Hi,
hope someone can help because we've been trying to solve that for a long time.

Problem: We are launching a new product. The question is: how many of those should we test to ensure that x% of the product will have no failure within the warranty period (5 years). Forecast are a 1000 unit per year will be sold. A 95% confidence level would be appropriate.

We've try to use statistical tool, but we don't know what test or hypothesis would be appropriate. Also, it's a new product with a new process involved and we don't have any historical data.

Usually, we're not testing any, because the process is always the same and our product development tools are good enough to ensure quality at launch.... but for this one, my confidence level is not good and I would like to rely on statistical tools not only on workers experience and actual tools.

Thanks!!!

Hello,

Check out this thread please: AQL - Can progressive acceptance be used with longer/bigger lots? (http://elsmar.com/Forums/showthread.php?t=18432) Once you are there, please also see the links to similar threads at the bottom of the page.

Please let us know if this does not answer your question.

This sounds like a classic reliability engineering question - and there are various standard methodologies.

You can test the parts on an accelerated schedule. For example, if a door will be opened on average 10 times a day for 300 days per year for 5 years, that is 15,000 openings. So you could get a test rig to open and close the door every other second for about 8 hr and get the same number of openings and closing. So in a couple days you could get some pretty good data on the reliability of the door's behavior for 5 years of actual use.

You can also test parts in extreme conditions. For example, suppose you have light bulbs designed for 120 V. If you run them at 200 V they will fail quickly - perhaps in a few minutes. Run them at 180 V and they will last longer - perhaps a few hours. At 160 V they might last a few days. With an appropriate model, you should be able to reasonably well predict how long they would last at 120 V.

Read up a bit on reliability engineering - that might give you some ideas. http://www.weibull.com/ is a pretty good site for an overview. Or it you have more specific questions ... fire away and hopefully you will get some more answers here.

Tim

Thanks Tim,
I've been reading on reliability for the next few days and the problem is always the same: I don't have any data because we are not producing enough prototype.

It's a money issue; prototype are expensive. In this particular case we will produce 4 prototype. If I test those 4 with accelerated method, how sure I am that product in the field will not fail? I would like to come up with a mathematical way to see if I have 98% or 30% of the chance that any will fail during the warranty? In fact, my goal is to prove that we need more prototype and more testing to be confident enough to launch the product, but right now I have nothing to negotiate with management.

We will sell about 1000 per year.

Thanks again!!