Statistical Cycle Count Method

[cdaringe]

Hi all:

I'm in search of the correct statistical technique to set the number of cycles I need to push my product through during test.

Historically these values have been arbitrarily set. Currently, I cycle it 500 times per test, which extends a full shift. That's a lot of NVA cycle time i'd like to cut back on.

I have failure data/unit--at what cycles failures occurred and had to be corrected.

I want to develop a model for gauging the # of cycles to catch X% of failures, likely with an alpha of 1-2%.

Many unit pass through fail-free, others with multiple failures/unit. Not sure what method to pursue! Would appreciate some tips!

Thanks,

-Chris

 

[roscoesplevins]

Is the historical data PASS/FAIL or is there a variable result with limit(s).

Tried DOE or ANOVA to determine cause and correlation to multiple failures?

 

[cdaringe]

Hey Rosco:

The failures are binary, as you mention. PASS/FAIL, not a continuous measure. Really the fail reasonons are a multitude of different reasons, but we treat them as simple PASS/FAIL. We do root-cause on most of them, but because of the dynamic nature of our product (not stable--we're high customization, low volume), we are chasing a moving target in regards to proper setups. DOE & ANOVA, unfortunately, won't get me too far here as I don't an explicit set of defined variables to work with.

Thus, we want to use historic pass/fail data exclusively to set our testing levels, regardless of the failure diversity we experience. It sounds dangerous and unwise, but based on the failure modes, we believe this to be safe and acceptable.

Thx!

 

[roscoesplevins]

Understood. Your situation may be similar to ours; setup dependent, short run, multitude of seldom run part numbers.

Use the address below. This is the MathDL page for Excel based calculation and graph of the hypergeometric sample size. Very easy to use, but not exactly what you need. This could however be good background justification for whatever plan you implement.

http://mathdl.maa.org/mathDL/4/?pa=content&sa=viewDocument&nodeId=428&pf=1