iSixSigma

I-MR chart for determining if process is in control

Six Sigma – iSixSigma Forums Old Forums General I-MR chart for determining if process is in control

Viewing 8 posts - 1 through 8 (of 8 total)
  • Author
    Posts
  • #43470

    davchr
    Participant

    I have several related questions on the use of I-MR charts;
    1. Is is valid to plot data from 32 consecutive samples on an I-MR chart and make a determination as to whether the process is in control or not? Since the samples are consecutive it seems like the sigma will be very small and the not representative of process variations over time.2. Does it make sense to chart several dependent measurements on an I-MR chart? If so, why? This is an injection molding operation. I think that one X and one Y dimension should be sufficient. If the overall dimensions are monitored, the smaller interior features will change with the overall dimensions.thanks
    davchr

    0
    #137918

    Elbrin
    Participant

    davchr,
    It sounds like you are asking this in the context of monitoring the performance of a process, rather than seeking improvements.  You are really the best person to answer that question.  You know your process, logical sub-grouping, and random sampling techniques are key to intellegent data analysis.  If 32 consecutive samples is not representative to of the of the possible population then change your sampling strategy.  I am somewhat familiar with injection molding, and in my opinion there are factors for variation that would not be apperant in 32 consecutive samples.

    0
    #137919

    davchr
    Participant

    The reason for my question is that our customer is using the data from 32 consecutive samples of our parts and determining that the molding process is out of control.  Based on these 32 samples, the sigma is quite small so it is easy to have one or more samples out of the limits.  Because of the small sigma, I think that some of the error may be measurement errors and that if we measured one part 32 times that it would be out of control as well.
    This is not my area of expertise.  If this is a gray area, I will investigate further so that I can present a reasonable argument against this criteria.
     
    thanks
    Davchr

    0
    #137920

    Romel
    Member

    Please try to check first if among the SPC charts, I-MR is the most feasible to be used on your end. This is only applied for processes whose output is homogeneous, has longer interval before another output emanates and the inspection is costly.
    The sample size will be ok as long as it meets the criteria above. Better be careful to use n=1 in determining UCL and LCL.
    As for the dimension, irregardless if it is an X or Y as long as it is an important CTQ. 

    0
    #137945

    Ken Feldman
    Participant

    Mr. Romel,
    I looked everywhere but couldn’t find a value for n=1 in any table of constants for control charts.  Can you help me out?

    0
    #137972

    Elbrin
    Participant

    Forget about the control charts if you are not meeting your customers demands it does not matter if your process is in control or not.

    0
    #137973

    hk
    Participant

    Davchr
    I think you might have several problem.
    1. You measure variation. If you same sample measure 32 times and will be out of control, which means your measurement system might have repeatability problem. You should solve this problem before plotting any control chart. That’s why six sigma having MSA in measure phase them control chart in control phase in DMAIC.
    2. What is you sampling frequency? Are you plotting data from individual data from the output? According to you, you plot 32 consective sample on I-MR chart, you are using MR (Moving range) to determine the I chart spread (Sigma). In this case, the “rational subgroup” is 2 subsequence sample. Thus, you should not have break in between the sampling. This method normally done for computering system for every sample measuring.
    3. “The sigma will be very small and the not representative of process variations over time.”. Ask yourselves a question, why long term variation is much great than short term? If you can us short term variation to detect long term variation, won’t it be good to capture the special cause and react to it? If you are saying the long term variation is well known greater than short term variation, and the long term variation is accepted by customer, management that could not be solved due to not feasible or technology limitation, you done plan to detect long term variation based on short term variation, then you might want to change your sampling frequency to long term sampling.
    4. One control chart should be used for one variable only.

    0
    #138158

    Joe Druecker
    Participant

    Absolutely – to both questions. Better than I can explain is Don Wheeler’s magnificent treatment of the robustness (robustivity?) of the 3 sigma control chart to handle abnormally distributed data and still be accurate in detecting a special cause.  The question of how many samples to use before charting is also definitively handled, as well as the myth of extrapolating ppm from a few data points. Understanding Statisitical Process Control is THE text on this subject. I used it in teaching adult con-ed evening courses for 5 years.  I had a wide variety of math abilities in my classes, and no one ever said the book was too hard to use or understand.  Anything Dr. Wheeler writes is real world and pure gold.

    0
Viewing 8 posts - 1 through 8 (of 8 total)

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