Sponsored By

Design of experiments reduces cycle times

DN Staff

June 12, 1995

4 Min Read
Design of experiments reduces cycle times

As auto companies develop new vehicles, suppliers like Engelhard Corp. must develop new catalytic converters to meet customer technology objectives and satisfy changing regulatory targets. The basic technology continually adapts in a process that often begins with experiments that identify the ideal combinations, concentrations, and amounts of the catalyst components. The objective: Determine which mixture will best meets the needs of each company in the most cost-effective way.

The large number of input variables in these responses can require an enormous amount of experimentation. The efficiency of the experimentation profoundly affects the cost and time it takes to bring a catalyst to the marketplace. Engelhard has begun such a campaign to integrate design of experiments (DOE) into its operations and retained DuPont's Quality Management & Technology Center (QM&TC) to provide DOE training and consulting.

What is DOE? Design of experiments is a methodology for applying statistics to experimentation. It can be used both in the laboratory to develop new products and processes and in manufacturing for process improvements. Engelhard sought to do both.

In DOE, experimenters use statistics to develop mathematical models that predict how changes in the input variables interact to produce changes in output variables or responses. The model, which shows how these variables and responses are related, is produced through a planned sequence of experiments called a design. The model enables experimenters to predict how responses change and interact at different variable settings. It also measures and controls experimental error.

DOE, still widely underutilized as a competitive weapon, maximizes the amount and accuracy of information the experimenter receives from a given set of experimental runs. It also provides insight into how variables and responses work.

Why DOE? Developing a Low Hydrocarbons Emissions System (LHES) often calls for the optimization of five or more variables simultaneously. It would be cost prohibitive to do that using the classical approach where you change one variable and hold all the others constant. You could never sort out the interactions among variables and responses. DOE, however, works because it is an empirical, statistical approach that applies in any case where you need to know how variables interact in a system to produce responses.

At Engelhard, one of the company converts claimed he could do fewer experiments to get more results. He also said it would help him work better with customers because he could show them the experimental design, a plan with objectives and predictions about costs and outcomes, and give them confidence. Engelhard claims the methodology pays dividends to the corporation and those who use it.


Ask the Manager

Q: What is the best way to implement a design-of-experiments?

A: "What we've learned is that training seminars alone are not enough," says Larry Dight, manager, new product development at Engelhard. "Like anything that's new, DOE will be stored in some mental closet unless you have a system for reinforcement and culture change so that it becomes second nature. We're looking at new initiatives, such as the development of internal consulting support, that will spread its use."

Q: If DOE is such a powerful methodology, then why isn't it snapped up by everyone who is exposed to it?

A: There is no easy answer. "It's mind boggling to me that more people don't pick up on DOE," says Pat Burk, research group leader at Engelhard. "Some people don't feel they have the time to do the up-front planning it requires. Others feel that once you set up a designed experiment, a bunch of monkeys could carry it out-and that's no fun.

"A lot of people expect to solve problems through pure intellect. They have this scientific pattern in their mind like a cartoon: Do an experiment, create a hypothesis. Do another experiment, refine the hypothesis. Do another experiment, refine the hypothesis further. This one-variable-at-a-time approach seems sacred.

"People don't change easily. They need reinforcement and a reward structure from management. It takes some nurturing. Scientists are strong-willed people who have a robust allegiance to their paradigms, and DOE is not in most scientific paradigms."

Sign up for the Design News Daily newsletter.

You May Also Like