iSixSigma

SPC for Non normal data

Six Sigma – iSixSigma Forums Old Forums General SPC for Non normal data

Viewing 9 posts - 1 through 9 (of 9 total)
  • Author
    Posts
  • #42309

    AN
    Participant

    Greetings!
    I am working on a project where i need to use robust SPC charts to monitor the process and i have a few questions.
    The presumption of SPC is that it is works well on (a)normally distributed data and
    (b)the data points are independent.
    I have non normal data and am not sure whether my data points are auto correlated.
    What kind of SPC charts do i use for Non Normal data?
    How do i eliminate the possibility of Auto correlation? If my data is auto correlated, how do i proceed to do SPC on them?These answers will be very valuable to me.
    Thanks in advance,
    AN

    0
    #133504

    Ken Feldman
    Participant

    If you search this site for threads regarding normality of data as a requirement for using control charts, you might find that your assumption is in error.  Control charts are designed to be rather robust to normality.  If there is significant autocorrelation, you might try a time series analysis.

    0
    #133508

    Heebeegeebee BB
    Participant
    #133525

    AN
    Participant

    Darth & Heebeegeebee BB,Thanks for your inputs.
    Is it better to convert the data to normal to perform SPC? Are there certain SPC charts suited for Non normal data (apart from X bar R), like there are specific Non Parametric tools for statistical analysis of Non normal data?Also can anyone clearly define the difference between QA and QC for a process? I find a lot of definition for QA/QC on projects and IT but nothing valuable related to a process.Your feedback is much appreciated,
    Thanks,
    AN

    0
    #133528

    Ken Feldman
    Participant

    NO!  Do not transform the data for use in a control chart.  The control limits will now be in units of the transformed values and be meaningless to anybody reading it.  To repeat, control charts are very robust to normality so stop worrying about it.  The only one that might be somewhat challenged with severe non-normal data is the I/MR chart.  The underlying distributions for the R chart and all attribute charts are not necessarily normal anyway.

    0
    #133529

    AN
    Participant

    Thanks Darth,That clarifies my point.AN

    0
    #133901

    tracker
    Member

    There is a great book “Normality and the Process Behavior Chart” by Dr Donald J. Wheeler that offers guidance for dealing with data and normality.  With SPC charts it is not necessary to check for normailty.  Good Luck! 

    0
    #133902

    Jerod
    Participant

    * NonNormal Data:
    Weibull will make the best-fit curve for NonNormal
    Non Normal Data use Weibull. (Capability Chart)
    Non Normal Data you cannot center
    Normal Data you can center
     
    Box-Cox Method =NonConforming / NonNormal = picks and runs best equation for your data. The Box-Cox transformation can be used for converting the data to a normal distribution, which then allows the process capability to be easily determined.
    *Normal data Cap Studies gives you Cpk & Ppk.
    *NonNormal data Cap Studies gives only Ppk.

    0
    #134104

    R.M.Parkhi
    Participant

    AN
    Best way will be to reduce variability by a DOE & then go for SPC.You may kindly refer to the book-‘ World Class Quality’ by Mr. Keki R.Bhote ( published by American Management Association ).Pl. study carefully the chapter on Multi- Vari analysis.
    Pl. contact,in case of a doubt.
    Regards,
    R.MParkhi

    0
Viewing 9 posts - 1 through 9 (of 9 total)

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