Discrete data and errors in some conditions bring great difficulties to data analysis, and even seriously destroy the data and bring wrong analysis results, which does exist under ideal conditions. Median is a very robust parameter in data profiling that is close to the uniform value of the data set when the data is not discrete.
When the data is discrete, the uniform value is far from the true value, while the median does not deviate from the true value, indicating that the median is robust. Therefore, the difference between the median war means can be used to determine whether discrete data can exist in the data.
Apply the improved HuberM method to process the data, if the uniform value is getting closer to the median, to clarify that there is no discrete data in the data; If the uniform value deviates more and more from the median, it indicates that there is discrete data in the data.
The degree of significance depends on the data request, according to modern statistics, the recommended value range is 0~0.1. Some of these theoretical needs are improved, such as the similarity between median and uniform value, how to determine the significance of the data, etc., there is still no theoretical basis.
According to the results of the improved HuberM method on the vibration data of C rolling bearings and the frictional moment data of A and B rolling bearings, it can be seen that within the significance degree 0~0.1, the variance of C rolling bearing vibration data is complex, and the variance of A and B rolling bearing frictional moment data is monotonically reduced.
