iSixSigma

Process Capability Preferences

Six Sigma – iSixSigma Forums Old Forums General Process Capability Preferences

This topic contains 24 replies, has 6 voices, and was last updated by  “Ken” 14 years, 3 months ago.

Viewing 25 posts - 1 through 25 (of 25 total)
  • Author
    Posts
  • #40174

    annon
    Participant

    Is there a benefit to using one process capability metric over another (assume you have a choice)? Why?
    Thank you.

    0
    #123854

    “Ken”
    Participant

    annon,You ask a lot of basic questions that can be answered with a little work on your part. Try that first, then come back to the forum.Ken

    0
    #123863

    annon
    Participant

    Will do.  Thanks for the lesson in protocol.

    0
    #123868

    Gabriel
    Participant

    Yes. Cp and Ppk together make a good set.
    Why? Because Stan sais so.

    0
    #123873

    annon
    Participant

    Thank you.

    0
    #123874

    “Ken”
    Participant

    G,So, it okay to mix short-term and long-term variation in capability measures? Did anyone give a reason to you other than that’s how it should be done?If so, what’s the reason?KenQuestion on the side: If a process is not stable, then what claims can be made using any of the capability indices?

    0
    #123881

    “Ken”
    Participant

    annon,Remember this: You get from this forum as much value as you paid for. That is not to say everything you read here is useless, because there are some very reasonable and potentially useful responses. The real concern you should have is how to tell the good stuff from the crap. What do you use to filter out the crap, and get the gems. It would be good for you to do a bit of verification before you accept anything you read. This is especially true when receiving information from one source who uses another source as reliable and credible. Ask yourself these questions whenever you read anything online in this forum. Who is the person giving these answers? What is their background and training? Have they published any peer reviewed work that has met with acceptance? Have they obtained enough experience to truly acquire the knowledge and ideas they provide? How do you know? You don’t know any of us, and just that alone should compel you to question anything you read on this forum. If the answers mean the difference between success or failure on your job, buyer beware! Again, do a little homework first before using anything you read here…Just some friendly advice from a guy who has been burned many times by claimed experts.Ken

    0
    #123883

    Mr IAM
    Participant

    Thats the best advice ever posted on the forum!

    0
    #123886

    Gabriel
    Participant

    Cp tells you what your process could be if it was fully centered (it will ever be) and fully stable (it will never be either).
    Ppk shows your actual process performance, with its actual unstability level and its actuall offset.
    Cp is the upper limit for Ppk as you work to center the process and removing the special causes of variation. If you want an improvement beyond Cp, look for a new peocess.
    And to the other issue, perfect stability is neither a realistic expectation nor a desirable one (go AIAG’s SPC handbook). So where is the limit of “acceptable level of stability” before you can make a claim? And who said something about a claim anyway?

    0
    #123894

    Mike Carnell
    Participant

    Ken,
    I don’t see what is being mixed. They are separate calculations. If a person uses only short term in general they will probably overstate the process capability.
    Long term is what a customer should care about because it is the number they will be dealing with over time. Once you have long term stuff then the short term stuff doesn’t mean much unless you are doing the 1.5 shift hustle.
    Stability as I said before is a figment of someones imagination. It guarantees you nothing about what happens in the next nano second and in the Long term capability an “unstable” process will show a larger std deviation. Beyond the size of the std deviaion what other information does stability create?
    In terms of the actual calculation of Cp,Cpk,Pp, and Ppk what value do they add. They are simply groups of 3 standard deviations. Does a Cpk of 1.33 tell me any more than if the number was just 4 std deviations?
    It is one of those windmills that is stupid but not worth the fight.
    Just my opinion.
    Regards

    0
    #123895

    Mr IAM
    Participant

    Mike, 
    I agree on the capability indexes – I think your thoughts are good there.
    I’m curious about the stability comments though.  Would you agree that a “stable” (in statistical control) process is more likely to be predictiable in the near future then one that is not? 
    I agree that stability is not a guarante of future performance but, I do think that “The best predictor of future performance is the most recent past”.
    But, I have seen many improvement projects fail by people spending so much time trying to achieve “statistical control”, that they get frustrated because they can’t do it, and quit improvement all together.  That, I think is a big mistake.  To wait to improve, until the process is “stable” – probably not value-added in my mind.
    I’d appreciate your thoughts, as well as Ken’s.  Thanks!
     
     

    0
    #123897

    CT
    Participant

    Is it not true that long term process capability can give a false since security about a products process since a part or a group of parts are less likely to affect the Capability, and Standard Deviation? Only when long term issues are noticed does process slowly shows sign of instability. While using short term capability, the process centering can be more effectively monitored and changes become apparent sooner. This is especially effective in Kanban operations or in JIT operations.
    CT

    0
    #123901

    Mike Carnell
    Participant

    Mr. IAM,
    I agree with Statistical Control the probability of getting something that is similar to what you have recently received is better. In general it is a false sense of security. It is a nice to know number – does it change my behavior or my responsibility to assure that I am deliviering a good product? No. If it doesn’t change the way I need to run then the value is marginal (to me as a customer).
    We are definately in agreement on the improvement versus stability issue.
    The use of Cp & Cpk never made any sense from the first time I saw it. When they created this fight by introducing Pp & Ppk it made an illogical situation worse. Like I said earlier the fight always seemed like a losing battle so you report it that way and all the status quo folks feel good.
    Just my opinion.
    Regards

    0
    #123903

    annon
    Participant

    Thank you Ken.  Duly Noted.

    0
    #123905

    “Ken”
    Participant

    MR IAM,I very much enjoy your lucid and honest answers that give me a sense of discussion rather than a force feed. Thanks for the request to participate!I don’t have quite as strong an opinion as Mike’s. I’ve seen the process control tools used well, and used poorly. Like anything else when their used poorly, it’s a disaster. I once fell prey to the stability game. Years ago while working for Alcoa I was required to develop and prove my processes were both stable and capable. I remember running 20 capability runs on a single process before giving up. I also remember learning later the software used to perform the stability tests used all of the Western Electric trend rules to do it. I computed the Type I error rate for all the rules to be in the vacinity of 45%, ouch! What a waste! So, I’ve learned what I can and cannot do with these tools. It’s taken a long time to get there.Is achieving stability of value? Absolutely! Is it always achievable? Not on your life! How can you tell when to quit? Difficult to answer, but most likely depends on experience. In past work with Hughes Research Center I developed thin films deposition processes on Germanium, yuck, and achieved control within +/-100 Angstroms. Could not have done it without the proper use of process control tools. Was the deposition process stable? You bet. We could have never built the devices if it wasn’t. Today, my approach is very different with respect to achieving process stability. To me the measure of stability is a launch point for the direction I take to achieve process improvement. Any unstable process has signals. It speaks to me, and tells me where to look for improvement. I next take the data and subject it to a bootstrap CuSum analysis which provides me with trend changes signaling change points in the process to investigate for dependency among the inputs. What if the process is stable? Ah, its not talking to me. So, I subject the process to a series of runs designed to make it talk to me. Or, I partition the data to extract any patterns that may exist deep within its structure. The achievement of stability is no longer a primary goal for me. I use process stability as one of the communication vehicles for achieving breakthrough improvement.But, I do hold to the formative work of Shewhart, Deming, Wheeler, and others. An unstable process is an unpredictable process. Using metrics that are derived from an unpredictable process is nonsense, and a waste of time. If a process is not stable, one or more of its inputs is unpredictable. That’s a foundation of process development and control for over 70 years.Enjoy the discussion. Keep it up.KenKen

    0
    #123909

    Mike Carnell
    Participant

    CT,
    We obviously calculate Std Deviation differently. When I calculate it and there are outliers or the spread on the data is wider my std deviation number gets bigger. I have noticed this is perfectly correlated. I also noticed a perfect correlation between my std deviation getting larger and my capabilty going down.
    Just my opinon.
    Good luck

    0
    #123915

    “Ken”
    Participant

    Mike,
    The mixture I spoke of referred to the use of Cp with Ppk.  One estimate uses SDshort-term, and the other uses the SDoverall-which some people like to claim as long-term.  A potential capability estimate relates to process potential, and the actual capability estimate relates to process location. 
    I agree in the final outcome, none of these measures matter unless they are used for something, i.e., provide some value as you put it.  So, first I want to focus on process improvement.  Suppose, I’ve already done all of the requsite customer work, and know where to improve.  Now, how will I measure process performance?  I have three possibles to choose from:  Sigma Quality Level, Cp/Cpk, Pp/Ppk for continuous variables measures.  Which do I use to measure improvement?  If I use SQL, I do not have any discrimination between location and scale(variation).  SQL is a composite quality measure for management.  If I use Pp/Ppk, or any one of the two, and the process is unstable the measure will be contaminated with random changes in between group variation, unstable shifts in the mean over time.  If I use Cp/Cpk, then the measure will provide a more consistent and reliable estimate of process performance, because these estimates will not be as sensitive to random changes in between group variation over longer periods of time.
    Where’s the value in using any of these measures?  First, to improve you need to measure something.  Certainly you agree with that statement.  To report an improvement you need to be able to compare the past-state performance to the improved-state performance.  So, I use Cp/Cpk for an internal measure of improvement.  The customer doesn’t care how I measure my process to improve–they just want better stuff from me.  But, where’s the value in these measures?  Well, Cp provides me an assessment of process variability in relation to customer requirements.  Cpk provides me an assessment of process location in relation to customer requirements.  Yes, the lower standard deviation inaccurately estimates the defect level.  But, that’s not important to me.  Why?  First, all estimates are prone to error.  So Cp/Cpk and for that matter Pp/Ppk will all over or under estimate the actual defect rate going to the customer, but that’s not the purpose for my using them in the first place…  I’m not interested in the single-point measures of any of these process performance estimates.  I’m interested in the differences between past and proposed improve-state.  That difference, taking into account the error interval, provides me knowledge that I’m moving in the right direction towards improvement, and providing customer value. 
    I use the capability estimates as guides to drive my improvement efforts.  If I have a low process Cp, and a stable process, then I know I need to work on reducing process variability first.  Nothing else takes presidence.  I can’t adjust the process to achieve greater customer value.  However, if the Cp (potential capability) is > 2, but Cpk is below 1.4, then I know I have a process targeting and control issue I need to address.  If the process is unstable, then all bets are off with improvement via targeting.  Instead, I should focus on variation reduction via the signals associated with special causes.  I’ve already adressed the methods used to achieve improvement of unstable processes.  Shigeo Shingo gave us a powerful method called Poke-Yoke (Mistake Proofing)–and there are many others.
    I use these metrics to both guide me in process improvement, and tracking the improvement effort.  Given my methods are sound, and I achieve the desire improvements the value-add to the customers and the business are self evident. 
    Mike, this is the way I’ve done business for 25 years.  None of it has failed me yet.  If a process is stable, the difference between Process Performance and Capability metrics are bounded within the confidence intervals of any of these estimates.  Meaning with a stable process, you can use any estimate you want.  This stuff has worked effectively for 70 years.  There are many articles, books, and other writs to support this statement.  It’s worked for me and many of my colleagues, subordinates, and drinking buddies for over 20 years.  It’s hard for me to walk away from something that works so well.  Sorry!
    BTW, typing is no prob for me.  Took a typing class in high school as a way of getting dates.  It worked!  I got a couple of nice dates, and a bonus.  I learned how to touch type.  Most of the time what I think goes to my fingers before I can stop it.  That’s has been a curse at times in the past.  Have a great day tomorrow!
    Cheers,
    Ken

    0
    #123916

    “Ken”
    Participant

    IAM,
    Appreciate the positive feedback!  I like you unbiased, thoughful approach to communication.  Don’t leave the forum.  We need more folks like you!
    Ken

    0
    #123917

    “Ken”
    Participant

    Gabriel,
    I take no issue with any of your comments.  Process capability/performance metrics should be used for process improvement.  It doesn’t much matter which ones you use, because with either SDshort-term or SDoverall the confidence intervals typically overlap.  If they don’t, then the process is chaotic, and quibbling over which process metrics to use is a waste of time!
    I recognize AIAG has it’s set of quality requirements via QS 9000.  I also recognize that any automotive supplier has to understand and use them.  This said, US auto quality ranks fairly low in the world.  I’ve lived in many parts of the world, and driven many foreign vehicles, and this is an under-statement.  So, I’m not certain that AIAG has the line on providing requirements that necessarily drive quality.  Instead, I think AIAG is about control and profit at the expense of quality.  Sorry, for being so blunt, it just slipped out.  Don’t ask me what I think of ISO Standards in relation to quality… 
    Ken

    0
    #123949

    Mr IAM
    Participant

    Mike,
    Thanks for taking time to reply – glad to read there is some feeling of predicability to control charts, granted there are no “absolutes” in real life.  Good discussion!  Thanks again –

    0
    #123951

    Mr IAM
    Participant

    Ken, Thanks for the response.  I agree, and I like your approach.  Have you used Multi-Vari analysis much for identifying likely causes of variation in your processes? Sounds like you have used that type of analysis.  I also like to look at the process data overtime, I think it does provide some value, every time.  I also agree with Mike however, that sometimes we go over-board on “stability” analysis meanwhile, the process continues to suffer and no changes are being made.
    I ran a wave solder operation for a few years where we were charting DPU for solder shorts… could never get the process stable but, learned some good things…. like; there was no significant differences between part numbers, between operators, etc… so, for that it is valuable.
    Take care for now, thks –

    0
    #123957

    “Ken”
    Participant

    Mr IAM,
    Agree with you perspective completely, and I think you caught my drift on the process stability concern.  When I ran into a snag there I didn’t throw away the methodology.  Instead, I tried to understand where it failed me.  The incorporation of stability analysis is fundamental to process understanding.  Discarding it due to bad experiences, or heresay of bad experiences is akin to throwing out the baby with the bath water.
    Yes, I used Multi-Vari analysis in my past work–both in passive and active data collection modes.  It’s very effective in identifying key variation contributors via variance components analysis and alike. 
    Great discussion!
    Cheers,
    Ken

    0
    #123964

    Mike Carnell
    Participant

    Ken,
    I am not trying to get you to turn your back on anything. It is a discussion group and we have different opinions.
    I also have no issue with a control chart other than for the most part they are rarely implemented with any real thought behind them. They also bring a maintenance factor with them that becomes an issue. The reaction to them tends to be the real down fall.
    I am getting ready to head back out of the country so my participation will become a function of schedule and right now I am behind.
    Thanks for the discussion.
    Regards

    0
    #123965

    Mike Carnell
    Participant

    Mr. IAM,
    I am glad we agree on the lack of activity related to stability. My issue with stability doesn’t come from Control Charts. Check out some of the previous posts on the calculation. It takes a years worth of data. If I have a process that is tearing me up I am not waiting a year to begin fixing it, if it isn’t stable waiting for another Quarter of data and I certainly want something more sensitive than that to manitor a process.
    Control charts have their place. Unfortunately they also carry baggage.
    I gotta go. Time to head for the Southern Hemisphere for a month or so.
    Regards

    0
    #123980

    “Ken”
    Participant

    Mike,
    Not going to give up the ship just yet, so don’t worry.  You’re right, it’s only a discussion.  I agree companies rarely implement process control properly.  I’m curious about the statement, ” the reaction to them [control charts] tends to be a real down fall.”  Again, I think this statement serves to support the first from your observations, but not sure.
    My experience in working with SPC for 25 years is that if the management and work level teams understand how to use SPC properly there is a much better chance of successful use.  Perhaps there is an underlying cause for why you’ve had a much different experience with process control methods than I, i.e., industries you’ve supported, etc., but this is just speculation on my part.  I am left with at least one nagging question that I must ask.
    If process control methods using control charts are not your favorite, what guidance do you provide others to establish process control in the Control phase of a DMAIC project?
    Have a safe trip back to the US.
    Regards,
    Ken

    0
Viewing 25 posts - 1 through 25 (of 25 total)

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