Abstract

Alcoa Warrick Operations had a process that was initially in control but not capable of meeting new customer requirements. Management created a team to remedy the situation and decided to implement a design of experiment to sort out the important variables or combination variables that influence the process output.

The team selected seven independent variables at lower and higher than normal levels. They accepted six two-way interactions (3 pairs) that were confounded. 32 runs provided sufficient information about what affects earing. Other process variables on the list were going to be held constant at their nominal settings and two randomization sequences were needed. The proposed design easily met the minimum sample size requirement of 14. Average earing would be 2.0% with a standard deviation of 0.4% at this process setup. The team performed a verification study for confirmation.

The study confirmed that the earing average and variability could be influenced by these seven variables and that a combination of these seven variables did lower the average and variability. Management decided to use setup 1 and production data continued to verify that the process was still in control and capable.