EEVblog #1036 - PSU Fire PCB Repair

[EEVblog]

PCB Repair of the RD Tech DPS5020 Power supply module after the flaming capacitor failure.

[Faith]

Thanks for the follow-up!~ Was originally going to get one of these; but now I'm not so sure.

Mistakes happen but what made me was when RD Tech admitted that this isn't the first time they've encountered precisely this method of failure but yet never bothered to correct the issue until now.

Oops.

[bjcuizon]

Sorry but this is just regarding the video image-wise .
IMO The video quality is great and all shots look nice, with only the important bits in focus. 
..And I think the lighting is a bit different. It looks brighter and things look clear.

[EEVblog]

Sorry but this is just regarding the video image-wise .
IMO The video quality is great and all shots look nice, with only the important bits in focus. 

I personally don't like that, and usually deliberately set the depth of field on my camera to avoid any bokeh effect.

..And I think the lighting is a bit different. It looks brighter and things look clear.

Lighting is the same, just a different camera and lens. The early shots on the video were actually shot with a macro lens.

The audio sounded a bit more echoey to me as I'm basically behind a shotgun mic instead of on top of the top mounted mic like on my Canon.

[Muttley Snickers]

These videos were excellent with drama, suspense and a happy ending.   

If one of these capacitors can cause this much damage and potentially become a fire hazard by going short across a high current output then wouldn't it be a better idea to add these filter capacitors by way of dedicated smaller sacrificial tracks, the only down fall I see in doing the circuit this way might be a bit more noise and the risk that these tracks could blow out or go open without giving any feedback as the main output terminals would remain intact.   

[trophosphere]

Some ceramic capacitor manufacturers have MLCC SMD offerings which are much more resistant to PCB flexing and have a failure mode with a preference for failing open rather than being a short. Examples being AVX and Kemet.

[EEVblog]

Some ceramic capacitor manufacturers have MLCC SMD offerings which are much more resistant to PCB flexing and have a failure mode with a preference for failing open rather than being a short. Examples being AVX and Kemet.

Yeah, forgot about those. Would have made a nice addition to the video

[tszaboo]

OK, he stopped selling it, good.
But: Placing capacitors in high stress area?
Placing fuses between connectors? You know, something that you actually have to access to replace.
And placing fuses in parallel?? Seriously?
Placing components before the fuse?
How many death traps are there left in this circuit?

A working circuit and a proper circuit is two very different things.

[Brumby]

Thank you for the follow up.

It is good to see the issue is a straightforward one, the repair simple (even if a little inelegant) and the result of success.

Even better that corrective action is being taken from the manufacturing end.


These modules are back on my "of interest" list.

[firewalker]

I have seem high rating fuses (car type) paralleled in many good quality UPSs. Something like 3x30A fuses.

Alexander.

[Psi]

One way to fix the design with minimal changes might be some routed slots around the cap so board flex doesn't result in much cap flex.

[Brumby]

I'd move the cap, myself.

[tszaboo]

I have seem high rating fuses (car type) paralleled in many good quality UPSs. Something like 3x30A fuses.

Alexander.

If you are using a bigger, not SMD fuse, the story is different.
For a blade type fuse, the connection and the trace leading to it is massive, and it has small resistance, while the fusing element has a big resistance. For example it is almost 3.5 mOhm for Littlefuse ATOF.

If you are using a tiny SMD fuse, the connection the soldering and everything else has a big portion of the total resistance. You need to route high current through a 1206 component. The Littlefuse 501 has less than 2 mOhms resistance. It will dissipate 1W at rated current.
If you connect two SMD fuses in parallel, you get a fuse, which has "whatever, probably higher than just one" rating both for voltage and for current.

And this made me realize, I never saw and SMD fuse (chip type, not glass type) with 50+ DC voltage rating and high current.

[MisterDiodes]

If you are using a bigger, not SMD fuse, the story is different.
For a blade type fuse, the connection and the trace leading to it is massive, and it has small resistance, while the fusing element has a big resistance. For example it is almost 3.5 mOhm for Littlefuse ATOF.

If you are using a tiny SMD fuse, the connection the soldering and everything else has a big portion of the total resistance. You need to route high current through a 1206 component. The Littlefuse 501 has less than 2 mOhms resistance. It will dissipate 1W at rated current.
If you connect two SMD fuses in parallel, you get a fuse, which has "whatever, probably higher than just one" rating both for voltage and for current.

And this made me realize, I never saw and SMD fuse (chip type, not glass type) with 50+ DC voltage rating and high current.

Thank you.  Parallel SMD fuses are not good design practice, and generally they won't be rated for high power break / higher voltage.  A real fuse of the correct rating is called for here.

Also the advice to use two SMT MLCC caps in series is a bit bogus - that might delay the fire, does not eliminate the problem.  Even if you use different manufacturers for each cap it is just a race to see which one fails first in the series group.  Those caps that supposedly fail "open" can be just as bad we've found in our tests - and if they do fail "open" that can cause other problems on the board.  So we try to avoid all SMT MLCC caps on a high power rail in the first place, use good quality TH caps if possible, or else plan a proper reliable circuit protection method if you have to use SMT ceramics on a high available power rail.

I love how the manufacturer said it happened before, but they never bothered to fix it until it shows up on EEVblog! 

Big Thanks to Dave for the follow up video - that made the whole story even more interesting.  It just goes to show if you pay sh*t for a power supply, you'll probably get sh*t.  This one worked like sh*t and even smells like sh*t now.  There is a pattern there...

[IanMacdonald]

The actual safety margin on fuses varies quite a bit between types.  If the two fuses in parallel have a safety margin of 1.5x (typical for mains fuses) then with two you could be looking at as much as 60A to blow them in a reasonable time, which is way too much.

The other option would be to use a polyswitch. They also have a safety margin before they open, but at least you could use a single 20A rated one without that being such a nuisance if it does trip occasionally. The best answer  for this sort of current though is an MCB, space permitting.

Also I cringe at leaving burned board in place. Better to cut the whole lot out and either rebuild it with glassfibre and hardwiring, or else mount the components elsewhere.

[ChrisLX200]

Given the modest cost of these things I would not have attempted a repair anyway and just binned it. Plus - the manufacturer should send you a new modified one FOC (at least). How many of these things are out there owned by folk who don't watch EEVBlog on YouTube or read this forum? Has a recall been issued for this potentially hazardous device, can buyers get a refund if they can prove they bought one? The manufacturer has admitted fault so they don't really have a leg to stand on.

[Freelander]

that made the whole story even more interesting.  It just goes to show if you pay sh*t for a power supply, you'll probably get sh*t.  This one worked like sh*t and even smells like sh*t now.  There is a pattern there...

  What a happy soul.. ... The only pattern seems to be the sh*t that you are writing there. 
 

[gardner]

Dave mentioned torsional force on the screws as a contributing factor.  I would say that it's more likely flexing of the board due to downward force of the screwdriver.  The board is quite rigid torsionally, but thin and flexy the other, normal, direction, and it is supported off to the sides, not directly underneath the screw terminals.  Pressing down with the screwdriver puts the MLCC under tension.

[Cerebus]

I'm feeling very smug, I called it right (low resistance through the charring)  in the thread that discussed the previous video. 

[Freelander]

I'm feeling very smug, I called it right (low resistance through the charring)  in the thread that discussed the previous video. 

nice one 

Note also the unit running at a measly 100W and the tiny heatsink fan isn't even running. These are superbly efficient little units.  The cap issue is only on the 5020 units and has already been corrected. (it is not fitted to the smaller units). AFAIK there was just one previous instance of an issue in the factory that was put down to a faulty cap (live n learn). The clickbait flames video leader was a tad OTT and not really fair but I am sure the clicky baity part will and has increase(d) sales even more as people see the durability and shear value for money and performance of these little gems. A huge untapped market there and nothing even close competing. I bet there are a a few people wishing they had come close to designing and producing something even remotely as good and with such massive sales potential.
A little gold mine.
I have mounted the cap from my 5020c on a little plinth with a painted mushroom cloud over it...............

[Mr. Scram]

Thanks for the follow-up!~ Was originally going to get one of these; but now I'm not so sure.

Mistakes happen but what made me was when RD Tech admitted that this isn't the first time they've encountered precisely this method of failure but yet never bothered to correct the issue until now.

Oops.

It took three Boeing 737s going down with the same problem before they picked up on the rudder issue that caused them.

[jnissen]

As for the case fan being noisy... I also purchased a box very similar to the one in the video. The fan operates off a 5V LDO on the input board. I reduced the voltage that is generated by the small LDO so the fan only operates at 4V. The result is the fan is slower and not as noisy. It still moves a lot of air so I'm not concerned it's a bad thing.

Replace the 3K ohm SMT resistor (labeled 302) with a 2.2K ohm 603 resistor. The voltage to the fan is calculated by:  Fan Voltage = 1.25V (1+R2/R1) 

The R2 is the 3K ohm while the R1 is a 1K ohm on my board.  With the factory values 1.25*(1+3) gives you 5V. So to get 4V just use a 2.2K ohm resistor!

New Fan = 1.25 (1+2.2K/1K) or 4V

You can try lower values but I found my fan needs about 4V to properly startup. Too low and the fan will cog once and not spin.

[Someone]

So we try to avoid all SMT MLCC caps on a high power rail in the first place, use good quality TH caps if possible, or else plan a proper reliable circuit protection method if you have to use SMT ceramics on a high available power rail.

Its not a problem exclusive to "high power" rails, a supply rated for just a few watts is capable of violently destroying MLCC caps as the heating is extremely localised.

[Brumby]

Given the modest cost of these things I would not have attempted a repair anyway and just binned it.

A lot of people looking at a low cost solution like these modules is not going to be so cavalier.  They will see a problem like this and want to repair the module, if at all reasonably possible.

By having one of these units "blow up", Dave has done the manufacturer a favour.  He has caught the attention of a lot of people who view such cheap modules with suspicion and has activated their "AHA!!" button.

By then successfully pursuing the repair, he has demonstrated the basic robustness of the design - other than the poor placement of a single MLCC cap (which only occurs on the 50V 20A version).  The response from the designer/manufacturer has only added to the positive vibe about them.

The repair itself is another plus.  It was very simple (for this fault) and could be done by anyone with a modicum of skill.


All these things add up to a more positive view of these modules.  Sure, something went wrong - but by having something go wrong, we have been able to see the Achilles' heel we so often suspect is lurking in low cost gear - and we can fix it.  Better still, we can prevent it.

Tell me you don't have a more positive attitude about these modules because of all this.....

[Tek_TDS220]

The printed circuit board shouldn't burn (and continue to burn).  Whoever made the board is cutting corners on flame retardants.  My guess is that the board would not pass the tests necessary to import them into the US and the EU.  I don't know about Australia.