topic Re: Interpration of DoE results in Discussions

Interpration of DoE results

MartinY — Wed, 09 Aug 2017 06:53:02 GMT

Martin Here:

I have to brother JMP staff with my recent DOE project once more.

I am developing a cementitious adhesives, I used JMP custom design; I picked three factors:

Blue circle (white cement): 0.6898~0.7431 (proportion)

Ternal white (HAC cement): 0.2564~0.3009 (proportion)

sodium citrate (retarder): 0.0023~0.0093 ( proportion)

these three factors are entred in custom design as Mixture, added up proportionally to 1.

The responses are three adhesions.

adhesion after immersion (>=1MPa)

adhesion after 6hrs (>=0.5 MPa)

adhesion after 30 mins open time (>=0.5MPa).

The design and response results are listed in below table:

Factors	Response
Blue circle	Ternal white	Sodium citrate	Adhesion after imm (Mpa)	6 hrs adhesion (Mpa)	30 mins open time (Mpa)
0.7389	0.2546	0.0065	1.02	0.68	1.136
0.7192	0.2785	0.0023	0.85	0.692	0.121
0.6940	0.3009	0.0051	0.843	0.432	0.664
0.7431	0.2546	0.0023	0.922	0.572	0.092
0.6968	0.3009	0.0023	0.89	0.716	0.201
0.7165	0.2778	0.0058	0.917	0.574	0.923
0.7431	0.2546	0.0023	0.909	0.678	0.096
0.7165	0.2778	0.0058	0.88	0.548	0.804
0.6898	0.3009	0.0093	0.88	0.224	1.079
0.7138	0.2769	0.0093	0.911	0.278	0.839
0.7361	0.2546	0.0093	0.941	0.428	0.953
0.6898	0.3009	0.0093	0.859	0.21	1.264

The first formulation gives best results, adhesion after immersion=1.02 MPa, adhesion after 6 hrs =0.68 MPa, and adhesion after 30 mins open time 1.13 MPa.

The rest all failed adhesion after water immersion.

Could any of JMP staff briefly run this design? I have trouble to interperate the JMP regression results, since there are no significant factors or interactions in JMP output.

Big thanks!

Re: Interpration of DoE results

stephen_pearson — Wed, 09 Aug 2017 15:22:55 GMT

Has the design not answered your question? Over the ranges studied (which seem very narrow) one mixture gives suitable results. The ranges you have picked may not be wide enough for the effect size to be modelled.
Have you repeated the entire test procedure (including creating the adhesive) on the first mixture a number of times (say 3-5) to check the result is a real outcome? Was it the first mixture you made? Is the test procedure well established?

Re: Interpration of DoE results

louv — Thu, 10 Aug 2017 15:22:15 GMT

Martin,

I have attached a JMP file that I constructed from your data.

I have scripted the models to the datatable. Run those scripts to see the analysis.

You may have had difficulty with the regression models without the use of the Cox Mixture Model found under Estimates>Cox Mixture. I also included the Summary of Fit and ANOVA for your design (both under the red triangle).

Lou

Re: Interpration of DoE results

MartinY — Wed, 09 Aug 2017 23:52:34 GMT

Dear Stephen:

Thanks for taking time reviewing my question.

You are right that my ranges of three factors are very narrow. The reason for it is it is very difficult to pass adhesion after 6hrs ( >=0.5MPa) and adhesion after immersion (>=1MPa)

simultaneouely. In an industrial environment, it is imperative to get workable formulation that passes all the requirement listed in the standard, in this case, I get a workable formulation ,but because of its narrow range, I have great difficulty of interpreting the mechanism behind it.

Do I need to make runs that I know before hand it will fail certain requirements? I am a little bit too target driven in this project, may be, hehe.

The tests procedures are well established, I am in the process of validating the workable formulation, but even it works, I am also interested why it works. that is why I want to

salvage something from this DOE, and plan a future one which will shed light on the mechanism of interaction between three factors.

Thanks

Martin

Re: Interpration of DoE results

MartinY — Thu, 10 Aug 2017 00:06:12 GMT

Hi Lou:

First I really appreciate your help.

The model I derived from JMP is in conflict with common sense of cement chemistry, therefore I have great difficulty of interpreting it.

Your model is much superior to my model, it agrees with the many established facts of cement chemistry.

I will study Cox mixture later on, but I could not resist to ask you directly how do you manage to convert it to a better model.

In custom design, if the factors are entered as "Mixture", the model does not give (Bule circle-0.6898)/0.053; in my model it is simply: Blue circle.

How this can be done?

Thanks again for your time

Martin

Re: Interpration of DoE results

louv — Thu, 10 Aug 2017 00:16:35 GMT

Martin,

Formulation Mixture Designs are very unique from a modeling standpoint. I set up your design from your note using the custom design platform and the default was 12 runs. By setting up the design in JMP all of the column information is coded correctly and the model script is saved to the data table. I then simply copied and pasted your settings in the design that I generated and ran the model. The difference is that when examining the model I utilize the output that I have shared namely, Summary of Fit, ANOVA and assess signifcance of the factors from the Cox Model. The Prediction Profiler is the most visual way to assess the situation.

Lou

Re: Interpration of DoE results

MartinY — Thu, 10 Aug 2017 02:29:58 GMT

Thanks, Lou.

at which step should I use Cox mixture? see below is my interface for model generating, I press Run, what I get is different model from you.

then when I go to estimate, pick cox mixture, fill in entries for factors, I got wrong message " Cox mixture can not handle higher order than quadric", then Cox mixture button becomes grey.

Could you let me know how I got it wrong?

Big thanks!

Re: Interpration of DoE results

MartinY — Thu, 10 Aug 2017 04:29:34 GMT

I see more complex script in your model, what I mean is below picture.

Do I actually need to write script to get Cox Mixture Model? in my model the script is only a few lines.

Re: Interpration of DoE results

MartinY — Thu, 10 Aug 2017 06:46:01 GMT

Hi Lou:

I always get wrong message at Cox mixture interface. after the wrong message, Cox mixture button turn grey.

Re: Interpration of DoE results

louv — Thu, 10 Aug 2017 13:43:31 GMT

Remove three-way interaction in your model dialog and try again.

Also please check your column info for your factors. Does the meta data look like this?

Also...Did you download my JMP file? Two of your rows in the data that you shared did not add up to unity and had to be adjusted accordingly.

Finally what version of JMP are you using?

Re: Interpration of DoE results

louv — Thu, 10 Aug 2017 17:38:54 GMT

Hi Martin,

No the script looks complex because it has been recorded after I have manipulated the model through my mouse clicking activities. So although it appears more complex I did not script anything but rather saved the script to the data table after I had closed down some windows and opened some others and obtained the output that was the most meaningful to assess your experiment.

Lou

Re: Interpration of DoE results

louv — Thu, 10 Aug 2017 17:40:44 GMT

If you remove the three-way interaction term in the model the Cox Mixture Model will not give this error message.

Re: Interpration of DoE results

MartinY — Fri, 11 Aug 2017 01:12:13 GMT

It turned out that I did not load Mixture and tick Pseudocomponent.

But I could not run Cox mixture under Pseudocomponent setting, I get wrong message from JMP " Cox mixture not supported with Pseudocompent coding", I guess it has

something to do with my JMP Version which is 12.0.1.

One question about Cox mixture, do we run some risk to eliminate three ways interaction? The basic cement chemistry is the retarder, sodium citrate, regulates the speed of

reaction between the two cements, Blue circle & Ternal white. Certainly the most facinating thing is to shed some light on the interaction of three ingredients.

could I add three way interaction to the model? certainly, we can not use Cox model, is there any other way to do it?

Re: Interpration of DoE results

louv — Fri, 11 Aug 2017 11:55:22 GMT

Martin,

In version 12, uncheck the pseudo components in the column info metadata and then try to run the Cox Mixture Model. That affords the same analysis as what I have shared earlier from JMP 13.1. The addition of the three-way interaction does not seem to improve the model in the design that you shared.

Lou