Subscribe Bookmark
louv

Staff

Joined:

Jun 23, 2011

Why design of experiments keeps the science in science

Custom Design dialog box in JMP

Design of experiments complemented my work as a chemist.

In a recent discussion in the LinkedIn DOE group, I learned that some scientists resist the use of design of experiments because they believe that using DOE would be "taking the science out of science.” I believe this type of resistance comes from the minds of those who fear change. The unfortunate effect of this fear is that scientists end up chasing noise and experimenting less efficiently than they could by leveraging the synergy of statistics and science.

A George Box quote that I love is "Discovering the unexpected is more important than confirming the known." As a chemist at Kodak for 28 years, I was involved in hundreds of designed experiments in the chemistry space.

The patents that we obtained during those years were due to finding the exceptions to the rules rather than confirming the known. The non-intuitive findings and efficiencies gained using DOE gave us a significant competitive advantage. We always had the threat of outsourcing chemical manufacturing hanging over our heads, and we found out that rarely could anyone elsewhere compete with the quality and cost of our chemicals. (Stay tuned for some non-intuitive discoveries that I will share in a upcoming blog post celebrating the 25th anniversary of JMP.)

The science was paramount at the critical stage of discovering the synthetic route – which is the most efficient pathway to produce a compound – and chemistry. Once you have a synthetic route that accounts for all of the health, safety and environmental considerations and you have a chemical compound that is fit for use, then you need to switch your focus from feasibility to manufacturability. You gain significant efficiencies by knowing when to apply DOE to the identified process and synthetic route.

During this phase, DOE helps you understand the impact of the variation of the inputs on the outputs on the manufacturing process and the nuances of the entire measurement system. This understanding is key to being able to deliver on a commercialization timeline. And this, in turn, is critical to running a business with the lowest inventory, which you are now able to do because your processes are well-defined and predictable.

The bottom line is that DOE and science – in my case, chemistry – are complementary, not contradictory. Scientists can and should use DOE to make their work in developing and improving processes much more efficient. They’ll have more time to focus on the science they love.

9 Comments
Community Member

Peter Bartell wrote:

One other aspect that I always found valuable at the manufacturability end of the commercialization process that DOE provided was a view into the factor space window of operability. Every manufacturing engineer knew that the set points of the process were just that...set points, and variation about them was inevitable. The question was, when the process varies...How much can we tolerate and still be confident we are producing conforming product? DOE is the most efficient path to discovering this factor space window.

Community Member

Dennis Fronheiser wrote:

The "Scientific Method" has been a long recognized path to discovery and advancement. Its steps include forming a hypothests and testing that hypothesis. DOE gives us an unbiased framework for conducting such hypoyhesis and objectively assessing results. DOE keeps the SCIENCE in science.

Community Member

Dino Aimino wrote:

Right on the money as usual with respect to DOE, Lou. Can't tell how many of my DOE's have debunked the "everyone knows that x controls this or that...." myths. DOE isn't a replacement for being creative or inventive but they do provide the most efficient method for process understanding and optimization! Keep up the good work, Mr. V!

Community Member

Lou V wrote:

Spot on Peter! We coined these designs as "Tolerancing Designs". It was great to experiment in a space where the signal/noise of the test allowed for such understanding. It also was refreshing to be data driven and understand common cause variation and avoid the inefficiencies of treating the noise as a signal.

Community Member

Lou V wrote:

Dennis, Agreed...Thanks for sharing!

Community Member

Lou V wrote:

Dino, Thanks!,

Isn't it fun to recall some of our preconceived notions that were put to bed at the completion of the design. Many cases where non-intuitive results led to further understanding of the nuances of the process.

Community Member

John Hunter wrote:

Science without design of experiments is often bad science or really really slow scientific discovery. Design of experiments used properly just speeds up how quickly scientific insight can be gained and confirmed.

When interactions are significant (which is quite a lot of the time, especially in biological system - health care for example) design of experiments thinking is so useful it is hard to imagine how scientists resisting using it can succeed. I am actually amazed how successful many scientists have been using methods that are inferior.

Community Member

Lou V wrote:

Thanks John,

So good to hear from you! I prefer understanding and knowledge around sensitivity to luck:)

Community Member

Peter Bartell wrote:

Dino: Agreed...one memorable example of the dispelling of what I called 'urban manufacturing myth' happened years ago when, via DOE, after repeated experiments that demonstrated that a solution's viscosity did not have an effect on the response's critical to quality characteristics, I kept asking the domain experts, "Why do you insist on using solution viscosity as a control factor in the manufacturing process?" Finally, one of them blurted out, "Because we can!"

Article Tags