EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: Pack34 on June 23, 2016, 03:49:39 pm
-
I've been asked to get some rough and general MTBF numbers for a circuit board. How exactly would I go about doing this?
I would assume I would take some reliability information from critical parts of the board, add in expected operating temperature, and some assembly information from the board house.
Is there an off-the-cuff formula to do this? I haven't been able to find any reliability information on the datasheets for components. I would assume I would send some emails to vendors or manufacturers requesting the info. What exactly would I request?
-
You make 1000 circuit boards, get them all up running and wait until they've all failed. Then do the statistics O0
-
I've been asked to get some rough and general MTBF numbers for a circuit board. How exactly would I go about doing this?
I would assume I would take some reliability information from critical parts of the board, add in expected operating temperature, and some assembly information from the board house.
Is there an off-the-cuff formula to do this? I haven't been able to find any reliability information on the datasheets for components. I would assume I would send some emails to vendors or manufacturers requesting the info. What exactly would I request?
There is no off-the-cuff formula, per se, and any MTBF more accurate than a WAG requires a huge effort and some environmental (specifically, accelerated life) testing.
Sanjaya Maniktala does a decent job of covering MTBF estimation of power supplies in one of his books, Switching Power Supplies A to Z.
-
take the lowest MTBF Rating of your components.
-
Some light reading, MIL-HDBK-217F "Reliability Prediction of Electronic Equipment".
Here's a summary of what people do from a software company:
http://www.reliasoft.com/newsletter/v9i1/prediction_methods.htm (http://www.reliasoft.com/newsletter/v9i1/prediction_methods.htm)
-
Thanks!
The big question, is where can I get general MTBF numbers for the individual components? I've seen numbers in calculations such as: 2,290,000,000 hrs for an RS232 transceiver and 98,039,215 hrs for a diode. Is this information readily available somewhere? If so, where do I get it?
-
You do high temperature accelerated aging testing. Calculate the MTBF from that test. When someone ask if this is really the correct method to do it, just nod. If they really question your method, then send links to articles. If, after that they dont leave you alone, go to a soundproof meeting room, tell them they are right, but please, leave you alone with your made up numbers, becuase there isnt a good method to make good numbers in short time.
-
I was afraid of that. I was asked to get some rough numbers for a circuit board with two days notice and no boards to test.
-
I was afraid of that. I was asked to get some rough numbers for a circuit board with two days notice and no boards to test.
(https://bianalystblog.files.wordpress.com/2011/06/media_httpdilbertcomd_dvcpc-scaled1000.gif)
-
You bring up a nightmare topic. For military and automotive we would use 217. Ground mobile for automotive. When I started out, I wrote my own program using IBMs BASIC compiler that had the equations in it and would create a printout. Trivial compared to collecting the data. Later I used a commercial program form Powertronic Systems which was still based off 217. Automotive decided the numbers from 217 were not realistic for that application. A group was formed and they created PREL. That would have been late 90's early 2000. There should be some info on SAE about it.
If you put a resistor in a desktop home radio, it may live longer than under the hood of your car. If you have a 0.5W resistor dissipating 0.01W it may live longer than the same part running at 0.48W. If you have a resistor rated for 50V and you run it at 1V, it may live longer than at 40V. Resistor package, type... all go into this.. There's your one resistor. A typical report would be several hundred pages. Some of the places would review line by line....
I have no idea what best practices are today..
Some light reading, MIL-HDBK-217F "Reliability Prediction of Electronic Equipment".
-
You hire a consultant. Give him the BOM of your board and the stress levels of each component. He plugs that into his program and gives you a number.
-
Using milspec parts you can estimate purely by math, a MTBF, since the parts are rigidly made to standards and have known predicted life spans:
http://www.sohar.com/reliability-software/free_mtbf.html (http://www.sohar.com/reliability-software/free_mtbf.html)
Simple version based on actual runtime data:
http://www.rfwireless-world.com/calculators/MTBF-MTTF-Calculator.html (http://www.rfwireless-world.com/calculators/MTBF-MTTF-Calculator.html)
Thanks!
The big question, is where can I get general MTBF numbers for the individual components? I've seen numbers in calculations such as: 2,290,000,000 hrs for an RS232 transceiver and 98,039,215 hrs for a diode. Is this information readily available somewhere? If so, where do I get it?
-
For quick and dirty analysis the MIL 217 approach is in my opinion the best game in town. And may be slightly better than throwing darts at a board. It has the benefit that it is a known and fairly widely accepted methodology. And has the drawback that it's tables of data have little bearing on modern electronics. You can sometimes patch the holes with manufacturers data, but more often you are just hiding your guesses behind layers of analysis. Hiring an expert may give you an answer that is slightly more likely to reflect reality, but the real value of hiring the expert is the artificial credibility that it gives the answer. The expert will also have a much larger library of existing data/guesses already at hand.
You won't know the real MTBF of your board until it has been in the field long enough to start seeing failures. Then the differences between your model and the real environment (which includes the actual components used, the assembler used, the real environment and use pattern they are exposed to and the ability and interest of the user in detecting and reporting failure.) will show up. It is not unusual to find discrepancies of several orders of magnitude.
The most famous example of these discrepancies is the US Space Shuttle. Many, many thousands of hours of analysis and component level testing were used to calculate a system failure rate causing death to astronauts was less than one in 100,000 while the actual number was closer to one in 50. The analysis was actually done twice, once during program development, and again after the first shuttle failure.
-
its all nearly useless because of electrical overstress caused by ESD, EMI, etc. How on earth can you calculate a MTBF if some some subsystem or w/e starts spewing damaging transients, slowly errording transistors?
the best way is to make something, and document repairs (and your documentation will suck balls unless you police it heavily).
you might get some general data like "communications failure". is it because your protection system is inadequate and not clamping enough? thermal aging? Your repair people will find a comm chip that went bad and write down comm fail. Then you will be scratching your head as to why it happened. Or you can heavily scrutinize each return and have managers yelling about backlogs (because who wants to hire a big repair department? it looks bad to visitors). The electronics layman will say something like "oh it was not designed right your product sucks" and not realize that things relating to electrical protection etc require individual attention and FEEDBACK to implement correctly*. You bascially need an entire department dedicated to the job cooperating with the design engineers. The amount of data you get might also be so small that things like employee laziness might totally skew your data. Not to mention that useful data takes years to collect and that NO ONE will want to do anything about it anyway :palm:
There is no cost effective one size fits all solution for electrical protection and you will always have people that find problems due to weird electrical things that happen in their facility (take the case of the mini fridge that turned into a 6GHz radio jammer!!!!!!!!!!!! (google it). Imagine the PCB you made started taking a hammering from 6GHz HV! )
Every "non robust" solution in your PCB is there only because of cheap design and lack of engineering time/skill.
-
You hire a consultant. Give him the BOM of your board and the stress levels of each component. He plugs that into his program and gives you a number.
and have a report thats 90% useless bullshit and 10% weird situational things he saw once in his life that costed you 250$ an hour to generate. This is literally like hiring a advisor to help you bet on horses (this being someone on your design team wanting someone to reassure them that the DCDC converter running with no filter through a 1 amp linear regulator won't fail at 850mA (ignoring inrush currents) because they are too cheap to add $1.50 to the BOM to get a 2 amp linear regulator LOL :-DD |O
Now, if you are doing something that is near theoretical limits (deep horizons probe, super sonic low altitude missile (SLAM, google it) ) then sure, it makes sense to hire some expensive advisors to think of stuff. Otherwise you are just looking for someone to reassure your cheapness.
What is a consultant gonna tell you about stress levels in some weird semiconductor part you choose because it was cheap that has a 2 page datasheet (where linear technologies would have 20, including test methodologies), with stuff going on unique to that manufacturing process (lets say, depending on how the bond wire was attached to the die)? Not to mention to do their job they will need advanced knowledge of your system behavior (i.e. code behavior, so they understand power cycling over time (that you might not want competitors to know about).
I have seen consultants/tradespeople in my job that come in and bullshit with employees for two hours lol! :palm: they act like they get a salary (while they rob the company by the hour) (and this is not guys that you want to have a strong professional relationship with, like customers or vendors). It's good that this interaction occurs to a degree for company culture, but often excessive.
another problem is that the jobs they perform take wayyyyyyy longer then estimated, so you get management freaking out about how long they will be present for due to cost. why not have your employees learn something and add value to the comapny?