I know what you mean by cost and time pressure. Now, I am struggling with the design which would help me to find trends in product defects elimination. My problem is that my defect ratio is low but still too high to be acceptable. My goal is to get down to 50 PPM's but my current estimates are 1-5K PPM's. My sponsors suggested DoE which just aim to obtain good product in the end. However, my fear is that due to the limited number of samples that we afford to test (100-120 per run for 2^3 runs) we may end up having zero or one defects (go/no go gage!) for chosen runs and I will not be able to differentiate one run from another. This DoE would be a waste of time and money then.
My idea is to set factor's levels at slightly elevated level to observe increased number of defects. Then I would be able to see trends. What do you think? Any experience in such cases?

What you are proposing may help identify important control factors, but may not predict well when trying to lower the defect level.

Is there any way at all of measuring a variable factor instead of defects? Any additional information would be helpful.

the approach for very rare defect rates can be significantly different than that for 'normal' defect rates.

Since there are many options - it will help us if you can give us more details about the nature of the defects youare expereinceing and the process or product that is having the defects.

My product is extruded HDPE bottle and the test is drop test. I would rather not replace this test by wall thickness measurements as I suspect that bottle split line is not perfect. There are at least two vital process parameters identified so performing DoE in order to find trends to make sure that I am closer to optimum point would be satisfactory for now - even if I would not be able to proof that DPMO < 50.


the approach for very rare defect rates can be significantly different than that for 'normal' defect rates.

Since there are many options - it will help us if you can give us more details about the nature of the defects youare expereinceing and the process or product that is having the defects.

What specific defects do you see? Do these defects relate to anything measurable?

You mention a drop test and split line (mold parting line?). Do the defects relate to a fracture along this line? Can some type of pull or shear test be performed along this joint? Variable data will provide much more meaningful information.

What specific defects do you see? Do these defects relate to anything measurable?

You mention a drop test and split line (mold parting line?). Do the defects relate to a fracture along this line? Can some type of pull or shear test be performed along this joint? Variable data will provide much more meaningful information.
Thanks a lot, that maybe sounds silly but I did not think of performing shear stress as I do not have the right equipment to perform such test. However, it may be worth considering. The good thing is that I would have a number instead of go/no go test. This would be more time-consuming and still no limit value but still worthwhile. Thanks!

Building on Miner post, You may want to consider using a 'burst test' and its data in your DOE, as it would be a measurable variable instead of just a go/ngo measure. The pressure readings would be an indicator of the quality of the 'parting line' prior to failure you may be able to correlate this to your modlign parameter. All you would need is a hose and a pressure gage if you wanted to keep costs down.don't forget a protective cage around the test sample.
Hope this idea helps.

I prefer the second option, however I am a bit afraid of introducing another error. I gain measurable factor however the bottle failure during drop test is caused (as far as I am concerned) by dynamic loads and shockwave. I am not sure whether this test is equivalent. Do you have any idea if there is any science that I may use to proof that what I plan to do is correct?

Building on Miner post, You may want to consider using a 'burst test' and its data in your DOE, as it would be a measurable variable instead of just a go/ngo measure. The pressure readings would be an indicator of the quality of the 'parting line' prior to failure you may be able to correlate this to your modlign parameter. All you would need is a hose and a pressure gage if you wanted to keep costs down.don't forget a protective cage around the test sample.
Hope this idea helps.

A burst test will apply consistent tensile stresses on the parting line. I would speculate that drop stresses would primarily be shear caused by compressive shock induced buckling with some possible tensile stresses.

Have you ever microscopically analyzed the failed edges? Analysis of metal edges after failure will indicate the type of stresses that caused the failure.

In some cases the various strengths will roughly correlate with each other, but this is not always the case, so it is best to test using the same type of stresses that cause the failure. However, a drop test, by its very nature, will apply these stresses in a very random fashion.

Your very low PPM rate may indicate that each failure is the result of a "perfect storm" of stresses. That is, each time the test unit happens to fall exactly on the parting line, you get a percentage of failures, and no failures occur at any other time.