iSixSigma

Can we Improve process capability with DOE

Six Sigma – iSixSigma Forums Old Forums General Can we Improve process capability with DOE

Viewing 6 posts - 1 through 6 (of 6 total)
  • Author
    Posts
  • #32827

    k.bhadrayya
    Participant

      
    Are there any direct applications to improve process capability using DOE?.
     
    I wanted to use it. But I wanted to get a conformation if some one has used DOE for this purpose. I am thinking to use it based on the following logic. If some one can correct it or confirm it , it would help me.
     
    The process capability is assessed only when the process is in statistical control ie only the inherent variation exists. Say the process capability index is poor . The control charts can not improve the process. If there are some adjustment factors, they can be manipulated/ changed to improve the process. But even for this one needs to know whether they have to be reduced are increased. This knowledge can be drawn from  theoretical knowledge or experience. In any case it becomes purely hit and miss method rather than straight hitting the target and one will loose the production and the process stability is disturbed. Further neither theoretical knowledge  nor experience can disclose interactions of factors if they exist and play a havoc in the process adjustment.
     
    The best way to solve this is to conduct off line experiments with the factors affecting the process with sufficient replicates. From replicates one can fin out the variance and hence find out the standard deviation . In fact it should be possible to identify the factors that affect the process mean and standard deviation through this .The model derived from DOE calculations , can be simulated for process mean, standard deviation, and process capability for given USL (upper specicifications ) and LSL ( Lower specifications) for different levels of factors. Then it should be possible to select the satisfactory process capability along with the concerned parameters for verification and implementation .
     
      
    K.Bhadrayya
    [email protected]
     

    0
    #88145

    David Oakley
    Participant

    I have had success using DOE to improve a process. You are correct that it can be a problem if the product is expensive and you don’t want to make scrap or if there would be a hazard to upseting the process.
     
    In some cases, I’ve been able to make models that I could use in the lab to run the DOE. In other cases, I’ve had to work with Production and find some creative way to do What I needed to do.
     
    David

    0
    #88204

    Robert Butler
    Participant

    By definition you are going to upset a process when running a design.  After all the idea of a design is to check some range of process settings to determine the magnitude and direction of the impact that changing the variable has on the process.  You can set up a design so that the min and the max settings are all within the known control boundaries but you run the risk of not varying things enough in order to detect the impact of the change (if you are controlling some variable to some range it is very likely that that range was set so that, over the range of variation, nothing would happen to the process).
      In those instances where I have run a design on the production line I have found that the best approach is to set up the design and then sit down with managers and line personnel and go over the design, one experiment at a time, and ask for their thoughts about what a particular combination might produce (no real change, possible scrap, known scrap, unknown, etc.) At the end of the discussion everyone, including management, has a sense of the amount of bad product that the design is likely to make.  At that point you will either get buy in for running the design with the understanding that bad material will be produced or you won’t.  If your design isn’t accepted you may have developed enough understanding of the expected impact of your initial design so that you can go back and build a new one which takes into account the objections raised in the initial meeting.  You would then review this new design with the same group of people to see if it was acceptable.
      In your post you give the impression that you think that you need to make multiple runs of the design in order to assess both the impact on mean and on variance.  This is not the case.  You can certainly do it the way you suggest but that is very inefficient.  The method you need to check for assessing the variance without running complete design replicates is the Box-Meyer approach.  If you search this site on that keyword you will find a number of descriptions of the method as well as reference citations.
      Once you have the models built to estimate changes in mean and variance you can plug those models into any number of multivariate optimization routines to assess the possible changes in process capability.

    0
    #88236

    k.bhadrayya
    Participant

    Mr.David
    Will it be possible to send me an example with out disclosing identity of some features as probem ie factors and responses to my mail address: [email protected]
    can we know where you are working along with your mail ID
    thanks
    k.bhadrayya
     

    0
    #88242

    Michael Schlueter
    Participant

    Hi,
    You describe a good technique, which is known as Parameter Design or Quality Engineering, introduced by Dr. Taguchi at NTT in the 50’s. Its main objectives are:

    reduce variability
    adjust to-target
    at least cost
    quickly
    which results in improved process capability. The basic idea is to make the process (or better: its product) insensitive against variability of noise factors (things the engineer can never control) AND sensitive to control factors (settings the engineer has to decide upon). Specifically you will know numerically how and how much you will affect sigma and mean by changing different parameters.
    Replications today are either replaced or enhanced by introducing so called noise factors into the experiments – call it stress factors. Noise factors try to represent real bad process conditions, which will drive your output(s) off-target. Your optmized process should be least sensitive to those.
    Examles: Xerox used electrostatic forces to mimic the effect of variability in humidity, paper quality etc. for paper feeding mechanisms, provoking multiple or empty feeds. – A manufacturer fed his stamping process intentionally by materials with extremely high and low hardness materials. – In some experiments wear was accounted for by intentionally reducing certain dimensions.
    Replications do capture your current process situation. Experiments using noise factors lead to processes, which remain stable over a wide range of conditions, e.g. future ones.
    Unfortunately you will not find many case studies on Parameter Design on the internet. While I myself have hundreds available on paper, mainly in Japanese.
    When you plan your DOE, the most important step to do is this one:
    * identify the ideal final result (output)* in a measureable way.
    This requires some fresh air, because many times we are constrained by problems and limitations. Away with them. Assume, you are successful and your process runs on cpk=1000. What does it mean? How do I recognize the ideal product (output) in a sea of results?
    This has to do with “intention”. What is the processed product intended to do (under all stressy noise conditions)?
    In the examples above:
    * the feeding mechanism is intended to feed exactly 1 sheet of paper* the intended dimensions are exactly the stamped dimensions* the motor (not mentioned above) characteristics are exactly the same even after wear
    Your experiments will be most meaningful, if you switch to this kind of metrics. Defect oriented metrics are good for control purposes or for creating awareness, but most of the time poor for optimization purposes.
    I.e. if you would try to optimize on Cp or Cpk, you may obtain good results, which you most likely can not reuse for future changes. Think of the stamping example. When you optimize on stability of dimension mapping (intended dimension = stamped dimension), you may not only cover todays production of rectangular patterns, but even tomorrows market demands for circular patterns without any change – your process stamps reliably any dimension. While it may be difficult to transfer knowledge from a high cpk-optimized rectangular process to the new market demands.
    So, look a little ahead into your processes future, map it by noise factors and start. You are on a good track.
    Best regards, Michael Schlueter
     

    0
    #88246

    David Oakley
    Participant

    The plant where I was working has been closed and the work moved to another country. As a result, I do not have access to any of my project folders.
     
    As an example of improving a process using DOE, I had a cleaning process that used a chemical that we were going to have to stop using. I used test coupons to evaluate a large number of combinations of cleaning agents and times. After we had decided on the new cleaning agent, we used a DOE to prove the results from the test coupons and to finalize the concentrations and time. We used some small tanks that we had access to and made parts to scale with the tanks. There was a considerable cost to all of this but the alternative would have been to shut down the process.
     
    David

    0
Viewing 6 posts - 1 through 6 (of 6 total)

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