injection molding DOE: minitab 14
Six Sigma – iSixSigma › Forums › Old Forums › General › injection molding DOE: minitab 14
- This topic has 9 replies, 10 voices, and was last updated 15 years, 9 months ago by
KHIROD PATTANAIK.
-
AuthorPosts
-
December 3, 2004 at 2:36 pm #37741
Hello folks:
We have correctly identiifed all the factors in our process, all 11 of them. I do have a quandry however, three of the factors involve temperature changes that will take many minutes, maybe an hour to stabilize. There are hi/lo settings for these temp factors. If I randomize the experiment becomes too cumbersome for practical consideration. Is there a tactic, such as blocking, I can use to segregate the teperatures?0December 3, 2004 at 2:41 pm #111689
facemanParticipant@facemanInclude @faceman in your post and this person will
be notified via email.You can consider treating it as a random factor or a covariate.
Goos luck.0December 3, 2004 at 3:36 pm #111695
Robert ButlerParticipant@rbutlerInclude @rbutler in your post and this person will
be notified via email.Your problem is a very common one and the design of choice is a split-plot. You randomly choose a temperature setting (the split) and then randomize your runs within the split. Once you are finished you choose another temperature setting at random and repeat the procedure until all temperature settings have been run. In your case, it your description of the factors so affected suggest you will have to run a split-split plot. A split-split plot is the same idea as a split plot but it is more complex. I don’t know the capabilities of Minitab with respect to such designs.
An excellent reference for split and split-split plots is:
Analysis of Messy Data – Volume 1 – Milliken and Johnson
You should read up on these designs in order to both understand what the computer design is doing for you (assuming it can generate one) and what you need to do to properly run it an analyze the results.0December 3, 2004 at 5:55 pm #111697Use a Taguchi L12 placing the most difficult change in column 1, second most difficult in column 2 and so on.
The advice on randomization is impractical in most situations. Taguchi set up his arrays with this in mind.0December 3, 2004 at 7:36 pm #111701Stan is absolutely correct. Use the Taguchi DOE in your Minitab* 14 and use Hi/Low on your 11 factors. You can chose # of runs (please pay particular heed to Stan’s recommendation to use the first variable as the one which is hardest to change, second variable will be next hardest to change, etc.)
Once you are done, you can analyze using Taguchi’s bigger is better, nominal is better, smaller is better approach. I also learned that once you have the Taguchi DOE results in place, you can convert the design to a custom Response Surface Design.
PB
0December 9, 2004 at 8:49 am #112084Hi Thomas,
you could make 3 blocks, each one for the factor that involve temperature changes. Then in each block you can fix a temperature, so, in the experiment, you will do only 3 times this change. In this way the experiment will not be completely randomized, but you can “catch” the error come from the differents blocks.0December 9, 2004 at 10:57 am #112090Thomas, would you share what you determined to be the 11 significant variables?
Thanks
Tim0December 9, 2004 at 1:42 pm #112107
TierradentroParticipant@johnInclude @john in your post and this person will
be notified via email.thomas,
I was faced with the same situation you describe but in a different process industry. I researched the knowledged database of my statistical software supplier for technical articles on how to treat one “hard-to-change” factor. Not only did they have articles, they provided direct technical support via telephone and email in order to get a final design suitable for our needs. I have been using DOE techniques for many years, but in this case I wanted additional reassurance on how to set up my blocks, randomize, and then most importantly – how to correctly analyze the experiment. This approach worked well for me – it may be an avenue for you to pursue.0December 10, 2004 at 11:04 pm #112213
Kim NilesParticipant@KnilesInclude @Kniles in your post and this person will
be notified via email.Thomas:
Three thoughts, one is to make sure you explain what will become the assumptions you use in your final report so you can later go back if you want and analyze the effects of temperature changes. For example:
– One hour is assumed to be sufficient for temperature change stabilization.
– The split plot approach was chosen as it is assumed to sufficiently randomize temperature change stabilization error for our application.
My second thought is for you to note that the ideal injection molding response variable would be the viscosity of the injection liquid as it cools. Since you are not likely measuring that, you should know the correlation of the response to the control which may be different for each of the 11 factors and not necessarily high.
My third thought is that from experience, I have found anything temperature related to be extremely significant with injection molding experimentation (including very significant interactions). Time related variables are only very significant and pressure related variables are mildly significant. Along these lines, Ive also found results to be rather linear and obvious after youve run a couple experiments such that prediction and or extrapolation becomes more reliable than with other types of processes.
Good luck with it.
Sincerely,
KN http://www.KimNiles.com0August 9, 2006 at 6:17 am #141567
KHIROD PATTANAIKParticipant@KHIROD-PATTANAIKInclude @KHIROD-PATTANAIK in your post and this person will
be notified via email.Dear Sir,
In moulding we are collection the rejection which is visual measurement like sink mark,warpage. Which is controlled many variables and same input causes different output, in that case how we will use mini-tab.
thanks,
pattanaik0 -
AuthorPosts
The forum ‘General’ is closed to new topics and replies.