Weibull shape factor selection databases

[Brian Ray]

I have a fair amount of experience with reliability statistics (mostly Weibull analysis), but run into the same situation time and again when it comes to plan reliability tests. Namely, how do I select a reasonable Weibull shape parameter (beta) for a mechanical or electrical component to be tested without on-hand (lab testing or field) failure data? Obviously, a shape factor of 1 will yield the most conservative test plan, but this may present an "overtesting" situation in some cases. However, I cannot feel comfortable with the reduced testing that a larger shape factor dictates unless I can point to some reference.

I am aware that there are databases of typical Weibull parameters, but I don't have a good feel for how well they're accepted in industry, nor which ones are most appropriate. I'm wondering if the tables/databases in these sources would be appropriate for the components I am testing. Examples include: high accuracy syringes, electronic solenoid valves, electromechanical pump, flexible plastic parts, spring and detent systems, electrical connectors, etc.

Any suggestions or guidance would be very much appreciated.

[TRISTAN]

To achieve a comfortable feel when choosing the shape factor to plan reliability testing it is not necessary to have data from identical or similar components. My point of view is to pre-determine which kind of failures will be concerned; thus there only a limited number of typical failures modes that you will prevent, and so your gained "functionnal" experience on others previous products can be used to fix a justified related shape factor.(leakage, wear, breakdown, a.s.o shape factor will be applicable because it is a functionnal experience feeedback and not a product feedback). From my own experience in mechanical engineering a typical beta=2 would be a good basis to plan reliability tests. Reliability problems on electrical connectors a.s.o are too concern by mechanical difficulties (and if not , the best projection is beta=1...But we can think some failures modes will occured and will reduce "de facto" the conservatism hypothesis!)