iSixSigma

Destructive Design of Experiment

Six Sigma – iSixSigma Forums Old Forums General Destructive Design of Experiment

Viewing 11 posts - 1 through 11 (of 11 total)
  • Author
    Posts
  • #46855

    Sinnicks
    Participant

    All,
    Please advise.
    I am trying to conduct a DOE using three factors:
    1) Surface finish with 4 Levels
    2) Oil Type with 2 Levels (The levels are High-with & Low-without oil)
    3) Supplier Type with 2 Levels
    The nature is to test bolts to destruction with and without an oil coating, because it is destructive I cannot figure out how to obtain a replicate as I can’t use the same bolt twice.
    Am I barking up the wrong tree……….If someone has some advice then please let me know.
     
    Regards and thanks
    Mark

    0
    #155480

    Robert Butler
    Participant

      Bolt-to-bolt variation is going to be part of your final process so even if you could run all of the factors with the same bolt you wouldn’t want to because the results of those experiments would assume perfection with respect to bolt manufacture. 
     What you really want to know is do any of the three factors of interest matter in the presence of bolt-to-bolt variation.  I’d recommend the following: Take a sample of the bolts, randomly assign one bolt to each experimental condition, and analyze your data in the usual fashion.  The results of this analysis will tell you which factors, if any are significant and, if you run a Box-Meyers analysis, it will also give you a measure of the variation due to each factor.
      If none of the factors are significant, take a small sample of the bolts, set up a standard test condition and test these to destruction.  Use the results of this analysis to determine if the reason for lack of significance was due to excess bolt-to-bolt variation or if it was due to the fact that none of your factors really mattered.

    0
    #155767

    SB
    Member

    “because it is destructive I cannot figure out how to obtain a replicate as I can’t use the same bolt twice.”
    Using the same bolt twice, if you could do it,  is not a replicate but a repeat measurement. In order to create replicates you would use different bolts at the same settings. This would then give you an estimate of the error due to bolt to bolt variation.
    Regards.
     
     

    0
    #155770

    SB
    Member

    Hi Robert,
    I can’t agree with your method of carrying out all the bolt to bolt variation tests after the main DOX.  The replicates and the main DOX  should be considered together and the runs randomly distributed.
    Regards.

    0
    #155776

    Robert Butler
    Participant

      The only reason for my suggesting additional post hoc testing was to provide information should the results of the DOE be inconclusive (that is to say – none of the model terms are significant and the apportioning of variability via the Box-Meyers fails to account for a large part of the observed variability). 
      In those instances when this occurs people tend to spend a lot of time wondering about the causes of the magnitude of the unexplained variation. In this case the obvious first choice for a culprit would have been the bolts over which randomized our design.  All I was suggesting was that, should the above occur, a quick, post hoc, check of the bolts can confirm or deny this suspicion. 
      You could, as you suggested, sprinkle tests such as these through the design space and run a separate analysis on these tests. The reason I wouldn’t recommend this course of action is that there have only been a few times in my life where a design failed to deliver information on variables impacting mean location and/or variability.  With this kind of experience as a guide, routine tests for specific things that might contribute to the residual error (that is things over which the design was randomized such as bolt-to-bolt variability) would only consume time and money that, in all probability, could be better used elsewhere.

    0
    #155780

    Craig
    Participant

    Mark,
    It looks like you have some confusion regarding replication. (Probably stemming from your GRR training). In  GRR studies, you actually replicate by measuring repeatedly on the same units. The treatment combinations are defined by operator and part, and each condition is replicated 3 times. It gets confusing becasue you only have 10 parts, but 3 x 3 x 10 = 90 total observations. When you run an experiment on a process such as yours, you will have the same number of observations as parts (experimental units)
    In your example, the experimental unit appears to be “bolt”. I agree with Robert in that you would randomly assign a treamtement combination to your sample of bolts. Can you go to the stock room and randomly pick bolts from the two suppliers and also with the 4 different surface finishes? If that is true, then randomly assign the oil type to each of them and conduct the destructive tests. I would make sure the test sequence is randomized so you can block the effects of tester variation. When in doubt….randomize!
    Good luck.

    0
    #155784

    ezweld
    Participant

    Mark,
     
    I have found this forum to be extremely callous when it comes to dynamic measurements.
    The simple truth is you can’t repeat your measurements and you have by nature a large measurement variation. The same bolt will never break the same way twice. To be able to use DOE’s with dynamic measurements you have to increase your sample size. For the lucky people of this forum who can repeat their measurement they can factor out measurement noise. But for the filthily few more samples are needed to get to a comfortable mean.

    0
    #155809

    Craig
    Participant

    Ezweld,
    What is a “dynamic measurement”? I can’t say that I have heard that term before. In any case, replication is needed in DOE instead of repeated measurements. It sounds like the experimental units (bolts) would be pretty inexpensive and a large sample size can be used in the poster’s DOE. If the model comes up insignficant, it means that the noise outweights the signal.
    I would run at least 5 replicates, preferably 10 in a full factorial design and see how the model fitting goes. It is also possible to run a DOE on just one of the factors first before making the big investment in time and resources. (Oil, no oil, 5 replicates)

    0
    #155827

    ezweld
    Participant

    Haci,
    Wikipedia’s definition is the following.
    The dynamical system concept is a mathematical formalization for any fixed “rule” which describes the time dependence of a point’s position in its ambient space. The mathematical models used to describe the swinging of a clock pendulum, the flow of water in a pipe, or the number of fish each spring in a lake are examples of dynamical systems.
    A dynamical system has a state determined by a collection of real numbers, or more generally by a set of points in an appropriate state space. Small changes in the state of the system correspond to small changes in the numbers. The numbers are also the coordinates of a geometrical space—a manifold. The evolution rule of the dynamical system is a fixed rule that describes what future states follow from the current state. The rule is deterministic: for a given time interval only one future state follows from the current state.
     
    This basicly states that replication is not possible. But the range of natural variation can be observed by increasing the sample size thus giving you a picture of the process.
    Let’s say we are doing tensile testing on the bolt. We want to do the DOE on 3 factors
    1) Surface finish with 4 Levels
    2) Oil Type with 2 Levels (The levels are High-with & Low-without oil)
    3) Supplier Type with 2 Levels
     
    With a sample size of 1 your not going to get a true representation of the way a lot of bolts would test with the same factors. The bolts are not all the same. Even from the same supplier. The nature of the test is dynamic it can not be reproduced. Replications of the experiment would show the inhertent varitaion of the test. Replications may show the main effects of the mean to have a different outcome than if you used a 5 piece subgroup with the same number of replications.
    I know your next issure is the test instrumant is not adequate for the bolt. Please refer to an ASTM standard for tensile strength testing. Look at the part of the standard that tells what methods were used to arrive at their standard. The amount of variation between testing labs may surprise you.
     

    0
    #155830

    Craig
    Participant

    Ezweld,
    Replication means applying the same treatment combination to more than one experimental unit. There is no question that replication is possible. REPEATED MEASUREs would not be possible for destructive testing. My opinion at this point would be to run an experiment as shown below (4x2x2 full factorial with 5 replicates). If the test time prevents the experimenter from completing all the measurements in one set-up, he would have to add “blocks” to the experiment.

    Pattern

    Finish

    Supplier

    Oil

    Strength

    212

    2

    Supplier A

    Oil

    .

    412

    4

    Supplier A

    Oil

    .

    111

    1

    Supplier A

    No Oil

    .

    222

    2

    Supplier B

    Oil

    .

    311

    3

    Supplier A

    No Oil

    .

    212

    2

    Supplier A

    Oil

    .

    222

    2

    Supplier B

    Oil

    .

    112

    1

    Supplier A

    Oil

    .

    311

    3

    Supplier A

    No Oil

    .

    411

    4

    Supplier A

    No Oil

    .

    222

    2

    Supplier B

    Oil

    .

    112

    1

    Supplier A

    Oil

    .

    211

    2

    Supplier A

    No Oil

    .

    212

    2

    Supplier A

    Oil

    .

    211

    2

    Supplier A

    No Oil

    .

    312

    3

    Supplier A

    Oil

    .

    121

    1

    Supplier B

    No Oil

    .

    421

    4

    Supplier B

    No Oil

    .

    112

    1

    Supplier A

    Oil

    .

    321

    3

    Supplier B

    No Oil

    .

    222

    2

    Supplier B

    Oil

    .

    412

    4

    Supplier A

    Oil

    .

    421

    4

    Supplier B

    No Oil

    .

    321

    3

    Supplier B

    No Oil

    .

    411

    4

    Supplier A

    No Oil

    .

    311

    3

    Supplier A

    No Oil

    .

    122

    1

    Supplier B

    Oil

    .

    111

    1

    Supplier A

    No Oil

    .

    411

    4

    Supplier A

    No Oil

    .

    221

    2

    Supplier B

    No Oil

    .

    121

    1

    Supplier B

    No Oil

    .

    122

    1

    Supplier B

    Oil

    .

    411

    4

    Supplier A

    No Oil

    .

    221

    2

    Supplier B

    No Oil

    .

    422

    4

    Supplier B

    Oil

    .

    211

    2

    Supplier A

    No Oil

    .

    212

    2

    Supplier A

    Oil

    .

    322

    3

    Supplier B

    Oil

    .

    122

    1

    Supplier B

    Oil

    .

    121

    1

    Supplier B

    No Oil

    .

    311

    3

    Supplier A

    No Oil

    .

    112

    1

    Supplier A

    Oil

    .

    411

    4

    Supplier A

    No Oil

    .

    221

    2

    Supplier B

    No Oil

    .

    322

    3

    Supplier B

    Oil

    .

    321

    3

    Supplier B

    No Oil

    .

    221

    2

    Supplier B

    No Oil

    .

    422

    4

    Supplier B

    Oil

    .

    211

    2

    Supplier A

    No Oil

    .

    112

    1

    Supplier A

    Oil

    .

    311

    3

    Supplier A

    No Oil

    .

    221

    2

    Supplier B

    No Oil

    .

    321

    3

    Supplier B

    No Oil

    .

    312

    3

    Supplier A

    Oil

    .

    121

    1

    Supplier B

    No Oil

    .

    212

    2

    Supplier A

    Oil

    .

    122

    1

    Supplier B

    Oil

    .

    312

    3

    Supplier A

    Oil

    .

    421

    4

    Supplier B

    No Oil

    .

    111

    1

    Supplier A

    No Oil

    .

    111

    1

    Supplier A

    No Oil

    .

    322

    3

    Supplier B

    Oil

    .

    412

    4

    Supplier A

    Oil

    .

    421

    4

    Supplier B

    No Oil

    .

    312

    3

    Supplier A

    Oil

    .

    222

    2

    Supplier B

    Oil

    .

    321

    3

    Supplier B

    No Oil

    .

    422

    4

    Supplier B

    Oil

    .

    111

    1

    Supplier A

    No Oil

    .

    312

    3

    Supplier A

    Oil

    .

    312

    3

    Supplier A

    Oil

    .

    112

    1

    Supplier A

    Oil

    .

    211

    2

    Supplier A

    No Oil

    .

    322

    3

    Supplier B

    Oil

    .

    111

    1

    Supplier A

    No Oil

    .

    222

    2

    Supplier B

    Oil

    .

    311

    3

    Supplier A

    No Oil

    .

    121

    1

    Supplier B

    No Oil

    .

    322

    3

    Supplier B

    Oil

    .

    122

    1

    Supplier B

    Oil

    .

    411

    4

    Supplier A

    No Oil

    .

    122

    1

    Supplier B

    Oil

    .

    422

    4

    Supplier B

    Oil

    .

    421

    4

    Supplier B

    No Oil

    .

    412

    4

    Supplier A

    Oil

    .

    421

    4

    Supplier B

    No Oil

    .

    221

    2

    Supplier B

    No Oil

    .

    422

    4

    Supplier B

    Oil

    .

    422

    4

    Supplier B

    Oil

    .

    211

    2

    Supplier A

    No Oil

    .

    121

    1

    Supplier B

    No Oil

    .

    212

    2

    Supplier A

    Oil

    .

    412

    4

    Supplier A

    Oil

    .

    321

    3

    Supplier B

    No Oil

    .

    412

    4

    Supplier A

    Oil

    .

    322

    3

    Supplier B

    Oil

    .

    0
    #155833

    ezweld
    Participant

    Haci,
     
    Your right, I was thinking about Taguchi DOE. I would rather run a Taguchi with a higher subgroup and lower replication. Not that it’s any better it just what I would want to do.

    0
Viewing 11 posts - 1 through 11 (of 11 total)

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