iSixSigma

Cycle Count – Attribute or Variable

Six Sigma – iSixSigma Forums General Forums General Cycle Count – Attribute or Variable

Viewing 14 posts - 1 through 14 (of 14 total)
  • Author
    Posts
  • #54414

    Tim
    Participant

    Given a cutter or blade as a factor in a DOE or 1FAT, if I want to increase the life of the cutter and evaluate the number of cuts it is capable of before failure would the number of cuts be variable or attribute data? The number is an integer which would lead me to believe that it is attribute but its not a good/bad situation like a standard attribute. I tried using the inverse of the cutter count as a proportion but was told that this method was not an accurate method of transforming my data. Is this data attribute? And if so, what six sigma tools can I use to evaluate my results.

    0
    #195277

    MBBinWI
    Participant

    @timfran87 – Tim: How do you evaluate whether a cut is acceptable or not? I’m guessing that you have some continuous variable that you are using to judge acceptability. You can use this instead of the number of cuts (with correlation of number to variable measurement).

    0
    #195278

    Tim
    Participant

    It would be difficult to use the measurements that verify good/bad because the cause of bad is a function of multiple factors. We assemble components and machine two ends of the assembly flush. However after a heating application some of the components have moved after being cut and some have not. I am ultimately trying to determine the root cause of those components moving whether it be dull cutters, too hot of cooking temps, incoming component position, some other factor, or an interaction of multiple. Also some historical data contains the amount of cuts per cutter before it was replaced because of quality complaints (not validated measurements) of product running out of spec. More recently there has been a process change and I would also like to see if the cutter life has increases since the process change.

    0
    #195284

    MBBinWI
    Participant

    @timfran87 – you’ve just described the multiple variables that you need to evaluate (and ultimately either control or render innocuous). A good DOE will include these different variables and show you main effects (how the individual variable affects the outcome) as well as the interactions (how combinations of changes in variables affect the outcome). Find a competent DOE SME to help you design a rational study and you will be light years ahead in your understanding of the system.

    0
    #195285

    Tim
    Participant

    I have already performed the DOE and determined that out of all of the factors mentioned above that cutter condition was the only significant factor. That is when we ran the 1FAT to see what the life of a cutter would be if it was cutting off different amounts of material. This questions stems from historical data that I already have though. A process change was made about 1 month ago. On that date it appears that the cutter life increased but I do not know how to show if it is different. I have the cutter life for both sets of data (before and after the process change) but do not know how to compare the two data sets. For example before the process change the cutters averaged 2000 cuts before they needed changed and currently they average 3000 cuts before they need changed. The reason they are changed is because of the complaint of a quality auditor who claims that they are seeing defects caused by worn cutters. So how do I compare the 2000 to 3000 average cuts? Looking at it seems good, but should I use the number of cuts as an attribute or variable? And based on whichever type of data it is how should I evaluate it. If it is variable do I treat as normal variable? If it is attribute how do I convert the quantity of cuts to an attribute (I tried the inverse but do not know if that is correct)? I apologize if I have not explained this very well but this really has me stumped.

    0
    #195286

    Robert Butler
    Participant

    Are you saying that you have multiple counts of acceptable cuts and that when you take an average of these multiple counts before the process you get an average of 2000 cuts and after the process change you get an average of 3000 or are these two numbers just some kind of estimate. If they represent genuine averages of count measures (I find it odd/interesting that they would be these exact numbers but we’ll let that go for the time being)then a two sample t-test would suffice for comparison.

    What bothers me is the quality auditor claim of defects caused by worn cutters – how is this quantified and what is the reproducibility of the claim. As for the number of cuts – it’s count data not attribute so all of the usual methods for analysis of continuous data can be used when assessing it.

    0
    #195287

    Ali
    Participant

    Assuming that there is predefined consistent process to determine when the cutter need to change , I would use chi-square goodness of fit test (from stat-tables), stack before and after data in one column designated by category name.

    0
    #195288

    Ali
    Participant

    Assuming there is consistent process to judge when the cutter need to be changed , I would use chi-square goodness of fit test (from stat-tables).

    0
    #195289

    Chris Seider
    Participant

    Agreed @rbutler What is the real quality output we should be looking at?

    0
    #195292

    Bartosz Jankowski
    Guest

    Dear Tim,

    1. I would use number of cycles as variable (continous) data and analyze it as such. Why? With large enough sample there is no significant difference between normal distribution (most frequently used for continuous data) and Poisson distribution (for discrete data).
    In “variable” case, use 2-sample t-test (if normal distribution) with before-after comparison or use median comparison test (Kruskal-Wallis) if distribution is non-normal. This way, You will have a clear test for decision making.
    I would not use “attribute” approach, mainly because number of cuts is direct reflection (coding) of blade usage time (continuos data).
    For distribution identification, use Normality test (exists in most of the stat software apps) or distribution identification tests (Stat/Quality tools/Individual distribution identification in MTB 16).

    0
    #195296

    Tim
    Participant

    Thank you everyone for your comments. I wanted to take some time to address each question about my process and then ask some clarification questions.
    1. @rbutler- I used the values 2000-3000 as an example. the real values are like 1807 to 2856 but the data is at work and I am home fore the weekend. As far as the quality inspectors claim (this is going to open a can of worms) but I know it is not a repeatable or reproducible method of determining cutter life but it is all we have at the current time. Ideally we would automate the inspection with a vision system but I cannot justify the investment cost with the projected ROI because of the lost cost of the defect. Your second paragraph is my big investigation though. IS COUNT DATA REALLY CONTINUOUS?
    2. @thepig I have never used chi-square test. Is it meant for attribute data, variable data, or both and is it like a 2 sample-t or 2 proportions test?
    3. @cseider the defect is the exposure of a component of an assembly after it is cut. I have already shown that the exposure gets worse the longer a cutter is on, but the “determined” life of the cutter has increased suggesting that the issues is not presenting itself as early as it use to.
    4. @brotherdargon How can I determine if I have enough samples to know that there is not a significant difference between the normal and Poisson distributions? Do I test both and see if the tests yield the same results and if I do, then make that claim? I can test the normality easily when I get back to work but is there an attribute test for non-normal data to compare to the Kruskal-Wallis test to show that I have enough samples?

    Thank you all again for your comments this is really educational for me.

    0
    #195297

    Tim
    Participant

    Sorry correction

    2. @theqig I have never used chi-square test. Is it meant for attribute data, variable data, or both and is it like a 2 sample-t or 2 proportions test?

    0
    #195298

    Robert Butler
    Participant

    Count data can be viewed and treated as continuous data. When these discussions occur (and they do with predictable regularity) the thing that should be kept in mind is not the issue of discrete vs. continuous but that of interval vs. ordinal. Count data is definitely interval and if you have a large number of values that you count can assume then it is reasonable to use the statistical methods one would use for continuous data.

    From Agresti – Categorical Data Analysis, 2nd Edition, pp.3 “Variables are classified as continuous or discrete, according to the number of values they can take. Actual measurements of all variables occurs in a discrete manner, due to precision limitations in measuring instruments. The continuous-discrete classification, in practice, distinguishes between variables that take lots of values and variables that take few values. For instance, statisticians often treat discrete interval variables having a large number of values (such as test scores) as continuous, using them in methods for continuous responses.”

    The description of your problem meets the above criteria so running a two sample t-test would be appropriate. It should be noted that the t-test is robust with respect to non-normality so it should give you an acceptable answer. If you are really concerned about all of the internet noise concerning t-tests and the supposed need for data normality you should run the t-test side by side with a Wilcoxon-Mann-Whitney test (the non-parametric version of the t-test) and see how they compare. The odds are they will tell you the same thing although their actual P-values will be numerically different.

    You should also get yourself a copy (through the library) of The Design and Analysis of Industrial Experiments 2nd Edition by Owen Davies and read pages 51-60 on the effect of departure from normality on tests of significance to put your mind at ease with respect to the ability of the t-test and ANOVA to function in the presence of non-normal data.

    0
    #195300

    Ali
    Participant

    @Tim ,
    Chi Square Test Can be used for both discreet and variable data . Commonly used to show change or improvement when working with discrete data.
    It tells you how is observed distributions of your data (before & After ) fit the expected distributions under the null hypothesis.(if you are using Minitab , Make sure you check the graph options)
    Furthermore , you can use Kruskall wallis (although I brefere Moods median test specially if I have outliers in my data).

    0
Viewing 14 posts - 1 through 14 (of 14 total)

You must be logged in to reply to this topic.