Quality Control

M A N U F A C T U R E R is interested in producing a controlled product--one in which the deviations about the average level of quality are no larger than can be accounted for as a result of chance. The present paper gives simple detailed methods for determining from inspection data whether or not a product is being controlled in the sense of indicating the presence of assignable causes of variation. Naturally the inspection data constitutes a sample of the effects of the manufacturing causes and hence the interpretation of these data in terms of what may be expected in the future is a statistical problem. A controlled product is defined as one for which the frequency of deviations from the expected quality can be estimated by probability theory. To make such estimates, however, it is necessary to characterize or specify the distribution of quality which the manufacturer wishes to maintain. These specifications of the desired quality must be arrived at by methods customarily used in setting engineering standards, but when once they have been established the statistical methods amplified in this paper make possible the most economical control of this quality. The limits within which quality may be controlled with a given amount of inspection depend upon the standards adopted for the quality to be maintained. This paper interprets quality specifications in terms of five different types of constant systems of manufacturing causes. The five types chosen are sufficient to cover the entire range and it is believed t h a t only five types are necessary because sampling theory indicates that little practical advantage would be derived by endeavoring to subdivide one or more of these.