iSixSigma

DOE response analysis – multiple readings/ run

Six Sigma – iSixSigma Forums Old Forums General DOE response analysis – multiple readings/ run

Viewing 7 posts - 1 through 7 (of 7 total)
  • Author
    Posts
  • #37116

    Lomax
    Participant

    Can anyone please advise?
    I plan to conduct a DoE in which the response output for each run is a current trace waveform over time showing all the dips and peaks of the spindle motor current throughout a cutting cycle.
    Can Minitab process the various variable data readings extracted from the current trace for each run in terms of analysing the DoE?
    Any other suggestions?

    0
    #108631

    Anonymous
    Guest

    Neil,
    The only experience I have in this regard is multivariate ‘torque’ control. If you do not have an ‘torque’ encoder then I would not be able to offer you any assistance.
    Andy

    0
    #108633

    Tim F
    Member

    Neil,To answer this question, it would be helpful to know what you hope to achieve from this data. Assuming the current vs time data is in digital form, Minitab (or just Excel) could certainly extract a variety of information, like max, min, mean, st dev, etc. Or you could count what fraction of the time the current is above some desired limit. Any of these numbers could be used as a response for your DoE analysis.If you want to actually analyze the structure of the dip and peaks, you might try autocorrelation in Minitab. Another option would be Fourier analysis, but that might be more than you need (and Minitab doesn’t do it).Tim F

    0
    #108635

    Robert Butler
    Participant

      The measures you are taking are called repeat measures.  If the measures over time or distance matter and you wish to capture that as well as assess the impact of the variables in your design you will have to use a mixed models approach with repeated measures.  As far as I know Minitab will not handle data of this type. There may be other packages but the only one I know which can handle this kind of data is SAS.
      If you have access to SAS or to something comparable your coding for each experiment will have a variable for time or distance for each measure of current.  You will have to indicate, using some kind of class or grouping statement that time/distance and the experiment identifier associated with each experiment are class/group variables.  You will then have to indicate the repeats are on time/distance and that these are within each experiment.
      Because of the lack of independence with respect to successive measurements, you will also have to indicate the correlation structure of your repeat measures.  If you have a package that can do all of these things your final models should prove to be extremely valuable. 

    0
    #108637

    lin
    Participant

    Robert Butler,
    For a DOE run, could you please explain the difference between a repeat run and a replication run? How and when do you use them? Thank you!

    0
    #108642

    Robert Butler
    Participant

    In order for a run to be a replication you, as the experimenter, must reset your entire apparatus and then completely re-run the experiment.  If, for example, you are mixing several ingredients to make a plastic extrudate, in order to qualify for a replication you would need to go back to your raw materials, select appropriate quantities from these raw materials (if you are trying to include lot-to-lot raw material variation in your run then this would mean selecting the raw materials from lots different from those used for the first run), and make a new batch of compound from scratch. 
      If process variables were part of the design you would also have to have cooled down/shut down the extruder and then brought the process back to its settings for temperature, speed, etc.
      For a repeat or, as it is sometimes called, a duplicate run, in the above case you would just take another sample from the same mix and run it through your processing machinery without making any changes.
      A replicate run is an attempt to generate a measure of the variation of your process. Specifically, variation you could not capture by virtue of the fact you cannot control every single source of variation in the process.  The results of a replicate, when compared to its counterparts (other experimental runs with the same settings) is a measure of unassigned process variation.  It is this measure of variation that is used to test assertions concerning things such as the significance of process mean shifts.
      A repeat run can give you some measure of the minimum amout of variation you could expect to see in the process given that everything about the process never changed.  It can be used (or abused) as a rough measure of process possibilities.  The abuse comes from the fact that if you report out your repeat variation, over time people will forget the the word “repeat” and just remember variation. The scenario I’ve witnessed time and again is this: six months or a year after reporting this level of process variation the number shows up in an annual report with the word “repeat” missing. When an audit reveals the current level of variation is no where near this value, everyone and their grandmother gets raked over the coals because “our process variability went to —- during the last six months!”
      Repeat measures, continuous measures made on aspects of a single run, are often needed to characterize process behavior – stress-strain curves, shear curves, taking a patients BUN readings at 1,2, and 3 months after an operation, etc are often the way the output within a single experiment is measured.  As I mentioned in the previous post, these measurements require special handling.  If you don’t treat repeat measure or repeat runs properly what will happen is your computer program will assume the variation between repeats is actually the measure of your unexplained process variation. This variation will be very small and conesquently, practically every variable you changed, along with most of their interactions, will test significant and you will come to the conclusion that you have made some significant discoveries about the process when in fact all you have done is make a mistake.

    0
    #108683

    Manee
    Participant

    You could try rdExpert for multiple responses and analysis.
    Manee

    0
Viewing 7 posts - 1 through 7 (of 7 total)

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