Bearing Defect Detection with Unsupervised Neural Networks

Algorithm 2

Sample split based on normalized sample symmetry.

	Input: normalized samples of inner diameter sample with 760 × 760 pixels.
	Output: 12 shares samples along the center of the circle with labels from 1 to 12.
(1)	Divide the inner diameter sample into 12 shares evenly along the center of the circle;
(2)	Label the 12 shares with numbers 1–12
(3)	Rotate the samples 2–12 by a certain angle to the position of sample number 1.