# Design an experiment

Six Sigma – iSixSigma Forums Old Forums General Design an experiment

Viewing 16 posts - 1 through 16 (of 16 total)
• Author
Posts
• #29891

aush
Participant

To all
I have two variables and they both change continously.They both tend to rise and then stabilize after certain time period. Before the two factors stabilise there can be a fair ammount of production .
It is difficult to control these variables  to an exact value . I was interested to find out, which of the variable has a significant effect or if the two factors interact and have an significant effect .
How does the forum think I should approach this .

0
#77315

Ovidiu Contras
Participant

You can answer those questions by performing a full factorial DOE , 2 variables ,two levels (low – the lowest value for the respective variable / high -the highest value) ,that gives you 4 experiments , without replication and no center points . For all those experiments you measure your output . There is specialized software that designs and analyzes your experiment .
Good luck

0
#77321

aush
Participant

Thank you for your input. My problem is that I can not control the levels . When I design the Full factorial experiment  I will have to set the variables at the required levels, which seems to be difficult.
The only other way I find is to probably group he variables.
Will that be the right way to do it.

0
#77322

Ovidiu Contras
Participant

What if you set the “low” at the starting point (before starts increasing) and the “high” when it stabilizes ?

0
#77325

Robert Butler
Participant

If you can set your variables so that when they are at the “low setting” they drift over some range of values, all of which are lower than the lowest value of your “high setting” , you can go ahead and generate your experimental design in the usual fashion.  When it comes to analyzing the design you will have to normalize the range of low and high values.  The resulting design matrix will not be a field of -1’s and 1’s but it will be an array with values between -1 and 1.
Such a design will not be perfrectly orthogonal but you will have enough separation of effects to enable you to make statements concerning the effects of your main variables.  As for being able to identify interaction effects, the answer will depend on the scatter in your low and high settings.
To simulate this, set up a standard design.  Choose a low and a high value and then pick a range around these values and substitute random values from the low and the high values into your design.  Re-normalize the design and then check the matrix using some of the diagnostic tools available.  In particular, look at the aliasing structure.  If you have access to something like SAS, or you know someone who can run it for you, have them put the matrix through Proc Reg and run it with the “vif” and “collin” options.

0
#77328

Gabriel
Participant

The problem is that he can not make the -/+ and +/- combiations. At the beginning he has -/- and when it stabilize he has +/+. But if one factor stabilize faster than the other he may have +/- or -/+ (only one of them) between the start and the stabilization.

0
#77329

Robert Butler
Participant

I guess we need clarification from Aush.  The way I read his/her second post it is possible to set the two independently it’s just that they drift over time.  If this is the case then it is possible to set up the (+,+), (-,+) etc. combinations and run the experiment.  On the other hand, if they are linked so that only the (+,+) and
(-,-) combinations can be run then, of course, it is not possible to run a design with these two factors since they cannot be varied independently of one another.

0
#77334

aush
Participant

Robert
Thanks for your reply. Infact I am not sure  if I will be able to choose the range effectively giving us -/- of +/+ conditions.
I may have to try. As you have mentioned  there is a possibility that one factor reaches + condition before than the other or the vica versa. So I do not know if such an experiment is feasible ,since I will have to decide before hand ,the time frame for the trials.
Under such condition I was looking for a solution where it is possible to run a trial with maybe certain assumptions and then based on the results  justify the significance.

0
#77335

aush
Participant

Thanks Robert I think this may work

0
#77337

FGM
Participant

Have you tried a Regression analysis, including different settings of your variables and their effect on your output variable??
Of course, this must be based on past data.
FGM

0
#77338

Gabriel
Participant

Ok, you can not control those variables. Can you measure them? Someone proposed a regression analysis. With this you can find a sort of response surface in the range where those variables move. You can not perform an experiment to see what would happen if…, but it will tell you if there is a significant correlation between those Xs and your Y (in this range only). You don’t need to be able to control the variables to do this, but you need to be able to measure them (the two Xs and the Y at the same time).

0
#77344

Helluvabigun
Participant

Why would you bother.   If you cannot control the inputs at any level, why is it important what the outcome really is.  It will be  what it will be.
I would’nt waste anymore time about where the inputs might be significant toward the output until I had some control.
Sorry to be negative but you might as well be p*ssing against the wind.

0
#77345

aush
Participant

I never said they can not be controled. Yes they can be but after a fair ammount of financial investment.
BEfore taking  such a decision we will have to know if the variables are significant. Understanding how the factors interact or not and what is the significance of the variable  deciding a control system is not very feasible. Costs of control systems is dependent on what you specify.
So I think it is important to know the variables and their significance.
Regression analysis is a way out also. I will see how I can collect the data. The data itself is reliable even though at this moment not under control.
Thanks for your input once again.

0
#77354

Marc Richardson
Participant

Can you do this?
1) set both values to low (-/-), measure your output
2) allow both values drift up (+/+), measure your output
3) set the first value to the low, leave the second value at high (-/+), measure your output
4) allow the first value to drift up
5) set the second value to the low, leave the first value at high (+/-), measure your output
You said in your original post that the values tend to stabilize once they have risen. This may not be practical but I figured it was worth offering as an alternative.

Marc Richardson
Sr. Q.A. Eng.

0
#77359

Helluvabigun
Participant

Marc

I believe Aush has no controll at present on either input.  Please then explain how he would leave input 1 low and let input drift High?  and…..please explain how he would let input 1 drift high as input 2 is kept low.   ?????
Let the guy do his correlation.  If its strong, then get stuck in with regression analysis.  Its the only suggestion I can see.  DOE needs some controllable unless we carry out noise experiments. ( and even then ….some controllables are needed.)
Good luck.

0
#77723

Seregni
Participant

To All,
Remember that every statistical tool must satisfy certain statistical assumptions. DOE’s is not an exception. This states that your input variables (X’s) must shown some degree of statistical  control (common causes present only). Then you can manipulate them to run a very succesfully and predictable DOE. Otherwise the DOE results with this unstability will be confusing and misledading. The main purpose of a DOE is to undertand the behaviour and the effect of the red X’s on the response (Y)   My sugestion: Is this unstability a special cause? If yes then eliminate them. Then run your DOE.

0
Viewing 16 posts - 1 through 16 (of 16 total)

The forum ‘General’ is closed to new topics and replies.