Should an audio system be left on at all times?

Almarg - Supercomputers don't fail daily, otherwise we would have serious security problems (unless you talking about software crashes). Audio gear does not contain thousands of components - give me example of one. You will be lucky to find one that contains couple of hundred.

What is interesting about electronics commonly used is that if it does not fail in the first year most likely it won't fail for next 20 years (just opposite to MTBF) and when it does it's usually because of lightning or mechanical failure (switch, relay, motor, plug etc).
Hard discs that have MTBF in order of at least 5 years have 99.9% of time mechanical failure in spite of having a lot of electronics (ICs) on them.

MTBF does not apply to audio gear and probably should not be mentioned (unless we try to predict statistical failure in 500 years). Remember saying about lies big lies and statistics? Statistics already has proven that motorcycle accidents are caused by tatoos.

Kijanki -- What I meant about supercomputers was that a failure will occur somewhere in the system frequently, perhaps daily, but the systems are designed with redundancy that allows them to continue to function despite the failure, and while the failure is being isolated and repaired. In fact I believe some of them have associated diagnostic computers, whose only function is to detect and diagnose failures.

When I mentioned hundreds or possibly thousands of components, I was referring to the system as a whole, not to any one individual component. And I was including capacitors, resistors, etc., which these days are very small and can be very numerous.

I don't have schematics for any recent audio components. To some extent I'm extrapolating from my knowledge of computer motherboards, which typically contain zillions of tiny components and are probably somewhat indicative of the digital, microprocessor-based, parts of many modern audio components. But besides that, if you have ever looked under the chassis of a quality analog FM tuner, new or old, you will see many hundreds of discrete components. The ultimate example is probably the Marantz 10B of the 1960's, which I have seen the underside of, and it contains more components than I would want to count.

Your comment about not failing for 20 years or so after the first year is exactly what I was referring to earlier about failure rates being non-linear functions of device age. Failure rates are greatest during infancy and old age. They are much lower during the period following infancy and through middle age. That is well recognized in Reliability Engineering. But my point is that the relatively low failure rate during middle age can still be significant, because it will be degraded at the system level as a function of the number of devices (and potentially also by many other things, such as operating temperatures and the specific circuit designs).

The quote about there being lies, damned lies, and statistics, was actually originated by Mark Twain, and is one of my favorites. I hadn't heard the one about motorcycle accidents before -- good one!

Regards,
-- Al

When turning equipment off how far off does it go? Mine goes into a standby mode. If I turn on my cdp and eject a disc that has sat overnight the cd is warm. I'm figuring it's already warmed up. I believe the speakers require some movement for warm up more so than the electronics.

Almarg - I'm not clear on reliability and you can probably tell me where I'm making mistake but if we take device like transistor that eventually fails (everything fails) - let say in 100 years then using 1 milion of them would cause earlier failure, and in case of 1 billion of them one would fail every hour - am I right so far?

Now I have pencil on the desk in front of me. It doesn't last forever and will eventually rot - let say in 100 years (used or not). So if I take 1 milion of similar pencils one would rot in an hour after it was manufactured?

Doesn't reilability engineering assume that there might be some (very rare) faulty components (or connections)?

People believe in proportionality and therefore everything has to fail some time but for instance life of steel under stress is infinity of cycles (no fatigue) as long as strees is below certain level (not true for stainless steel and other metals).

Imagine basketball with a rope around it. Add 1m (3 feet) to it and rope will be loose by about foot everywhere (radius increases by 1 foot). Now do the same with our earth around equador, add 1 meter and you'l get axactly same result - gap of 1 foot everywhere around the earth.
r2-r1=(L+1)/2pi - L/2pi = 1/2pi L disappeared. It is very strange and counter intuitive but has certain practical implications. For instance clothing sizes for children cannot be in the same numbering scheme/scale as for adults.

Kijanki -- Good questions.

I don't want to get into non-electronic things, because I'm not particularly knowledgeable about them.

But consider a transistor or integrated circuit. Envision a plot of failure rate vs. component age, based on the assumption that it is being operated within its specifications (not always the case). A plot of failure rate vs. age will start out at some relatively high value (infant mortality), which in turn will depend on the degree of screening, burn-in, quality control, etc. (which is better for military gear than for consumer gear -- one reason consumer gear is so much cheaper). The curve will then go down to a lower level following the infancy period, and remain relatively constant until old age, when it will rise considerably.

What is most relevant to the questions we have been discussing is the middle period, where the failure rate is lowest, and is fairly constant over a considerable number of years (as you have pointed out). If we consider only that period, where the failure rate is essentially constant, then yes, if that failure rate at that point on the age curve is, say one failure per 100 years, and we have 100 components of that type, then we can expect one of those components to fail each year, on average. One set of 100 components may do better than average, and last 5 years without a single failure. Another set of 100 components may do worse than average, and have a failure within a month. But averaged across a large number of sets of 100 components, there will be 1 failure per year per set, based on the assumptions of constant linear failure rate, mtbf of 100 years, and 100 components.

If the failure rate vs. age curve is not linear and constant, such as during infancy and old age, the analysis is more complex. And as I indicated earlier, calculation of system mtbf has to take into account and properly weight the mtbf's of the different types of components, as well as their relative quantities in the system.

Hope that clarifies things more than it confuses them!

Regards,
-- Al