Most reliable low power (5-10W) SMPS design?

Hi everyone,

I have noticed that with modern consumer electronics, the switching mode power supply is almost always the weakest part regarding realiability. However, the laws and regulations are continously pushing designers from linear power supplies to SMPS because of energy effeciency. This means that we are forced to use SMPS even in devices that should reliably last for more than a couple of years. I am particulary worried about white goods, for example intelligent refridgerators, electric stoves/ovens, coffee machines, etc...

I have decided to start learning some SMPS design basics by building a simple 230VAC -> 5VDC, 5-10W power supply and thought that it would be interesting to try to optimize for reliability from the beginning. So, I present a couple of questions for the more experienced designers and hope to get some feedback.

1) What do you think are the weakest components in a SMPS of this power level? I think we all can agree that the electrolytic capacitors (in secondary circuit) are statistically the weakest link, but what comes after them? Maybe optocouplers because of their LED aging? What about the other components like diodes, ICs and FETs, do you think that we can list them in an order of "general" reliability?

2) What is the most reliable topology for this kind of SMPS? I found this primary-side regulated flyback configuration:

hxxp://www.powerint.com/sites/default/files/PDFFiles/rdr158.pdf

This looks very good to me because there is no optocoupler (one less failure prone component) and the overall component count is very low. But what is the catch? Why do almost all SMPS have an optocoupler? Is it just for saving money on the simpler transformer?

3) Even with all things considered and using high quality electrolytics, do you think that a SMPS will ever be able to reach the respectable 15-20 years of service life of a well designed linear power supply?

Did you ever hear the apocryphal story about Henry Ford (the founder of the american car manufactuer, Ford)? He sent someone round the scrap yeards to examine abandoned and broken old cars to find which parts only failed rarely, if at all.
His intention was to reduce the specification of those components which survived well, But you could do something similar from the opposite point of view, collect abandoned and failed SMPS units and note the frequency of each component's failures to determine which components fail most often.

There are some SMPSs in the workshop here which have been running continuously for over 20 years, day and night.
We had three units come in for repair recently which had been in use 24/7 for 15 years or longer, and all with SMPS failure. Replaced a few caps which had fallen below spec. and they were back in service. (One of them had other faults but not in the SMPS). I don't have the inclination but if I trawled through boxes of old PC PSUs I guess I'd come up with the same evidence.

Surely much of the answer to your question depends on what CAUSES of failure you want to anticipate in your design. Some units will be designed with better resilience to thermal problems than others, some will have over-rated components, bigger contact areas on their connectors, some will have mechanical strength and resistance to vibration, some just bigger & wider tracks on the boards. What are the causes of failure if they're not voltage, current, heat and stress?
If you tried to plan a generic resilient design, you could easily find yourself with 'the Henry Ford problem' - as soon as you increase the spec. of one common point of failure you will then find that the most common cause of failure is somewhere else!

Oh, I nearly forgot. The last factor for designs which fail is cost. Why do you think consumer products fail more readily than their industrial, medical and military equivalents?