133.33[机械工程—机械制造及自动化]