Optimization techniques

Six Sigma – iSixSigma Forums Old Forums General Optimization techniques

Viewing 17 posts - 1 through 17 (of 17 total)
• Author
Posts
• #29956

Nitin Agrawal
Participant

I want to know of different optimization techniques I can use to find out the best specs / value for my process parameters.

0
#77574

vidya Kulkarni
Member

hi Nitin,
I think DOE can be the best tool for optimisation of a process or spec value.
vidya

0
#77588

Ø6 Sigma BB Coordinator
Participant

When we know Y = f(Xs), we can optimize our process easily.
These are just some of the name of the techniques to find Y = f (Xs) and how to use it after discovery the relationship.
0. Variable Search
1. Factorial DOE
2. RSM
3. Scatter plot
4. Robust design
5. Parameter design
6. Tolerance design
7. Statistical tolerance analysis
8. Simulation
Hope this helps.
Six Sigma Black Belt Coordinator

0
#77600

Mike Carnell
Participant

Nitin,
DOE’s in and of themselves will point you in the right direction but your best treatment combination is not neccessarily an optimized solution. There are two techniques that can be used as a follow on to a DOE that will provide a systematic methodology to get to optimization. They are Evolutionary Operation (EVOP) and Steepest Ascent. They are similar in concept.
Good luck.

0
#77623

Nitin Agrawal
Participant

Dear Ms./Mr. BB Coordinator,
Thank you for your reply. Can you tell me more on the following techniques
1) RSM (Response Surface Methodology)
2) Statistical Tolerance Design
How can I use these techniques to find out the optimum condition for my process?

0
#77624

Nitin Agrawal
Participant

Dear Mike,
Thanks for introducing me to the two techniques of EVOP and Steepest Ascent. Can you send me some more details on the two so that I can develop some understanding of them on what is the basic funda of the two and how can they be used.

0
#77630

mcintosh
Participant

Don’t forget to consider the effects of process variation of your factors in the optimiztion effort.  The sensitivity of one factor vs another may drive you to different “ideal” zones.  EVOP and RSM only give extremely short term glimpses into variation. You definitely will want to do some Monte Carlo type modleing .  Remember group 6S is not just about the mean – vartiation is the key.

0
#77634

Manee
Participant

Nitin
Robust Design or Taguchi Method is the best technique for optimization of process parameters.
Manee

0
#77637

Erik L
Participant

Nitin,
Evolutionary Operation (EVOP) was a strategy of experimentation first espoused by Box and Draper.  The basic idea was to run incremental experimentation on full-scale manufacturing processes without disrupting the overall quality of the product and/or to negate bringing the process off-line to run text book applications of factorial experimentation.  The results of EVOP will be a continous improvement of the process, but will not result in breakthrough-type performance improvements.  The overall strategy is heavily dependent on process knowledge and stability of the process that is being targeted.  The basic elements include:

Series of sequential experiments
Simple factorial designs at 2**2 or 2**3
Small changes to factors
Large # of runs per factor combination
Simple graphical analysis
Steepest Ascent is an element within Response Surface Methodology (RSM).  DOE like other tools builds on a foundation of knowledge.  The maturity of DOE to a process goes from screening, to characterizing, and culminates with optimization.  RSM is the technique used to optimize a process.  At its basic level, RSM is looking to model curved relationships to find the optimum response through the quadratic model.  This is due to the fact that most functions can be approximated quite effectively by the 2nd order Taylor series expansion and further complexity is not typically needed.  Additionally, all of the x’s, for RSM, need to be continuos.  So what about this method of steepest ascent?  Well, it builds on the factorial-type DOE.  Let’s say that you’ve fractionated a design.  Picked a smaller set of variables and then followed up with a full factorial.  Once you have the Y=f(x) and S=g(x) you can look to see whick path takes you to more desirable response levels.  This is accomplished by either taking the practical approach of asking Minitab to make a cube plot and to pick off any trends of desirable improvement or you can go analytical and take the the 1st order derivative of the Y=f(x) model that you’ve developed.  This will provide you with a gradient vector (g) that will provide insight into where the next iteration of the experiment should be run.  So the cliff note version of RSM would be:

Run factorial designs
Look for significance of a quadratic factor (if missing move further up the gradient-if present add axial points.)
Hope this helps.  To do this topic justice we would need a discussion forum focused on just DOE, but this gives you some insight into the general concepts.  I’d highly recommend the following books:  Design and Analysis of Experiments, Montgomery, 5th ed. and Reponse Surface Methodology, Myers and Montgomery.
Regards,
Erik

0
#77638

Erik L
Participant

Manee,
I’d be cautious in advocating Taguchi methods as a preferred experimentation approach.  The designs, and analysis via S-N ratios, has been shown over and over to yield unnecessarily complex designs that are inefficient in ultimately yielding the Y=F(x) and S=g(x) models.  There is a lot of literature out there that reinforces the sequential building approach of factorial experimentation and design.
Regards,
Erik

0
#77644

Manee
Participant

Erik
I think you should read the book by Phadke “Quality Engineering using Robust Design”.  There is a comparison given in one chapter between DOE and Robust Design.  There are other books on that subject too
Manee

0
#77647

BB
Participant

Douglas Montgomery’s latest edition (5th?) of his Design & Analysis of Experiments has a chapter on Robust Design. His approach provides the modern (post Taguchi) approach based on Response Surface methods. I do wish that section provide a little more guidance and more examples though.

0
#77654

Erik L
Participant

Manee,
I have read Phadke’s book on the subject.  While I do recognize the loss function that Taguchi advocated and pushing for robust design, I do not believe the experiments that he ‘modified’, based on S/N ratios, to be as efficient and/or effective in deriving the optimal Y=f(x) and S=g(x) response models.
Thanks,
Erik

0
#77655

John J. Flaig
Participant

Erik,
Your analysis is absolutely correct. The Taguchi approach to optimization is fundamentally flawed. It may work on some problems, but not on others. When it works people say “see I told you it works”. But the two step optimization has serious heuristic problems. For example, Taguchi breaks the controllable factors down into control and signal. Control varibles affect the variance, signal variables affect the mean. OK suppose the only active variable you have affects both the mean and the variance, and as you reduce variance the mean is moved away from the target. Two step optimization will not work.
John Flaig, Ph.D.
Applied Technology
http://www.e-AT-USA.com

0
#77659

melvin
Participant

For more details / exploration of this area take a look at:
–  Myers, R.H. and Montgomery, D.C., Response Surface Methodology:  Process and Productr Optimization Using Designed Experiments
–  Hu, C.F.J. and Hamada, M., Experiments:  Planning, Analysis, and Parameter Design Optimization
Bob

0
#77801

Tom Dunn
Member

The best way I have found is by using a very awesome 6-sigma software package I purchased a couple years ago.  It is very easy to use and does a great job.  It actually is the best software for the entire six-sigma process I have ever seen.  It does it all from problem statement to DOE to SPC.  If you are intersted let me know and I will track down some contact info for you.  [email protected]
Best wishes to you,
Tom Dunn

0
#110320

MBB Help
Participant

If you’re still out there. . . can you tell me more about the software you mentioned.

0
Viewing 17 posts - 1 through 17 (of 17 total)

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