iSixSigma

Control Charts for Processing Time

Six Sigma – iSixSigma › Forums › Old Forums › General › Control Charts for Processing Time

Viewing 9 posts - 1 through 9 (of 9 total)
  • Author
    Posts
  • August 4, 2005 at 1:42 am #40226

    Gastala
    ★ 10 Years ★
    Participant

    HiI want to apply control charts to processing time. I have several years of historical data.A histogram of the dataset has a vaguely normal shape peaking at about 5 days, but with the left tail of the distribution chopped off because the minimum possible value is 1 day. The standard deviation is about 7 days.Can anybody advise on the best approach, or this, or more generally about applying control charts to processing time or queue type situations.ThanksGlen

    0
    August 4, 2005 at 1:53 am #124229

    “Ken”
    ★ 10 Years ★
    Participant

    Glen,
    Why do you want to apply control chart to processing time?
    What’s the goal that you expect to achieve doing this thing?
    Ken

    0
    August 4, 2005 at 6:27 am #124234

    Gastala
    ★ 10 Years ★
    Participant

    Hi Ken Thanks for responding.The main objective would be to identify trends in the average processing time – particularly to warn of increases. There are also a number of ‘sites’ and it would be useful to compare performance across the sites.I’m thinking that EWMA charts may be the best bet, because they are not sensitive to the normal approximation (Montgomery) and they look promising when I apply them to the historical data.RegardsGlen

    0
    August 4, 2005 at 8:23 am #124239

    “Ken”
    ★ 10 Years ★
    Participant

    Glen,
    Your thoughts are sound in using the past performance to establish a baseline.  EWMA charts are useful for detecting subtle changes that would otherwise occur within the control limits of standard Shewhart control charts.  EWMA is a better alternative to using the older Wester Electric Trend Rules.  But I have a better suggestion.  However, I’ve used a software product called Change Point Analyzer that performs both point and trend-wise evaluation of time-series data.  It can provide you with an approximate timing for trend changes, and a statistical confidence associated with each change observation.  The best part is it’s cost.  You can get a look at CPA, and download a 30-day trial at: http://www.variation.com  (I am not associated in any way with this product, but have considerable experience using it)
    You talk about comparing performance across different sites.  So, I suspect here you are speaking more of comparing process capabilities, rather than the raw trends.
    How the information helps.
    Cheers,
    Ken

    0
    August 4, 2005 at 9:09 am #124244

    Peppe
    ★ 10 Years ★
    Participant

    Glen, I think your main problems are : the spread of data (too wide), not how to measure it and if you have a target  and how much you are far from it, as first steps. Try to do some correlation exercises and data analysis on your historical data and try to understand if it is a common spread (for all ‘sites’) or just from some. After that you can think on which is the best way to monitor it.
    Rgs, Peppe  

    0
    August 11, 2005 at 1:32 pm #124809

    JENNIFER
    ★ 10 Years ★
    Participant

    Good morning,
    Do you have the computer program Minitab to help with the control charts?  If so, you can use the I-MR, Xbar-R, or Xbar-S charts depending upon what your group size is.  Minitab will determine your outliers, SD, and mean.  Otherwise, that part will have to be done by hand (Excel).  If you’re trying to prove a hypothesis, the way the left tail is chopped off will require additional means other than ANOVA. 

    0
    August 11, 2005 at 3:00 pm #124828

    Ropp
    ★ 10 Years ★
    Participant

    You can put a control chart on just about anything, there’s legitimate arguments across the board on which chart is most effective for a particular application.  The bottom line comes down to risk – what’s your risk of failing to detect a change versus a false positive.  Contrary to common belief, Shewart cc’s are still effective when not normal, though they do lose some power.  One caveat to note before implementing SPC is to be sure that your historical data is stable and clear of special causes, which is very possible since you’re using such a large time period.
     
    Minitab is a very good software, though it’s SPC weakness is that it’s not configured to be used for ongoing monitoring.  Infinity SPC is one choice that works reasonably well, you can use many different control charts and configurations, and it will alert you real time to alarms.  Check out their website: http://www.infinityqs.com
     
    Good luck,

    0
    August 12, 2005 at 6:03 am #124900

    melvin
    ★ 10 Years ★
    Participant

    What’s your objective?  Do you have a question to answer or hypothesis to test?
    With regard to using a control chart across several years of data.  —  One reason for the blurring of the normal distribution may be process drift over the time period.  Consider segregating the data into periods based on known process changes.  Alternatively, it may be difficult to assess all changes over that much time – you may be better off to consider a shorter, more recent period as baseline.  the drifting may also mean that the limits from looking at the entirety of the data set will not serve you well in distinguishing special cause variation from common cause variation.
    Bob

    0
    August 12, 2005 at 10:49 am #124913

    Damo
    ★ 10 Years ★
    Participant

    Glen,
    I have to agree with Bobs last comment. Looking back over such a long time period isn’t going to tell you that much because you won’t be able to understand what was behind any special causes. Personally i’d take the last 20-30 data points to establish some limits and then continue to track the data from there on in, again though making sure that no special cause were within the learning phase (process running as normal). Afterall. the use of control charts are a way of predicting the future performance of the process as opposed to looking at it historically. Minitab would help you calculate the limits or you can do it by hand but i wouldn’t be that tempted to start spending money on software to carry on. Whats wrong with a pen and paper people? At least using this approach you can assign the reason behind any signals as well as document the action taken by writting on the chart. The chart will then effectively become a working diary of the process.
    Have fun!

    0
  • Author
    Posts
Viewing 9 posts - 1 through 9 (of 9 total)

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