[Dividing Line Image]

Formulating Using Design of Experiments




Dom Ruggeri

August 2006:

As I have said many times, in the early Nineties I was working for a major formulating house.  The powers that be began a study on how long it took a formulator to develop a new product.  On the surface this is a noble objective however, these people came from various backgrounds none of which had anything to do with metalworking fluid formulating.  Therefore, they knew a lot about dollars and cents but had very little lab sense.

Their study revealed we all needed training in Design of Experiments.  Normally when one formulates, he changes one raw material at a time while holding the others constant.  This process continues until the new product is as close to perfect as possible.  It is also very time consuming.  Using these new DOE techniques, the formulator can evaluate any number of variables and responses statistically.  Thus cutting the formulating and testing time of new products.  The result is getting these products out into the marketplace faster.

So we began our training in these revolutionary experimental methods.  At first all seemed to be going well, and I admit a few new products were formulated in a short time.  However, as the projects got more involved and the variable lists got longer, these techniques were beginning to bog us down.  Once again, the powers that be intervened.  This time they purchased a software package.  With a few modifications, this package would not only determine which DOE technique to use but it would recommend a final formulation and give you a percent chance of success.  So just like the cast of Police Academy 2 we were back in training.

This time our mandate was to take the best computer generated formulation, tweak it so that it was marketable, and submit it to marketing so a field trial could be scheduled.  Every single product formulated using this technique failed in the field.  As we all know, good managers know how to delegate the blame, and there was enough of that to go around.  When the smoke from that fiasco cleared, the management team was gone.  So I guess something good came of that after all.

Experimental Design has been around for a very long time, the methods range from Placket Burman (12 Variables), to various designs that will measure the effects of as few as 2 variables.  But can one formulate a product based solely on DOE data?  In my experience, no.  Still, the data generated from DOE data can be invaluable in the formulation of a new product or in solving a current product problem. 

Let’s look at a situation where one could use a DOE:


The customer has an application and they are looking for the following properties:

    Corrosion protection when the coolant concentration gets down to 1%

     Falex results full load (4500 foot pounds) torque less then 800 at full load

     No microbiological problems for at lease one month

    The machines must be kept clean and free from buildup

Seems easy as a matter of fact.  You have a formulation that may do just that, however if it falls a bit short, it is perfect for a DOE.  Any good formulator could reformulate a product to meet those criteria however; a DOE will get you some direction a bit faster.

Step 1

Pick your variables.  In this case they would be:

Corrosion package - be it boric acid and amine, Corefree M-1 and amine, high molecular weight sulfonate, and the list goes on

The EP/ Lubrication package

Biocide /Fungicide package

The Surfactant Package

Step 2

Set up your maximums and minimums.  These should be wide enough apart that you will observe an effect.

Step 3

Run your DOE experiments and record your results.

Step 4

Calculate the results of your DOE and determine what raw materials are important to which response.

I will admit this is a simplistic approach, but then this is the first article on the subject matter.  I will get into more detail in subsequent articles.  As always, should you have any questions please feel free to e-mail me in care of the magazine.  Till next month,

Good Luck