Great note [response to comment on Drain in the Bathtub Curve on NoMTBF Linkedin Group] – yes, there is a place for parts count prediction — not to determine the mtbf, to encourage proper derating, thermal engineering, and parts reduction, etc. It’s a start and as you note only one part of the reliability program.
Even with parts count – I suggest using Reliability and not MTBF… state the probability of success or failure rate AND the time period.
The longer life products (10 years) and products with harsh environments and other examples do have wear out or accumulative damage failure mechanisms. It does take some work, engineering, experimentation, and research to properly estimate a product end of life. Running a couple of ‘standard’ tests will often have no real meaning and should be avoided. Design and run proper accelerated life tests. If you need to know if a product will last 10 years with a low probably of failure, then finding and characterizing the life distribution, is important.
ALT’s are often expensive. Focusing on the highest risk or most likely failure mechanisms may help focus the research, yet, there isn’t a short cut.
For example, if a 25 year expected life solar panel manufacturer, sets up 10,000 panels and operates them for one year, Using the assumptions involved with MTBF, would be able to claim 10,000 year MTBF. If at that point they imply the unit will have a low failure ate at 25 years…. they may or may not be correct. I suggest that the naive application of the exponential distribution is uncalled for – hence the this campaign and request for you and other intelligent reliability engineers support to eradicate MTBF.