Acceptance sampling design by computer - optimality concerns

Abstract There is a great deal of literature dealing with the use of the computer in designing acceptance sampling plans. The general approach is to use some approximation technique to generate minimum sample size plans whose OC curves will approximate the desired Producer's Risk and Consumer's Risk levels. Since only rarely will a single such approximation satisfy both (1-α) and β requirements some means, either averaging or selection, is used to select a plan. Plans so determined are acceptable but often not optimal, and planes with significantly smaller sample size may exist which are very close to optimal. This paper reports the development and use of a computer program which may be used to design single sampling plans using either the binomial or Poisson distribution. The program also finds alternate plans with smaller sample size, and gives a measure of the proximity of such alternate plan to optimality. Some rudimentary artificial intelligence techniques are employed in the search and selection of optimal plans and the near-optimal alternative plans. An extended version of the program also supports experimentation with a variety of criteria of optimality for the selection of candidate plans from those generated. The main program is written for use under MS/PC-DOS in both Turbo Pascal and Turbo C. The extended version uses only Pascal.