Process inherently out of Statistical Control

Six Sigma – iSixSigma Forums Old Forums General Process inherently out of Statistical Control

Viewing 6 posts - 1 through 6 (of 6 total)
  • Author
  • #32285


    The organisation that I work for has been using SPC on a number of process for some time now with few problems. However, recently we applied this quality improvement tool to a new process and found that the measured parameter steadily increases until it reaches then exceeds the UCL. Investigation has found that this is attributed to tool wear which can be accommodated for through machine adjustment but ultimately is unavoidable. Should we continue to use SPC given this problem (continuous plus trend) as I suspect it is preventing us identifying true special cause variation? And if not, what would be a suitable alternative? Peoples opinions would be greatly appreciated.



    You shouldn’t apply “SPC” to this process as you will obviously violate many of the rules, appearing to be “out of control”. (There is some verbage in QS Sec. 4.2 if you’re interested) You certainly could continue to control chart the process, however, using it as a tool change indicator. You will still be able to identify patterns that deviate from the positive trend signalling potential special causes. You’d probably be wise to “pick your spots” and only do this with operators that have a solid understanding of control charts.


    David King

    The origin of control charting began when statisticians got tired of having to repititive T-tests to show a difference.  They reasoned if I have an x and I plot repitive observations of x only when something happens that is so improbable that it cannot be attributed to chance alone do I need to investigate.  If you have found an assignable cause to this escursion and are seeking other assignable causes may I suggest you measure an X which is independent of tool wear to discover other possible causes to the variation. 
    Let us suppose at each observation Xi you concurrently measure the degree of tool wear Wi. Such a measure of tool wear can be made by taking a small subgroup sample at each period and comparing it to the long term mean.  the statistic (Xi-Wi) is then a statistic which is independent of tool wear and control charting this statistic may lead you to identify other possible sources of variation 



    There was a related thread once or two months ago.


    Sigma Singh

    I think your problem is that you want to make full use of your tool life and as the tool wears out with time you will have an obvious drift in the product parameter. So how to distinguisg assignable causes from this natural drift.
    The solution lies in the usage of a control chart with “Sloping control limits”. You can read this in detail in Douglas C Montegomery’s book on “Statistical Process control” under special control charts for continuous data. As the name suggests this chart will have a sloping control limits that account for drift due to sloping control limits leaving the ground clear for assignable causes.
    Sigma singh


    Lalit Upmanyu

    I have a suggestion that I believe you can give some thought on. I have read about a technique called Factor Analysis which aims to combine two factors that tend to follow same trajectory. Here since wear of the machine and product output are highly correlated they can be combined together using appropiate equation. this new factor can then be used to do all the calculations and the plotting of the data on charts.
    Would that be an appropiate way to address the issue. Please give your much needed feedback.

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

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