A proposal to apply Taguchi-Inspired Methods to the reduction of machining variance

Abstract

The Probability a Machined part will be defective increases with the variance of the machined dimensions. Even for parts within tolerance, the quality decreases with the variance. By reducing the variance of these dimensions better parts will be produced. Several factors, some of which are controllable, impact this variance, and they may also interact with each other. By choosing an appropriate level for each controllable factor we can minimize the variance. Since factors may interact with each other, a factorial experimental design is appropriate to optimize the levels of the factors; optimizing one factor at a time is not likely to yield the global optimum in the presence of interactions. The designed experiments must be conducted in stages: (1) at the preliminary stage one determines which factors and interactions are likely to be important; (2) at the exploratory stage one runs a fractional design to identify the factors that are really important and to verify/update the preliminary hypotheses about the interactions; (3) at the search stage one uses a sequence of experiments with varying levels of the factors that were identified as important to optimize their levels; finally, (4) at the verification stage one runs the process under the levels deemed to be optimal, to check if the process behaves according to the predictions. At any stage, after analyzing the current data, one may have to go back to a previous stage. This report presents a list of factors and Interactions (preliminary stage output), and a design for the exploratory stage. We also discuss how to interpret the results and conduct the search stage. The plan is designed as a generic blueprint that can be used at any machine shop.