The method of singular value processing for non-contact detection of rotating parts

In order to improve the efficiency of the contour size detection process of rotating parts, the methods of rotating parts, and the stepping movement of the laser sensor for non-contact detection was adopted. Due to the discontinuity of the detection area, the singularity of the detection data was inevitable, which results in the instability of the detection data. Aiming at this problem, the characteristics of data distribution are studied, the causes of singular values are summarized, and a method based on the combination of improved K-means clustering method and evaluation index was proposed. First, according to the characteristics of the data distribution, the negative data in the detection data was replaced by the adjacent value, and then the K value and the initial cluster center of the improved K-Means algorithm were calculated. Finally, the indexes of eligibility and singular value evaluation were established. The experimental results show that this method can accurately judge the qualification of the parts compared with the box plot method and 3$\sigma$ method, and effectively identify the singular value of the detection data and deal with it.