High reliability electronics - how to learn

[Tzolkin]

Hi!
This is my very first post on this forum. Excuse me for my bad english - i still learn this language

I'm looking for materials/ books/ ebooks/ standards/ tutorials/ arcitles about a high reliability electronics. I don't want to build a rocket or a car ABS system, I just have noticed that most books about electronics avoid this topic.

I'm looking for materials about:
- possible problems with particular types of components,
- PCB design and materials,
- what is important during uC selection for high reliability projects,
- software tips/tricks,
- etc.

Many thanks for your help!
Regards

[MosherIV]

There are 2 sides to reliable circuits :
1 well designed safe circuits
2 component reliability.

For 1 - this comes down to knowledge abiut electronics
2 is a bit easier to research, look up "mean time before failure" MTBF.
This a well know field

[bitslice]

- possible problems with particular types of components,
- PCB design and materials,

Look at how people test for failure, heat, humidity, temperature, vibration, radiation and load,
now think about how each component is adversely affected by each one and how the tolerances of each will begin to drift. Don't overlook things like connectors, usually the #1 failure
It's why Mil Spec components exist.

- software tips/tricks,

Code Complete, by Steve McConnell
I'm not sure if this is the book I was thinking of, but knowing the best practices for coding is a good start for reliable code.
Or you can prove code in a mathematical fashion if you have the time.

Knowing how code fails is useful, often it's passing a variable which makes no sense, hence the advantages of say ADA over BASIC

[tautech]

Welcome to the forum Tzolkin

A very good rule to offer high reliability with any design is to set yourself a % of the maximum rating of any part of product design to a % well below any parts maximum rating. Old EE's that I've talked to that worked in government agencies used 62% of ANY rating, beit electrical, mechanical, thermal, MTBF etc.

Modern design often uses in excess of 90% of maximum ratings and there is little wonder why much of what we buy does not last. We live in a throwaway society where the cheapest reliable product wins. 
Electronic design is an art, constantly balancing component stress against maximum ratings and much time is spent on it.

Many of the individual components are much more reliable these days, especially semiconductors but it makes no sense to torture any component capability for the sake of a few cents when reliability could be much improved and a brand then obtain a better reputation.

[barry14]

MTBF can be very indicative of the expected lifetime of a product or component. By definition, if a large number of components are in use, the MTBF (if it was calculated properly) will be close to the observed failure rate.  For many components, the MTBF is derived from experimental measurements on a large number of parts. However it is only a mean value: many components will fail sooner and, by the same token, many will last much longer.  In any case, the statement that no components will last the MTBF value is false.

[2N3055]

Bunch of books are available.. Many whitepapers and publications free on internet..
I attached one document that is very nice as overview, to give you a picture....

Enjoy..

[barry14]

I did say that the MTBF must be calculated properly.  I agree that it is unrealistic that a device containing a motor could have an MTBF of 1 million hours. However the use of accelerated testing has been used for many years and has yielded many good results.  In any case, I think it is still false to say that all MTBFs are unrealistic. Electronics can be very reliable if properly designed and used and can exceed its calculated MTBF.  I have a Tektronix Model 453 oscilloscope that is almost 50 years old and still works perfectly with all of the original parts. I have a Heathkit digital clock that I built in 1972 and have used continuously since then.  It has 22 transistors so their combined operating time is about 8 million hours.  In all that time, there has been only 1 failure.  None of the other components in the clock have failed. So MTBFs in the million hour range are not necessarily excessive. Look at the tens of millions of cars operating now, each with multiple processors and other electronics.  Their demonstrated MTBF is in the millions of hours; if not, consumers would be complaining. By the way I have several computers that are over 5 years old and their HDDs are still working.

[Circlotron]

Temperature affects things too. I have a 90mm Sunon 12 volt ball bearing fan that has been running 24/7 since November 2002 in an ambient temperature of 3 deg C in my refrigerator. That's 123,000 hours. We did some tests where I was working at the time at 70 deg C they would be lucky to last 1000 hours.

[EmmanuelFaure]

Guys, did you even pass Reliability 101? I don't think so. Classical reliability theory, with constant failure rate :

MTBF = When 63% of the components from the sample fails.

Remaining working components = exp(-t / MTBF). For example :
- 99.9% survivorship @ ~MTBF/1000
- 99% survivorship @ ~MTBF/100
- 90% survivorship @ ~MTBF/10

@ Tzolkin, a must-read : http://www.sre.org/pubs/Mil-Hdbk-338B.pdf

[arlipscomb]

While specific to hard drives, this may provide some insight to the question of the value of MTBF: http://knowledge.seagate.com/articles/en_US/FAQ/174791en?language=en_US

Good questions on the subject.

[ChunkyPastaSauce]

I'm looking for materials/ books/ ebooks/ standards/ tutorials/ arcitles about a high reliability electronics.

There are many standards/workmanship/guidelines docs available online from NASA.

[Mario Blunk]

Right, the failure rate (or the inverse, the MTBF) gets a real meaning in connection with time. Given the failure rate and the time that has passed, you get the reliability R. R says how many units still work after a given time. If it is about a single unit, then R stands for the probability that your unit works after lets say one year.
If you calculate the other way around, starting at a given R at a given time, you can calculate the total failure rate of the unit. Divide this figure by the number of devices and solder joints and you get an average failure rate per device. The reliability of critical parts in automotive braking systems or nuclear power plants ranges from 99.9 to 99.9999.... after lets say one year (about 10000h).

[Gyro]

Hi,

Welcome to the forum. It is worth indicating in your posts when you are bumping very old threads.

[Doctorandus_P]

You can look up specifications and design considerations for power supplies for medical use. These have lower leakage requirements, and are built in such a way that if they fail, they fail in a way that is not dangerous to the patient.

[SteveThackery]

2 is a bit easier to research, look up "mean time before failure" MTBF.

MTBF does NOT mean Mean Time Before Failure! It means Mean Time Between Failures, and is only relevant to repairable items. It's partner is MTTF: Mean Time To Failure, which is used for non-repairable items like individual components.

Both are statistical measures, and are the inverse of the failure rate when the device is in the bottom of its bathtub failure rate curve.

[SteveThackery]

MTBF can be very indicative of the expected lifetime of a product or component.

No. MTBF specifically avoids failures due to ageing mechanisms. It refers to the random failure rate during the flat part of the item's failure rate curve. It is quite common for an item's MTBF to exceed (sometimes greatly) the expected lifespan of the item.

By definition, if a large number of components are in use, the MTBF (if it was calculated properly) will be close to the observed failure rate. 

This is misleading. The MTBF is EXACTLY the inverse of the failure rate. That is how it is defined.

In any case, the statement that no components will last the MTBF value is false.

Correct. Statistically, 35% of a population of identical items will survive to their MTBF. 65% will fail before their MTBF. Both these figures exclude ageing effects.

[SteveThackery]

It is quite common for an item's MTBF to exceed (sometimes greatly) the expected lifespan of the item.   

For example, incandescent bulbs are exceptionally reliable, with an MTTF of 100 000 to 1 000 000 hours. However, their designed lifespan is only around 1000 hours.

LEDs have an MTTF of 5 to 10 million hours, but a lifespan of 100 000 hours if driven conservatively, down to 15 000 to 25 000 for typical domestic LED bulbs, and can be less than 10 000 hours if overdriven in the typical Chinese way.

[SteveThackery]

As explained above, a good way of improving reliability for a system is to build it with high quality (ie genuine) components from mainstream manufacturers, and ensure that all components are operated well within their SOAR (safe operating area).  You also need to consider the operating environment: that might include protection against water ingress, for example. Also, rarely, fire resistance might be important if the system has a vital safety role. The environment also includes the electrical environment; noisy and hostile environments like cars require the electronics to incorporate protection circuits on its inputs and outputs. This typically includes over-voltage protection and filtering.

Beyond that we need to consider measures to improve resilience. Redundancy is a popular and effective way to achieve that, but it is expensive and doubles the maintenance burden. Where redundancy is used, common mode failures may be a risk, so highly resilient systems will use redundancy in which there is diversity as well. That is, the main and standby systems are functionally identical but implemented differently - for example by two completely unrelated development teams working to the same specification.

Crucial to high reliability for complex systems - especially software systems - is high quality testing (a whole subject in itself).

Talking of testing, a redundant system is only effective if the fault detection works reliably. That is, failures in the main system need to be detected in order to initiate a changeover to the standby system. And this is more complicated than you might think: the standby system must itself be tested rigorously in order to ensure it is available should the main system fail. Not all faults in the main or the standby system will be detected by the fault detection system; this is known as "fault coverage" and obviously needs to be as high as possible. One way of managing this is to switch between the main and standby systems once every 24 hours (say). This exposes both systems to user workload, which is often the most thorough test regime, and if a fault is detected there is a very high probability that the changeover can switch back to a working system, bearing in mind it was working fine a few seconds ago.

When a failure does occur, good reliability does rely on a short downtime. That means having spares in stock ready for a rapid swap-out. Remember, if the main system fails and the standby system is now in use, you no longer have any redundancy until the failed system is repaired.

There is a lot more to be said, and feel free to ask more questions, but I expect you've got enough to think about for now!

EDIT: There is also the concept of software reliability. Software doesn't fail in the same way as hardware, nor is it repaired in the same way, but bugs can manifest as a failure in the system from the viewpoint of the user, so it does contribute to the overall performance and the 'perceived reliability' of the system.

[electron_plumber]

I'm looking for materials/ books/ ebooks/ standards/ tutorials/ arcitles about a high reliability electronics.

There are many standards/workmanship/guidelines docs available online from NASA.

+1 for NASA papers.

Try to understand concepts like "fault tolerance" and "failure mode effects analysis" (FMEA, or FMECA). I worked in aerospace and leveraged a lot of NASAs papers on these topics early in my career. I then took this knowledge with me to the automotive industry and had zero issues fitting into that design culture.

[SteveThackery]

I'm sorry for replying to an old thread (is that what they call necroposting?).

What's the protocol here - should we reply to dead threads or start new ones?

[Siwastaja]

I'm sorry for replying to an old thread (is that what they call necroposting?).

What's the protocol here - should we reply to dead threads or start new ones?

If you have something useful to add and feel like it - like you did here - then why not. If you feel like discussing about MTBF, I don't see what value it adds to table to make a new thread for it.

What is irritating is when people do not notice they are posting to ancient threads and are trying to specifically communicate with the OP (who has left the building years ago), asking clarifications or overall doing low-quality posts. This easily happens because only the first necroposter sees the big honking warning that they are necroposting; in this case, "Mario Blunk" chose to ignore it.