Robustness design

[#1]

A DoE workflow calls for prescreening, screening, optimisation then a robustness design.  The robustness design would inform around a control space within a broader design space mapped out by the optimisation design.  In JMP, is there (and if so where) a procedure for running robustness designs starting from an initial data set used for optimisation?  

[#2]

Hi @kjwx109prime,

Welcome in the Community !

There may be several definition of robustness and different ways to create a robustness DoE, depending against which factors you want to test the robustness of your experimental space/optimum point : against external noise factors, against variation in process/experiment factors, or against a combination of both factors. Here is a serie of articles from Stat-Ease explaining the type of designs according to which robustness situation you are interested in :

1. Robustness against external noise factors : https://statease.com/blog/achieving-robust-processes-via-three-experiment-design-options-part-1/
2. Robustness against variation in our set points for process factors : https://statease.com/blog/achieving-robust-processes-via-three-experiment-design-options-part-2/
3. A combination of the first two types : https://statease.com/blog/achieving-robust-processes-via-three-experiment-design-options-part-3/

I did a short presentation using sequential DoE for the fine-tuning of an algorithm (Random Forest) on a fixed dataset, using screening design, then optimization design, and finally using a simple factorial screening design to assess robustness of the factors range found previously : https://www.linkedin.com/posts/victorguiller_design-of-experiments-machine-learning-activity-7122469...

Some ressources that may be helpful as well :

How to Design for Robust Production Processes | JMP

Hope this first discussion starter may help you,

[#3]

Hello @Victor_G ,

Thanks for pointing out the ambiguity in the query.  I am actually interested in the scenario attached.

The outer cube relates to response surface modelling and it describes a design space.  Operating at any point within this cube completely should meet requirements for acceptability, according to the response surface model.

The centre point describes the setpoint condition.

I am interested in a design that looks at the inner cube.  In this case, this does not relate to robustness to small variation around the existing setpoint, but is a slightly smaller version of the outer surface.  We are confident that operating at any point within the outer surface is suitable, but we are even more confident if operation is restricted to within the smaller surface.

In JMP, is it possible to followup a response surface model with a design which looks at this more restricted operating space? Alternatively, can the more restricted space be studied in more detail at the outset, as part of the original response surface study.

Alan

[#4]

I'm a bit confused.  Of course you can set any levels you want and it can be analyzed in JMP.  The analysis is greatly simplified if the design points are "balanced" (e.g., CCD, face-centered, Box-Behnken), but regression can be used at any point.  Think of the geometry of the space.  Imagine you are trying to sample that space to create a contour map of the response.  Usually a balanced sampling of the space is preferred as it is less biased.

Robust designs are typically related to understanding how well the model works over changing noise.  In these designs, noise is varied in the experiment usually as blocks, split-plots or cross-product arrays.  The objective is to find factors whose effects are consistent over changing noise and the effect of the noise is mitigated.

[#5]

Hello @kjwx109prime,

The situation you described is similar to what I have done on the example mentioned earlier :

I did a short presentation using sequential DoE for the fine-tuning of an algorithm (Random Forest) on a fixed dataset, using screening design, then optimization design, and finally using a simple factorial screening design to assess robustness of the factors range found previously : https://www.linkedin.com/posts/victorguiller_design-of-experiments-machine-learning-activity-7122469...

This sequential methodology is of course possible in JMP. You can do it in 2 steps :

1. Identify the factors and their restricted ranges that match your specifications thanks to the previous optimization response surface design (and model) done and the Design Space Profiler.
2. Once the factors and ranges are identified, you can Augment Designs (and add a block for the new set of runs to better differentiate the results from the original DoE and the new experiments) or create a new Custom design with the factors and range identified earlier, and check that the results from your second set of experiments is in accordance with your response mean and variance acceptability specifications :
    

Hope this complementary answer will help you,

[#6]

@Victor_G 's guidance is very useful, as always.

I just thought it was worth clarifying that there is no specific platform or feature in JMP for creating robustness designs (I suspect this is the case in other DOE software as well).

A robustness design is usually just a simple 2-level design with narrow factor ranges around the optimum.

So (as Victor says) you could design this with Augment, Custom Design or even using the classical design capabilities.

Personally, I would use Augment as this will keep all of the data from the different stages of the experiment in the same data table.

Phil

[#7]

Hmmm, Phil, your definition of Robust Design and mine are quite different.  I think of robustness as insensitive to noise.

Taguchi did much to popularize this concept:

Taguchi, G. and D. Clausing, (1990)  “Robust Quality”.  Harvard Business Review, January-February, pp. 65-75

There are some other good papers on robust design:

Bisgaard, Søren, Murat Kulahei, (2001), “Robust Product Design: Saving Trials with Split-Plot Confounding”, Quality Engineering, 13(3), 525-530

Hunter, Stuart (1985) “Statistical Design Applied to Product Design”, Journal of Quality Technology, Vol. 17, No. 4, October 1985

Montgomery, Douglas, C., (1999), “Experimental Design for Product and Process Design and Development”, The Statistician, 48, Part 2, Pg. 159-177

Box, G.E.P., Stephen Jones (1992), “Split-plot designs for robust product experimentation”, Journal of Applied Statistics, Vol. 19, No. 1

Robust Design has less to do with the design structure and more to do with introducing Noise into your experiments.