Solving the optimal process target problem using computer-generated experimental designs

Abstract

For any given manufacturing system, the initial setting of the process target is critical in preventing excessive product rejection and rework costs. Often referred to as the 'process target problem', the traditional approach relies first on the assumption of certain values for the process mean and variance, prior to identifying the optimal setting. This paper, in contrast, proposes integrating estimated response surface functions developed for the mean and variance based upon observations made on a given process, thus removing any assumption on the parameters. In addition, this paper considers non-standard experimental regions, where constraints may exist on the factor space or restrictions are implemented on the number of experimental runs conducted. In doing so, greater flexibility is obtained in finding solutions to process target problems and the scope of the research field is broadened. Non-linear programming methods are used to facilitate this approach, and numerical examples are provided to illustrate findings.[