EEVblog #324 - DC-DC Converter Testing - USB Supply Part 4

[tom66]

Due to the instability of the 50 Hz oscillation, I'm thinking it's a coincidence it lines up with the mains frequency.

I had a 5V power supply (wall wart thing) that I thought was an iron transformer based design. I measured the output as rippling around 60 Hz under no load. But I thought, we use 50 Hz in this country, what's up? I put it under load and it increased to several kHz... Maybe some control loop in the constant power circuit has a 50 Hz bandwidth.

[mlopes]

Just one consideration about the MTBF:
A MTBF of 350,000 hours means the component will have a reliability (working without any failure) of just 37% in the first 350,000 hours. If you do the math considering a reliability of 95% (probability of working without any failure), this value is reduced to just 2 years! [ R(t) = e^(-t / MTBF) ].
Thank you very much for all your effort developing the blog.
You do a really nice job.

It's even a little bit more complex :-) Most people understand MTBF the wrong way. MTBF = MTTF (mean time to failure) + MTTR (mean time to repair). A MTBF of 1000h could mean that a device brakes just after 1 hour usage and the repair takes 999 hours on average. If we want to specify the average time a device works until it breaks we should use MTTF! MTBF also implies that a broken device can be repaired (not just swapping the broken device).

Just one consideration about the MTBF:
A MTBF of 350,000 hours means the component will have a reliability (working without any failure) of just 37% in the first 350,000 hours. If you do the math considering a reliability of 95% (probability of working without any failure), this value is reduced to just 2 years! [ R(t) = e^(-t / MTBF) ].
Thank you very much for all your effort developing the blog.
You do a really nice job.

It's even a little bit more complex :-) Most people understand MTBF the wrong way. MTBF = MTTF (mean time to failure) + MTTR (mean time to repair). A MTBF of 1000h could mean that a device brakes just after 1 hour usage and the repair takes 999 hours on average. If we want to specify the average time a device works until it breaks we should use MTTF! MTBF also implies that a broken device can be repaired (not just swapping the broken device).

In this case I am supposing the MTTR << MTTF, so I can consider MTBF approximately the MTTF. Only the MTBF is available in the manufacturer’s datasheet and it is not feasible for him to predict the MTTR (once he does not know the type of application and repair complexity).

[mikeselectricstuff]

I have a hard time believing that any MTBF figure has any real validity or usefulness - is it really possible to evaluate every conceivable failure mode and assign a realistic probability to it?

[EEVblog]

I have a hard time believing that any MTBF figure has any real validity or usefulness - is it really possible to evaluate every conceivable failure mode and assign a realistic probability to it?

It looks good on paper 

Dave.

[hlavac]

I have a bad feeling about the health of Dave's electronic load since the power supply incident in EEVBlog #315.
Dave you should check it out, the 50Hz mystery is not normal...
Or is it the scope sitting on top of it and picking something up from the load?

[sagdahl]

I was not aware that the uCurrent limits at 300mA. What is the cure for the burden voltage at 400mA, Dave?

Maybe a power supply with sense inputs where you connect the sense-wires after the current-meter?

A power-supply with a good built-in current meter?

An external shunt-resistor of 0,1 Ohm, creating an maximum voltage of 40 mV?

uCurrent V2.0 which can handle 1A?

I agree with Dave that this is really strange that Fluke and other quality instruments hasn't solved this drawback. Fluke-guys, come up with a solution!

[IanB]

I was not aware that the uCurrent limits at 300mA. What is the cure for the burden voltage at 400mA, Dave?

Unless you are really in need of accuracy just use the 10 A range on the meter. That's what everyone else does.

[EEVblog]

I was not aware that the uCurrent limits at 300mA. What is the cure for the burden voltage at 400mA, Dave?

I showed it. Use the 10A range. Or roll your own solution if needed.

Dave.

[BobC]

I'm glad Dave is an EE instead of, say, a veterinarian.

Imagine the fuss if he had done all this to a kitten...

[Rufus]

What is the cure for the burden voltage at 400mA

In this case turn up the supply voltage to compensate or connect the PSU sense inputs to do it automatically (or just read current from the power supply in the first place).

[SteveyG]

... connect the PSU sense inputs to do it automatically (or just read current from the power supply in the first place).

While I was watching this I was surprised that he had no PSU with separate sense terminals. 

[madires]

I have a hard time believing that any MTBF figure has any real validity or usefulness - is it really possible to evaluate every conceivable failure mode and assign a realistic probability to it?

It's more like guessing a probability based on your experience, knowledge and statistics you may have. Then you check the dependencies of all atomic elements of your component or device and calculate the total reliability. But that may end up being a quite complex task. I've done that for communication networks. If you promise in your SLA such and such a reliability for your network, you should be able to deliver that. A vendor sold ATM switches with a reliability of 99.9999% (could be even some 9s more :-). Common sense tells me that's BS but I had to use that number for the calculation because there were no other numbers. If something goes wrong you can blame the vendor for giving you wrong numbers. Job done :-)

[sagdahl]

You are of course right. There is no need for better resolution then the 10A range will give us.
Looking forward to next episode.

[mikeselectricstuff]

I have a hard time believing that any MTBF figure has any real validity or usefulness - is it really possible to evaluate every conceivable failure mode and assign a realistic probability to it?

It's more like guessing a probability based on your experience, knowledge and statistics you may have. Then you check the dependencies of all atomic elements of your component or device and calculate the total reliability. But that may end up being a quite complex task. I've done that for communication networks. If you promise in your SLA such and such a reliability for your network, you should be able to deliver that. A vendor sold ATM switches with a reliability of 99.9999% (could be even some 9s more :-). Common sense tells me that's BS but I had to use that number for the calculation because there were no other numbers. If something goes wrong you can blame the vendor for giving you wrong numbers. Job done :-)

And any failure to meet the MTBF can be blamed on insufficient sample size...!

[LaurenceW]

Dave, I too think your BK electronic load may itself be contributing to these 50Hz ripples at low current, and the step overshoots at switch on. I don't think it's mains pickup, but instability in the system. Simple test - can you stick a 12V battery (simple linear chemical device! no active electronics!) and maybe a series resistor of 10R across your electronic load, and measure the voltage on your scope across the BK when set to a current of 1 mA? Also, apply the 12V at a power limit of 1W, and see what voltage it loads the battery down to, when you abruptly switch the load on?.

An interesting video, all the same. Let's see you testing that 1KV isolation voltage!

[EEVblog]

(or just read current from the power supply in the first place).

Not generally a good idea, as the current can change depending upon the actual supply voltage to the chip.
In this case when you are calculating power, you need to know the current and voltage, or you'll end up with significant error.

Dave.

[rbk17c]

About the *evil* shorting of output: Others (ReCom) specify a max 10s shot, else-wise get the ~10% more expensive '/P' model, witch has 'Short Circuit Protection'.

Oh, and you can now get 3W in a (smaller than Dave's) SIP4 package like MuRata's MEE3S0515SC.
While this device also has a 1kV isolation, note that this is tested for only 1s, and that a continual 1kV will kill the device.
Do you think this is a common limitation? Me too.
/holger

[RJSC]

Oh well...
I was hoping you'd be using an Adjustable DC-DC converter as a tracking pre-regulator so that the supply would go over 5V...
Will you consider it?

Someone can argue we could use a fixed 20V converter, but than we'd be wasting a large amount of power at the lower voltages and limited to 100 mA current mo matter the voltage.

[EEVblog]

Oh well...
I was hoping you'd be using an Adjustable DC-DC converter as a tracking pre-regulator so that the supply would go over 5V...

That's exactly what I'm doing. This is just for the isolation.

Dave.