INCREASING THE ADDED VALUE OF EXPERIMENTATION IN PROCESS MODELING

摘要

To remain competitive with World Class technology today requires continuing process and product improvement. In particular, reducing process variation along with ever shorter improvement cycles is needed. Understanding the contribution of each process variable to the overall variation in the process is also demanded. Success in such efforts can be aided by maximizing the information obtained with every experiment. This desired "added value" in experimentation can be obtained by coupling the fundamental physical understanding of the process (or product) with such analytic statistical tools as (PC based) Multiple Regression and D-optimal Experimental Design. The methodology is illustrated using published data on impact energy of PM sintered parts containing various amounts of nickel, copper, carbon, and MnS at various densities. First the contribution of the five most important variables to process variation are found by analyzing data from a traditional 19 experiment Resolution V design. The use of D-optimal design to select the next, best, experiment is then illustrated with "random" data in an additional set of 15 experiments carried out at the same time. With this one extra experiment the need for carrying out the 19 experiment Resolution V design is eliminated.