COMPREHENSIVE APPROACH FOR EVALUATION AND CONSTRUCTION OF ACCEPTANCE SAMPLING PROCEDURES.

Abstract

Acceptance sampling procedures are widely used in industry as part of the total quality control activities. The acceptance procedure is usually constructed based on a set of statistical and/or economic requirements specified by the producer and/or the consumer. After the acceptance procedure is determined, the users are interested in evaluating its statistical and economic characteristics. This dissertation presents a comprehensive approach for constructing and evaluating acceptance sampling procedures. A large variety of statistical and economic characteristics is studied, from both the producer's and consumer's viewpoints. A part of the acceptance procedure is the sampling plan. Various statistical characteristics of the sampling plan are studied. The statistical evaluation of the acceptance procedure consists of analyzing these characteristics. The economic analysis includes identification of the possible actions during the acceptance procedure and evaluation of the producer's profit and the consumer's cost functions associated with each action. Guidelines for applying the statistical and economic characteristics in the evaluation process are presented. In a real situation, sampling may be subjected to inspection errors, which can affect the statistical and economic characteristics of the acceptance procedure; so all the characteristics were restudied for an error-prone sampling inspection. The statistical and economic characteristics are used to specify sets of requirements for constructing acceptance procedures. Selection of an appropriate set is based on the needs of the user, the available data, and the conditions under which the procedure is to be applied. The concluding step is to combine the construction and evaluation methods into an overall analysis cycle of "construct-evaluate-reconstruct." Computer programs are given to facilitate application of the evaluation and construction processes. This study deals explicitly with single sampling plans for attributes. The analysis is based on the Bayesian approach in which the prior distribution is a mixed binomial with a beta weight function. However, the presented approach can be applied to any type of sampling and prior distribution. The results of the study can be used by decision makers as a tool to improve the use of acceptance procedures in a large variety of scenarios.