Dave started out wanting to investigate the power-on spike on the Rigol DP832 Lab Power Supply, but ended up hunting down a reset bug that uncovered a bad thermal design mistake in the supply.
Watch Dave hunt it down step by step, and almost get duped by some marginal oscilloscope triggering.
How will Rigol respond to this?
If they can make such a fundamental mistake as dissipating 5W in a small package you have to wonder about the quality of the rest of the design...
I wonder if maybe it was a feature-creep issue, where the reg originally supplied a lower current then someone added other stuff without telling the PSU guy.
Do you have a variac? I so, try winding up the mains voltage to the max rating and see how hot the reg gets.
(if not, use a mains transformer as an autotransformer with LV secondary in series with mains to boost it up)
The obvious easy fix is to replace the reg with a switcher - there are plenty of drop-in 7805 replacements.
If they can make such a fundamental mistake as dissipating 5W in a small package you have to wonder about the quality of the rest of the design...
And the rest of their total product portfolio. I was a Rigol fan but watching this makes me think twice next time
It might have originally been a switcher, but given how unusually sensitive the logic was (rebooting from such small ripple) maybe they swapped it out for the LM, and never upped the heat sink accordingly.
Hi Dave,
The 317 dissipation looks crazy, but the increased ripple does not look that critical to me to justify a reset on the SoC.
On the other side, I had some bad experiences in the past with the i.MX28, caused by the poor efficiency of the integrated power supply that leaded to a really high operating temperature of the whole SoC. My suspect would be that the front panel would suffer from the reduced airflow when operating without the chassis, the i.MX28 overheats and reset and the temporarily increased 5V ripple would be just a consequence of the integrated OS booting and initializing stuff.
Are you still able to check the temperature of that? The back of the BGA may be accessible enough to measure.
Thanks for the video!
Fabio
Hi Dave,
The 317 dissipation looks crazy, but the increased ripple does not look that critical to me to justify a reset on the SoC.
Perhaps. But I cannot find anything else obvious to explain it. I'll leave that up to Rigol.
My suspect would be that the front panel would suffer from the reduced airflow when operating without the chassis, the i.MX28 overheats and reset and the temporarily increased 5V ripple would be just a consequence of the integrated OS booting and initializing stuff.
That would be an amazing coincidence that two parts are running super hot and overheating.
Also would not explain why pushing external air over the LM317 fixed the reset problem.
Also would not explain why pushing external air over the LM317 fixed the reset problem.
Well... it's not that your bad-ass fume extractor was sharpshooting just at the regulator... :-)
Well... it's not that your bad-ass fume extractor was sharpshooting just at the regulator... :-)
Yeah, but it was going nowhere near the front panel.
If Rigol wanted to be as cheap as possible then, as a hotfix, they could probably send all of the current users one of those 20W 8R power resistors that they use for testing audio systems. Just stick that sucker inline between the +12V output and the LM317's input and it would probably dissipate enough heat to solve the problem.
Perhaps. But I cannot find anything else obvious to explain it. I'll leave that up to Rigol.
I'm curious about Rigol's response. Sometimes shit happens, maybe first the CPU board needed much less power and then they enhanced it and forgot to check the supply. But the manufacturer should communicate it clearly and provide fixes for it (maybe a repair kit with a bigger heatsink, as you suggested) or do a product recall.
Agilent is a good example for very good customer support: Last time when I reported a trigger problem in auto-mode for a signal on my Agilent DSO, they reproduced the signal (which was not very difficult, just a digital burst sequence), confirmed the problem and fixed it with the next firmware release.
Do you have a variac? I so, try winding up the mains voltage to the max rating and see how hot the reg gets.
I'm already not far off the max mains voltage here in the lab I think.
I'd have to do that for all mains voltage selection scenarios to be thorough.
It's not going to make a huge difference in dissipation anyway. Enough to cause units to fail?, maybe. But in the end there are many variables at play here, so I'll leave that up to Rigol to work out and fix.
Without looking in detail, my guess is a lot of that power is going to the LCD backlight, which could just as easily have been run from the unregulated 12V supply (assuming they actually have a proper backlight driver and not just some Rs from 5V....!)
..and I suppose there is a possibility that they have mis-configured the CPU to run in a much more power-hungry mode then intended - maybe they'll issue a firmware 'fix' that slows it down to a crawl so the UI then sucks ass...
they could probably send all of the current users one of those 20W 8R power resistors that they use for testing audio systems. Just stick that sucker inline between the +12V output and the LM317's input and it would probably dissipate enough heat to solve the problem.
The heat still needs to go somewhere - The R would need decent mounting where there's airflow. 5W is still 5W, whatever device you burn it in - it's mostly about surface area.
I'm already not far off the max mains voltage here in the lab I think.
I'd have to do that for all mains voltage selection scenarios to be thorough.
It's not going to make a huge difference in dissipation anyway. Enough to cause units to fail?, maybe. But in the end there are many variables at play here, so I'll leave that up to Rigol to work out and fix.
What does the mains voltage have to do with the 12-to-5 volt regulator? If a lower mains voltage means lower temperatures somehow, does that mean that a US (120V) system will be less likely to fail than your Aussie one?
You could try feeding the 5V from another lab power supply to see if it's really the issue.
If it works, then maybe just put one of the cheap 5$ switcher between the bridge output and the LM317 input (or bypass the lm317 all together), this should allow to cut the dissipated power by more than half, hopefully bringing it to a more reasonable temp.
Cheers,
Sylvain
How will Rigol respond to this?
i bet they will tell you "Dave, this is high-end precission power supply"
Sorry, but i couldn't resist
Without looking in detail, my guess is a lot of that power is going to the LCD backlight, which could just as easily have been run from the unregulated 12V supply (assuming they actually have a proper backlight driver and not just some Rs from 5V....!)
..and I suppose there is a possibility that they have mis-configured the CPU to run in a much more power-hungry mode then intended - maybe they'll issue a firmware 'fix' that slows it down to a crawl so the UI then sucks ass...
they could probably send all of the current users one of those 20W 8R power resistors that they use for testing audio systems. Just stick that sucker inline between the +12V output and the LM317's input and it would probably dissipate enough heat to solve the problem.
The heat still needs to go somewhere - The R would need decent mounting where there's airflow. 5W is still 5W, whatever device you burn it in - it's mostly about surface area.
Yep. I'm envisioning a wiring harness that allows the resistor to be placed back by the fan, perhaps. Also, I didn't really run the numbers so far as size goes. It might need to be more like 4R. Anyway, just thinking out loud.
What does the mains voltage have to do with the 12-to-5 volt regulator? If a lower mains voltage means lower temperatures somehow, does that mean that a US (120V) system will be less likely to fail than your Aussie one?
This is a linear supply, not a switcher with PFC. So if you increase the mains voltage by 10%, from 220 V to 242 V, all the secondary voltages will also go up by 10%. So dissipation will go up by ~5%. Not really something that will make much of a difference, but it might just be enough to tip the scales and have the unit reset with the case fully assembled. For 120 V the taps would be configured differently, so you wouldn't see only 6 V secondary voltage with 120 V AC in
.
PS: Nice video, Dave! I liked the combination of troubleshooting, review and tutorial.
Dave,
If you intend to investigate further, then one idea may be to stick a small, calibrated SMD thermistor on the tab of the 80N15 pass device for one of the 30V rails. Close the whole PSU fully up, load down all the rails, and measure the tab temperature.
At your mains voltage and 22V @ 3A output, just before the first tap switches, the 80N15 needs to dissipate 72W. With Rth,JC at 0.35 K/W you need the HS temp at 100oC or less, with no margin for error. You casually measured about 70oC or so at 48W dissipation (30V @ 3A load), so the large HS may seem slightly small as well. Particularly so if you load down all three rails, causing the transformer to 'pre-heat' the cooling air a bit before it reaches the large heat sinks.
PS: According to the IXYS datasheet I could quickly find, then the 80N15 MOSFET is housed in a TO-264 case, not a TO-220 as we see in the video.
PPS: The over temperature protection numbers I have seen, suggests the device in question usually shuts down when the die reaches roughly 175oC.
----
Edit. The IXYS 80N15 has a TJ,max of 150oC. Not sure I'd like to run it at that, but...
What does the mains voltage have to do with the 12-to-5 volt regulator? If a lower mains voltage means lower temperatures somehow, does that mean that a US (120V) system will be less likely to fail than your Aussie one?
No because of the tabs on the transformer it would still give out 12 volts no mater if it intakes 120V or 230V.
What he meant is that the mains voltage can go up and down around 5% or more depending on your area electric network.
Therefore the output voltage on the transformer can also be higher than 12V and that adds to the dissipated power altough not much I wager.
If it was MY power supply, the first thing I'd do is replace the LM317 with an LM1085. It has a theta J-C of 0.7 degC/W, a saving of over 20 degrees.
Next, I would stick a BGA heat sink on the front, such as an AAVID 375424B00034G.
My judgment might be colored by the fact that I have those parts in stock.
Kerry
It's a real shame, I was seriously going to buy one of these. But now, I've been put off.
So how many other Rigol units have a similar design where they are supplying 5v using a LDO at near 5Watt's? You would think that a company like Rigol would design reuse across the board and thus this might not be a one off.
Does it mean that their Oscilloscopes could potentially be doing the same? Makes you wonder.
It's a shame you didn't bother to just replace the small heatsink with a larger one. There is room for one both sideways and vertically, at least as far as I can see.
The heatsink in the picture below is a 7.6 C/w one and as you can see, it would accept a TO-220 just fine and would probably fit there.
Yeah, it would still be 5 watts dissipated there, and those capacitors would slowly cook there... but at least it would decrease the chance of resetting.
(picture was made for someone else to show the rectifier, but it just happens the heatsink is in the picture so ...
So how many other Rigol units have a similar design where they are supplying 5v using a LDO at near 5Watt's? You would think that a company like Rigol would design reuse across the board and thus this might not be a one off.
Scopes will have a switching power supply, so they're unlikely to use an LM317 to create a 5 V rail from 12 V. The Rigol DMMs could share this design, but they probably draw less power on the grounded 5 V rail, since the display is smaller. The LM317 is not generally regarded as low dropout, by the way.
I would definitely put any orders on hold until Rigol has come up with a solution and distributed this fix to their distributors. I wouldn't be surprised if they would indeed come up with a firmware bodge job like CPU power saving or fan speed.
I wonder how many DP832s are running in racks. Do people already trust Rigol for heavy-duty system applications? I would probably trust the scopes, but most of their other products are too new to be used for important jobs, in my opinion. Fine for design work where you just grab another supply if one dies (unless it sends out a spike and kills your DUT, but that would never happen
), but not in an application where your production line grinds to a halt if the thing fails.
The heatsink in the picture below is a 7.6 C/w one and as you can see, it would accept a TO-220 just fine, and would go around the large capacitors and the rectifier on the left side of the LM317.
That's still not very nice for those caps in my opinion. They may be 105°C rated, but they will only survive a few thousand hours at that temperature, so even 60°C is not very nice.
Hi Dave,
The 317 dissipation looks crazy, but the increased ripple does not look that critical to me to justify a reset on the SoC.
Perhaps. But I cannot find anything else obvious to explain it. I'll leave that up to Rigol.
one thing i noticed from watching the video again : the metal tab on that lm317 is very thin .... much thinner than normal ... fake lm317 ?