Abstract
Generally the distribution of re-classification is variant on each sorting, about a same class of works which were automatically measured and classified on their characteristic values. The variation can be forecasted on amount by statistical consideration. For instance, in case of sorting the size of shape, measured and classified work is supposed to have a Gaussian distribution N (μ0, σ02), concluded systematic error of measuring terms, accidental error of machine, and error of shape. Now, pay attention to the class which is classified in the first sorting, these works are supposed to have a uniform distribution with class range. And this class range c is divided into k and thin pillars are set at central location on mean value, and each pillar has distribution of normal population N (μ0, σ02) with variance σ02. The variation of distribution on repetition sorting is considered to be the area of normal distribution out of the limit of both side of uniform distribution. The propriety of this conception is ascertained by experiences, and recognized of excellent uniformity. From this result, it is elucidated that the probability is easily found out by calculation, of re-distribution of re-sorting with autosortor.
