iSixSigma

Capability Study

Six Sigma – iSixSigma Forums Old Forums General Capability Study

Viewing 7 posts - 1 through 7 (of 7 total)
  • Author
    Posts
  • #46640

    Rick L.
    Member

    I am involved in an ongoing discourse regarding the maximum required number of parts required to do a capability study on a grouping of parts. Or better put a number by which the relative result is no longer effected by having additional measurements. We have a mold that produces 8 of the same part at a time. Each part is produced from a different “cavity”. We are attempting “certify” a Cpk on each individual cavity as well as across the lot of cavities. There is a discussion on whether a quantity of 32 parts per cavity in the study is as effective as 50 parts per cavity. While one would think that this information and reference would be easy to locate, it has been harder than we have anticipated. Someone has pointed to the Central limit theorem but I am not sure this applies. Any thoughts on this or any other formula or theorum that would apply a number for this? Thank you.

    0
    #154464

    plastic accrington
    Participant

    Traditionally, a preliminary process capability study ( i.e. to qualify a new process) would be conducted by collecting a minimum of 20 subgroups of 3 – 5 pieces from the process, under normal operating conditions. So, to achieve what you want to do would require a minimum of 60 parts per cavity collected in sub – groups.
    These data are then plotted on a X bar – R chart to determine whether the process is in statistical control. Given the latter condition, then you can check for normality, and calculate preliminary Cp and Cpk (in the old days we called this Pp and Ppk, but those terms have been hijacked  to avoid the tricky business of getting a process stable).
    However, this will only give an indication of what the true process capability might be. Unless your study incorporates process changes such as mould cleaning, setting, materials changes, cooling rate/ temp changes, power surges/dips, etc.,  (unlikely in the short term), your indices will be  meaningless. If you want a true measure of your process capability, you will need to set up process controls to study long term variation. The Central Limit Theorem, DPMO’s, 3.4 shift, t – tests, multicoloured belts, etc., won’t help you.
    In my injection moulding days (13 years ago), we used DOE (mainly Taguchi) to optimise process settings, then used control charts to monitor and maintain process settings. We could only measure the finished product dimensions with any accuracy several hours after moulding, so relied on plug gauges for in – process inspection, and used a CMM machine to measure accurate dimensions for process capability studies (i.e. form and fit), which we carried out at monthly intervals.
    Get a copy of Davis Bothe’s book; it might give you some ideas.

    0
    #154736

    Jonathon Andell
    Participant

    I have dealt with multi-caity molds and capability studies before. There’s never a single right answer, because it has to do with finding the critical sources of variation – which differ from one kind of part, material, press, tool, etc., to another.
    I find that a good starting place is a multi-vari test. It allows you to explore for sources of variation before you try to mess with capability.
    If you are not familiar with this kind of test, contact me at [email protected].

    0
    #154737

    CT
    Participant

    Agree Jonathon. I too have struggled with multi cavity molds, diecast. Past experience has told me to start with a gage R&R on 32 pcs from each cavity. This is a very good way to start identifing variation in the measuring process. If your Gage R&R is good then 32 pc samples for your Capability Study is OK. I assume you have already did this work. My next question is what is your CPk goal? Does it change if you add more parts to the study? A test run of one or two cavities should be able to prove your theory about number of parts to sample

    0
    #154739

    Jonathon Andell
    Participant

    Mea culpa for overlooking gauge R&R as a precursor. Thanks for catching it!
    I like the multi-vari study because it does two things for me:
    – It gives insights into what kind of rational subgroup scheme will provide relevant data for the specific study, and
    – Identifying primary sources of variation sometimes (not always) enables me to make quick improvements without a full-blown DMAIC project – we just never know when such a fortuitous outcome will or won’t occur

    0
    #154744

    Neo
    Participant

    Rick,
    You’ve gotten some good real-world advice but I’m not sure they answered your question.  All you need to calculate a Cpk is a mean and standard deviation of your process (plus your spec limits).  I also understood from your question that your choices were 50 parts or 32 parts.  Since standard deviation goes as the inverse square root of the number of measurements, 50 measurements will give you 1.25 times more resolution than 32 measurements will.  For example, if your true standard deviation is 1.0, and you sample 32 units multiple times, your average standard deviation will be 1.0 +/- 0.13.  If you increase your sample size to 50 units, the average standard deviation will be 1.0 +/- 0.10.
    You need to know what level of resolution you need, and then decide if the extra resolution you get from the extra 18 samples improves your confidence level significantly.
    neo

    0
    #154809

    Jimmy Douglas
    Participant

    The best source for questions regarding process of machine capability studies, is the book “Machine/Process Capability Study” by Mario Perez-Wilson.  You can buy it at http://www.mpcps.com, or order it through any library.  It is the best source for all you questions.  Otherwise send him an email.  He always responses.  In this forum, you will probably get an erroneous answer…

    0
Viewing 7 posts - 1 through 7 (of 7 total)

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