# Process shift/Drift identification

Six Sigma – iSixSigma › Forums › General Forums › General › Process shift/Drift identification

- This topic has 4 replies, 4 voices, and was last updated 9 years, 2 months ago by MBBinWI.

- AuthorPosts
- January 17, 2011 at 7:58 am #53697

RasheedParticipant@mamoon_pk**Include @mamoon_pk in your post and this person will**

be notified via email.On a continuous process, if the variation starts slightly and drifting the control limits in days then how can we identify and by what chart?

Like today our mean is 2.5, after two day it is 2.7 after 5 days it is 2.9. Which will also change the Upper and lower control limits. Assuming the sub group does not have any impact on the RANGE.

Further please confirm if i am not wrong, using any software for making the control charts the UCL and LCL will be marked as per current data? I think it is not the limits (constant) which it calculated on the very first day?

Please reply ASAP.

0January 17, 2011 at 1:21 pm #191150

Robert ButlerParticipant@rbutler**Include @rbutler in your post and this person will**

be notified via email.As written your question doesn’t make a lot of sense. If you set up an Xbar and R chart correctly then what you should have had before you started is a measure of things when the process was in control. You would use the data from that time frame to set up your initial control limits for both the Xbar and the R chart and then you would go forward from there.

If you are now seeing a situation where the (daily?) subgroup mean has moved from 2.5 to 2.7 and then to 2.9 this may or may not mean anything – the question is what is this doing relative to the prior limits and what is the story with the corresponding R values?

0January 17, 2011 at 1:25 pm #191151

Eric MaassParticipant@poetengineer**Include @poetengineer in your post and this person will**

be notified via email.Hello Karma,

One of the main intentions of Statistical Control Charts is to detect the difference between common cause variation and special cause varation. As part of that, the control limits and rules are intended to prevent over-reaction. In fact, the rules and control limits for many control charts have a relatively low alpha risk (the risk of over-reaction) and a corresponding beta risk (the risk of missing a drift) that many find surprisingly high.

An alternative control chart you could consider would be to use an EWMA chart (Exponentially Weighted Moving Average). EWMA uses an approach that is related to time series modeling, in which you use the historical trend to forecast the next point – and see if the process is likely to drift beyond the control limits in the near future.

Here are a few links to learn more about EWMA as a possible alternative:

http://www.itl.nist.gov/div898/software/dataplot/refman1/auxillar/ewma.htm

http://www.itl.nist.gov/div898/handbook/pmc/section3/pmc324.htm

http://en.wikipedia.org/wiki/EWMA_chart

http://www.qualityamerica.com/knowledgecente/knowctrWhen_to_Use_an_EWMA_Chart.htm

Best regards,

Eric Maass0January 17, 2011 at 1:43 pm #191154

RasheedParticipant@mamoon_pk**Include @mamoon_pk in your post and this person will**

be notified via email.Robert, it is precisely for a system which is fetching the data from the line directly and calculating the UCL and LCL by it self on a certian group to present it on a monitoring dashboard. here are two points.

1) if there is a shift/drift in a system that could be with normal distribution then how we can identify, because on daily basis system is taking the reading from line and calculating the UCL and LCL, as it is shift then all points are coming with in the control limits

2) if we will fix the UCL and LCL 9as constant) then the drift can be identified but it would not show the reason (Speacial or common) properly as UCL and LCL are constant.

I hope that the solution given by Eric Maass may work.

Also advice what is the best practice, point 1 or 2. calculation the UCL and LCL all times are take them as constant.

thanks

Mamoon ur Rasheed0January 17, 2011 at 3:20 pm #191155

MBBinWIParticipant@MBBinWI**Include @MBBinWI in your post and this person will**

be notified via email.Mamoon: Any STABLE process is going to have some instances of runs where the successive values measured all go in one direction or the other (hopefully you have learned this from studying statistics). This does not mean that the process has shifted/drifted in and of itself. However, as these values continue increasing or decreasing, the probability of each successive value retaining the pattern gets smaller and smaller. If you research the Western Electric SPC rules, you will find codified a number of different rules by which to judge whether a specific pattern should “sound the alarm” or not (I’m not going to get into the alpha/beta/confidence level discussion here – you should read up on those so that you learn them better). Even if you “sound the alarm” this does not necessarily mean that your process is out of control, merely that you should examine it (smaller and smaller probability events do actually happen). You do not need a computerized SPC program to do this (although they make it easier and automatic), merely examine the points in question and evaluate based on the criteria.

Another issue you may be observing is a short term vs longer term phenomenon. Let’s say you have a temperature effect to the process that happens to be sensitive to the temperature range between 50 – 80 Deg F, and your process is operating in a season where the temperature changes from the low of that range during the nighttime to the high end of that range during the daytime. You would see the output readings have a similar rise and fall. Now we get into the question as to whether that amount of variation in the output is acceptable or not, and how much it will cost to reduce/eliminate this causal factor. So, as you watch your process readings, do they have longer term rises/falls? If so, do they have a pattern? You may need to take data over longer periods to discern what may be causing the cycling (could be lots of raw materials, labor shifts, specific workers, environmental issues, etc.).

If, after examining data for a sufficiently long period of time, you still see a trend that is increasing/decreasing and not averaging out to zero, then you do not have a stable process and control charts are meaningless.

You might want to check out “Measuring Process Capability” by Davis Bothe – an excellent reference on the subject.0 - AuthorPosts

You must be logged in to reply to this topic.