iSixSigma

Process Capability

Six Sigma – iSixSigma Forums Old Forums General Process Capability

Viewing 15 posts - 1 through 15 (of 15 total)
  • Author
    Posts
  • #48867

    Hemant
    Participant

    i am facing a problem in calculating process capability of a process. its a business process where we are measuring delay happening in the process. sometimes we are able to complete process well before time but sometimes huge delays happens. there are 10-15 transactions every month in process. data is coming out to be non normal. can you suggest some ideas. whether to go by DPMO method or Capability Analysis.
    Regards
    Hemant

    0
    #165848

    Adrian P. Smith
    Participant

    Is the process stable?
    If not, you should identify and eliminate root causes of instability before trying to calculate capability.
    Hope this helps.

    0
    #165855

    Evil
    Participant

    hemant,
    When calculating process capability and the data appears to be non-normal, the first order of business is to ask yourself if the data should be inherently non-normal. Sometimes, the data appears non-normal only because its so unstable OR a mix of different process conditions. In the case of cycle time data on a business process, non-normality is usually the case since it is zero-bounded (can’t complete in negative time). If you feel those huge delay points are truly the process operating “normally” (as opposed to some special operating condition), then you have some choices…
    1) transform the data using Box-Cox or Johnson transform
    2) Identify a distribution that fits and use the appropriate capability function
    3) Convert to attribute and compute defects/opps x 1e6. (as a last resort).All the above assumes you have access to minitab or a suitable alternative.e

    0
    #165889

    annon
    Participant

    APS gives good advice…using the 4S before baselining your process capability is a useful practice (spread, senter, stablity, shape).
    Although I would ask you why are you determining capability?  It appears as if you are working with a standard, not a specification, where ‘more/less is better’ is the project goal.  If this is indeed the case, why not simply work with a VSM approach to optimize CT either in isolation or relative to takt and build your charter off this approach?
    Good luck.

    0
    #165890

    Dr P K
    Participant

    Non normal processes are quite normal. The only “Normal” processes are purely theoretical.  You don’t need normal processes to use control charts … read Shewhart !
    DPMO is total nonsense.
    Capability studies can be misleading.

    0
    #165900

    annon
    Participant

    Dr PK,
    So you advise using CC such as IMR with a non-normal distribution?

    0
    #165902

    Whitehurst
    Participant

    Dr. Donald Wheeling points out in his book “Introduction to Statistical Process Control” that a distribution as badly sckewed as a right trianagular distribution has little affect on the accuracy or usefulness of making predictions based on the data you computed the statistics from.  However, you can not simply ignor non normal distributions and say that they are normal.  You have to make a decision on how critical the application is.  I would not want to ride on an airplane whose capability was based on left handed skewed poisson distribution of data representing the failure rate ball bearings failing.  Many software packages are now applying curve fitting and calculating CpK based on the number of standard deviations the mean is from the upper process limit.  You can also do curve fitting and convert the PPM to a CpK. 

    0
    #165921

    Hemant
    Participant

    Can you pls suggest some way
    My data points are
    300200700015656110000147000

    0
    #165922

    Lolita
    Participant

    Strange  data:14,56,70
    Check  again 

    0
    #165924

    Six Sigma guy
    Member

    Hi Hemant,
    Is this data points failures or defects? In that case why dont you do DPMO or DPU? Having units or oppurtunities by side will help..
    Else you can try mean time for defects if you want to do process capability for continous data. Whichever makes sense.
    Pst: above suggestion applies if you really want to do capability analysis to show to management how the process sucks currently and you want to do a project to improve the same by whatever means. Capability analysis/Zscore/Sigma score truly doesnt make any sense.
    Cheers….

    0
    #165927

    Six Sigma guy
    Member

    where every delay would be a defect or the actual time taken against a spec would give u the current process capability

    0
    #165931

    Hemant
    Participant

    Its actual data in terms of no of days delay happened in process.
    Thats where i am facing problem.
    Z value is going negative.

    0
    #165938

    Dhananjay Hegde
    Participant

    Hi
    The delay time would be a positively skewed distribution.In cases you should go with DPMO method.
    Dhananjay

    0
    #165940

    Robert Butler
    Participant

    A quick eyeballing of a plot of your data on normal probability paper suggests your equivalent 6 sigma spread is 0 – 150 (Corresponding to the .135 and 99.865 point on the normal curve).  Thus, if all of the usual caveats apply (data representative, process stable, etc.) your equivalent process capability would be
    Equivalent Cp = (USL – LSL)/(150 – 0)
    See Measuring Process Capability – Bothe – Chapter 8 for details.
     It is interesting to note that, if the data is representative, your process would seem to function very well most of the time but when it goes wrong it really goes wrong.  It would be interesting to know of 70, 56, and 56 all have something in common which is not present in the other cases.
     
     

    0
    #165952

    Neo
    Participant

    Noticed that half your non-zero delay times are in exactly 1 week increments.  Is there some reason to expect that from the nature of your process?  I believe that this could complicate the analysis of everything from normality to capability, but I’d like to hear more on that from the experts.  How do you deal with highly quantized data?

    0
Viewing 15 posts - 1 through 15 (of 15 total)

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