Setting Quality Objective for Go/No Go Leak Test
October 31, 2009 at 7:43 pm #52861
New poster – hoping to harvest the knowledge of this
powerful community..We make containers that have a lid that when
assembled should not leak. The containers have an
oring around the diameter to help accomplish this.As a result of variances within the container,
oring, and cover, the lids may leak. The QA testing to reveal the leaks is time
consuming. We are trying to determine what
statistical tools and what number of samples we will
need to prove out any redesign.At the moment we have no better test than a “go/no
go” leak test, and realize statistical tests for
this are not very sensitive. Does anyone have any input on how to implement a
more sensitive, variables-based test, that will help
us reduce the number of samples required? And what is the most effective statistical tool for
what we have now to help us set a quality objective? We essentially want to assert with a certain
confidence level that these covers will not leak.
the same cover may leak or not leak depending on how
they are installed, which needs to be factored into
this…0November 1, 2009 at 2:21 pm #186511
First off, what is the approximate leak rate that you are seeing? That drives required sample sizes in the go-no-go realm of measurement. The higher the rate of failure, the easier to remain in the go-no-go measurements, while as the rate drops down it would be much more convenient to have a continuous measure to rely upon.
Since you suggest the variation of dimensions is one of the contributors to this, which should be verified by a Pareto of leak related failures, it would likely be helpful to correlate the leak failures (p’s versus f’s) to the dimensional set of characteristics of the lids, etc. This would take some effort and data analysis, but it would drive your learning on what combination of sizes causes your leaks to occur.
Finally, the last comment about installation leaks is somewhat troubling, and you need to assess how much the leak issue is determined by the manufacturing issues, versus the installation process. Again, a Pareto of leak related failures would likely be very helpful to get some perspective on where to place your resources.0November 1, 2009 at 11:32 pm #186522
About a 4% leak rate out of 100 tests. The test is
fairly simple- put the cover on, if it leaks, it
fails. the same cover could pass and then fail on
the next trial. we know the diameter is key to
fixing the problem, as it determines the amount of
compression on the oring.what we dont know how to do yet, is determine a
criteria for which we can say that our new design is
statistically different at sealing than the old
design. that is the main question…we dont want to run 100’s of tests to prove the
difference, if we can avoid it…0November 2, 2009 at 1:01 am #186524
Unfortunately, a two sample test of proportions from Minitab’s calculator shows sample sizes of n=376 per group, and that is to see a difference from a 4% null hypothesis to an alternative rate of 1% on the new design (alpha=0.05, power=0.75). For 2%, the sample size jumps to about a 1000 per group so unless you are fairly certain the new design is better it does not appear this is the way to go as you have surmised.0November 2, 2009 at 2:37 am #186525
SeverinoParticipant@Jsev607 Include @Jsev607 in your post and this person will
be notified via email.
Obviously you need to make a change to your measurement technique. One option would be to implement pressure decay testing. You could bring the packaging up to a certain set internal pressure and then measure the time it takes to drop by X pressure. Presumably, as your seal improves the time it takes to achieve a set pressure decay should increase. If you can implement such a test, then you now crossover from attribute data to variable data and need a significantly lower sample size.
Another method may be to just ramp the internal pressure and measure the peak at which the seal fails. The only issue here is that you may need to constrain the lid if there is a risk that it will shoot off and become a projectile (and also you want to measure the quality of the seal it seems).
Once you’ve established a good measurement system your next step should be to conduct a thorough baselining of the current design so that you know where you truly are. Once this is established then you can use the information to determine a measureable improvement goal. The goal can then be used in your statistical power calculation to determine your sample size.0November 3, 2009 at 2:38 am #186552
The forum ‘General’ is closed to new topics and replies.