cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Browse apps to extend the software in the new JMP Marketplace
Choose Language Hide Translation Bar
PValueEnemy
Level II

How to design an experiment to compare multiple suppliers with different input ranges?

Hi!

Suppose I want to test a product yield from different suppliers. I want to test how temperature (TMP) and rotating speed (SPD) affect yield (YLD) in suppliers (SPL) A and B. However, supplier A had different rotating speed setup than supplier B.

 

Initially we propose a RSM DoE with input variables TMP, SPD and SPL, with SPD as percentage of their full capacity, let's say varying from 40% to 100%, but their absolute speeds are different. For example, supplier A at 40% capacity is 100rpm and supplier B at 40% is 120rmp.

 

I wonder if this is enough to

  1. Compare if suppliers A and B are different in terms of yield.
  2. Have the full overvier on how TMP and SPD impact YLD for each supplier.

If not, how could I do that?

2 REPLIES 2

Re: How to design an experiment to compare multiple suppliers with different input ranges?

It is practical to use the percentage in your design. It is an operational definition. I would include an interaction term for speed with the supplier. You can then optimize the speed for each supplier or use the supplier for comparisons. You can add a column with a formula to convert the percentage to absolute levels for each supplier.

This situation is not uncommon. Many community members have the experience to help you further with this challenge.

statman
Super User

Re: How to design an experiment to compare multiple suppliers with different input ranges?

There are multiple options, much depends on what you want to achieve (Are you going to choose one supplier over the other?, Do you want each supplier to improve their yields independently/simultaneously?  How good are your supplier relationships?  What is yield and how is it measured? Is this a batch process?  If so do you understand within batch and batch to batch variation? etc.).  There is not enough information provided to provide specific advice, but here are my initial thoughts:

1. Sampling.  Perform a component of variation study with components of measurement system, within supplier and between supplier at a minimum (you could also include within batch and between batch).  This has the advantage of assessing the measurement system, consistency within supplier and differences between suppliers.  In this case the SPD and TMP are nested within supplier. Perhaps before you start experimentation.

2. It seems unlikely there are only 2 factors affecting yield.  What about variation in raw materials, ambient conditions, duration of agitation, duration at temperature, etc.?  Perhaps run separate screening experiments with each supplier and then iterate.

3. Run nested experiments where the levels for the factors are nested within supplier.  This would not allow for assessing interactions between suppliers, but who cares?  I'm not sure you can manage an interaction between suppliers and x's other than optimizing them independently (see option 2).

"All models are wrong, some are useful" G.E.P. Box