Outlier detection and data processing of bearing dimension detection data

HE Gao-qing, XIAO Jian
(School of Mechanical Engineering, Hefei University of Technology, Hefei 23009, China)

Abstract: Aiming at the problems of the existence of outliers and data fluctuations in bearing dimension detection system, the method of outliers detection and the distribution characteristics of data of bearing dimension detection were analyzed, the causes of outliers and data fluctuation were summarized, a method based on boxplot theory combining outliers detection with least square polynomial fitting was proposed. Firstly, the boxplot theory was used to filtrate the outliers. Then the median of the detection data was used to substitute outliers. The least square polynomial fitting method was used to correct the abnormal data, and the detection data was reassessed by this method. Finally, the test was verified by bearing detection equipment. The experimental results indicate that this method can identify the outliers in the bearing dimension detection data quickly and efficiently, and reduce the fluctuation of the detection data effectively. The outlier recognition rate of boxplot method is 7.5%, 2.3% higher than that of 3σ method. The least square polynomial fitting method can reduce the fluctuations of detection data by 50% and improve the accuracy of detection results significantly.
Key words: bearing dimension detection; outliers; boxplot; median; least square polynomial fit