THE L3 SYSTEM — QUALITY CONTROL REQUIREMENTS 



965 



However, the manner in which the three-cell method serves as a con- 

 tinuous corrective influence over the distribution of delivered product 

 can be indicated by a few diagrams, for all of which a Normal distribu- 

 tion with a' = 0.3 A is assumed. 



The running average of small segments of product delivered in pack- 

 ages of 5 is held closer to the nominal by the three-cell method than by 

 the control chart method or the batch method, even w^hen the process is 

 statistically controlled at the nominal. A comparison with the control 

 chart method is illustrated in Fig. 9. In the upper chart are shown av- 

 erages of random samples of 5 units each, plotted on a control chart Avith 

 A5 and PA limits. These are samples obtained experimentally from a 

 Normal distribution whose average, X', was at the nominal for the first 

 20 samples (Series A). For the next 20 samples (Series B) the average 

 was 0.15A above the nominal and for the last 20 samples (Series C) the 

 average was 0.1 5A below the nominal. The same units were then classified 

 and packaged by the three-cell method. In this experiment, as the units 

 of each sample were classified, as many units were packaged in 1-3-1 or 

 0-5-0 distributions as possible. Of the first 100 units (20 groups of 5), 

 95 were packaged. After 200 units (40 groups of 5) were sorted into cells, 



tu 

 < 



I N 



UJ 



3- CELL METHOD 



£^^^5^%z^;:?s;^ 



|<- GROUP t >\ \* GROUP 2 *\ \*- GROUP 3 ■*{ 



Fig. 9 — Comparison of control chart and three-cell method, averages of 

 packages of 5. 



