Cycles to Failure Data

[john vlahakis]

The cycles to failure data from an initial reliability test conducted on n=5 samples looks like this: 107, 178, 170, 236 and 188. I computed mttf of 175.5 cycles, and I obtained a Weibull shape parameter of 3.7. Is it ok to use this estimated value of beta to compute sample size for a follow-up success-test using the following relation:

n=[ln(1-c)]/m^(beta)ln(R)]-1, assuming my test-to-field ratio "m" is equal to 1.5 (i.e., the usage during testing is 50% grater than that incurred by the device in real-world, actual use? I assumed an R by C requirement of R97C95, and computed n = 20.9 = 21 samples. Are there any holes in this reasoning?

[Pantelis]

John,
I think I am missing something. Got the fact that you want to assume a prior beta. Now, how long are you running the success test? I am assuming a zero failure test. Also the 97% Reliability that you are trying to demonstrate is at what time?

[john vlahakis]

We don't have a usage spec., though the physicians deploy the device, on average, 20 actuations, and apply 3600 seconds of burn time (4 pulmonary veins @ 900 seconds, max., per vein). My initial study, of which I posted the cycle-to-failure data, was just on the mechanical actuation of the device, and had a mttf of 175 cycles. I don't have cycles-to-failure data for the electrical part of the device, just the mechanical deployment part (I'm working on developing a method to check electrical reliability, including capacitance/resistance/leakage failures as a function of energize time).

Are there reliability standards for medical devices? If so, are B10 or R by C levels given? I'm new to reliability: I have a degree in plastics engineering, and a strong background in DOE and process validation. On my own, I have read several text books on the subject, and researched this web-site. I arbitrarily selected the 97% reliability requirement. Average deployments by the users of the device are 20 deployments, but can go as high as 30. I estimated MTTF of the weak-joint in the design at 175 cycles.

Knowing that 30 cycles is the max. used by the doctors, a do you have any guidance or suggestions for setting the reliability requirement for number of cycles?

Yianni

[Pantelis]

Yianni,

I think I got it now. First with respect to a reliability spec I would look at the FDA for a specific R and C. I have been involved with similar products in the past and I know there were numbers there that the FDA considered acceptable. Unfortunately I can’t get into any more details on this; however this is a good place to start.

Now with respect to the analysis and say you want to set the spec at 30 – the max. If you put confidence intervals on your prior analysis you should get an answer. You may need to increase the sample size to hit what you are looking for by a few units. With regards to going back and doing another test by assuming beta, at the end of the day you would not learn anything more than your first test, in other words it may make more sense to run few more to failure – in addition to the ones you already did – and augment your sample size. I did not run the numbers but just from looking at them I think you should be close. Alternatively if it takes too long to run the test you could look at a success test. I computed R97C95 at 30 cycles with your beta and get about 15 units for 50 cycles.

Hope this makes sense.