Design of Experiments

ment in a trial-and-error manner, changing one factor at a time, the way Edison did in developing the lightbulb. A far more effective method is to apply a computer-enhanced, systematic approach to experimentation, one that considers all factors simultaneously. That approach is called design of experiments (DOE), and corporat ions across the nation are adopting it as a cost-effective way to solve serious problems afflicting their operations. DOE provides information about the interaction of factors and the way the total system works, something not obtainable through testing one factor at a time (OFAT) while holding other factors constant. Another advantage of DOE is that it shows how interconnected factors respond over a wide range of values, without requiring the testing of all possible values directly. DOE fits response data to mathematical equations. Collectively, these equations serve as models to predict what will happen for any given combinat ion of va lues . With these models, it is possible to optimize critical responses and find the best combination of va lues . DOE software is tremendously fast, enabling the testing of many factors in just days. That’s drastically shorter than the months, if not years, spent testing only a handful of factors using the traditional method. DOE takes some of the art out of experimentation, replaces it with science, and yields better results. Consider the experience of the John Deere Engine Works (Waterloo, Michigan) and its search for a way to improve paint adhesion to aluminum components while eliminating the increasingly expensive chromateconversion coating process as a pretreatment to painting. It was difficult to find a basis for improving the paint’s adhesion because John Deere’s data did not clearly indicate the cause of the paint’s limited performance, says supply management engineer Wayne Mills. John Deere first ran a screening design to identify the important variables among the 12 that its experimental team had established, which included the casting method, pretreatment process, paint type, and the concentration of adhesive ingredients. The results surprised everyone. Chromate conversion had very little effect on paint adhesion, but paint type proved a very significant factor. No one at John Deere had considered paint type as a major player. More tests narrowed the variables to three: chromate conversion, paint type, and surface treatment. A three-dimensional-cube plot generated by the DOE software clearly showed how the three variables interacted. The results again identified paint type as the major problem. To confirm the software’s results, Mills and his colleagues performed a traditional one-variable test with several data points. This final test convinced all the members of the experiment team. As a result, John Deere solved its paint adhesion problem and eliminated the chromate-conversion pretreatment process from aluminum parts—a change that has saved the company nearly $500,000 annually, according to Mills.