okashira
10 kW
We already have the avalanche taken care of... Ive documented it pretty well
A $3 diode and $2.5 TVS diode takes care of it
A $3 diode and $2.5 TVS diode takes care of it
JP spot-welder, FET-switched, timed adj. pulse
- Thread starter okashira
- Start date
atomek1000 said:Hey guys i've got a quick question. Did anybody had an issue with blown resistors going to mosfet gate leg? I have the 1-gen riba spotwelder with 6x AUIRF1324 and they seem to be ok but i heard a big puff and found out that 2 of this 820 resistors are blown up.
What are these resistors for? Should i just replace them and go on or do you recommend modyfing/checking something before welding again? I heard about diode mod but i have regular car battery about 700A and haven't had any problems for 1 year so i don't think this was issue of fly back current or was it?
...
Gate resistors don't fail unless the MOSFET has failed, so if one or more of the resistors popped then so did their respective MOSFET(s).
And yes, it is entirely possible for the MOSFETs to survive a number of avalanche cycles before failing. The relevant graph from the AUIRF1234 datasheet is attached with highlighting in red of the applicable values. Note that the recommended peak drain current is 195A per MOSFET so 6 in parallel would only be good for about 1200A peak which I would think is insufficient for this application unless the lead-acid battery is tired and/or fairly small (say, under 40Ah of capacity). [EDIT] - also note that the repetitive avalanche energy is 175mJ per pulse so 1.05J total assuming equal division of avalanche energy (which is a poor/likely fatal assumption). If you plug this energy rating into the equation I=(E/(0.5*L))^0.5 you get an allowed peak current of 1024A to stay under the total avalanche rating (though, again, one should not assume that paralleled MOSFETs will share avalanche energy evenly - in fact, they are all but guaranteed not to!)
Also, 820 ohms is way too high for a gate resistor and while higher gate resistor values do slow down the rate of rise in drain voltage during turn-off, that doesn't actually reduce the energy during avalanche as the latter is strictly a function of peak current and cable inductance (0.5LI²).
atomek1000
100 W
Tesseract said:atomek1000 said:Hey guys i've got a quick question. Did anybody had an issue with blown resistors going to mosfet gate leg? I have the 1-gen riba spotwelder with 6x AUIRF1324 and they seem to be ok but i heard a big puff and found out that 2 of this 820 resistors are blown up.
What are these resistors for? Should i just replace them and go on or do you recommend modyfing/checking something before welding again? I heard about diode mod but i have regular car battery about 700A and haven't had any problems for 1 year so i don't think this was issue of fly back current or was it?
...
Gate resistors don't fail unless the MOSFET has failed, so if one or more of the resistors popped then so did their respective MOSFET(s).
And yes, it is entirely possible for the MOSFETs to survive a number of avalanche cycles before failing. The relevant graph from the AUIRF1234 datasheet is attached with highlighting in red of the applicable values. Note that the recommended peak drain current is 195A per MOSFET so 6 in parallel would only be good for about 1200A peak which I would think is insufficient for this application unless the lead-acid battery is tired and/or fairly small (say, under 40Ah of capacity). [EDIT] - also note that the repetitive avalanche energy is 175mJ per pulse so 1.05J total assuming equal division of avalanche energy (which is a poor/likely fatal assumption). If you plug this energy rating into the equation I=(E/(0.5*L))^0.5 you get an allowed peak current of 1024A to stay under the total avalanche rating (though, again, one should not assume that paralleled MOSFETs will share avalanche energy evenly - in fact, they are all but guaranteed not to!)
Also, 820 ohms is way too high for a gate resistor and while higher gate resistor values do slow down the rate of rise in drain voltage during turn-off, that doesn't actually reduce the energy during avalanche as the latter is strictly a function of peak current and cable inductance (0.5LI²).
Thanks for very informative answer
, the mosfets in this configuration handled about 35 000 weld pulses but recently i started to weld while charging lead-acid battery and voltage was little higher than always, maybe that's what lead to failure.Gate resistors are 82 ohms (820 marking on them).
pguk
100 W
okashira said:
I updated my post on previous page with a scope shot of TVS diode added along with my homebrew freewheeler.
Which Schottky have you used at $3?
Tesseract said:
So this is a new idea. If I were to add a decoupling capacitor for the freewheeling diode, do you mean I should connect it in parrallel with the schottky? Or else from drains to GND?
atomek1000 said:Thanks for very informative answer
Gate resistors are 82 ohms (820 marking on them).
35k shots is pretty good, all things considered; much better than I would have predicted, actually. If the MOSFETs are the original spec'ed (AU)IRF1324 then they were - and still are - an excellent choice, so I'd use those for the replacements, but replace all of them even if just 1 or 2 are bad. I suspect much as you suspected that leaving the charger on while welding is likely what tipped them over the edge. As for the gate resistors, 82 ohms is a tad high, but not unreasonably so. Finally, a good freewheeling diode would be an MBR8020R which is an 80A/20V Schottky diode in a stud package with the anode as the stud. Bolt the stud to the output bus bar connected to all the MOSFET drains and then solder a fat, short jumper from the cathode terminal to the output bus bar connected to the positive terminal of the battery. I don't think any $3 Schottky's are going to cut it here, but I didn't expect the MOSFETs to survive 35k shots, either, so can't hurt to try the cheaper one first. Just remember to always wear your safety glasses when you play with stuff like this...
pguk said:
Erf...Neither, though I can see how what I wrote might be interpreted that way now... The decoupling capacitor goes across the battery, but needs to be on the board to do any good - it "decouples" the inductance of the battery side wiring so it doesn't cause an excessive droop when the MOSFETs turn on, nor an excessive voltage spike when they turn off (most important, that). Please look at the screenshot I attached to my initial response and all will be clear!
okashira
10 kW
pguk said:
This one works great. http://www.mouser.com/Search/ProductDetail.aspx?R=VS-100BGQ045virtualkey61370000virtualkey844-100BGQ045
I wouldn't bother with a cap. It would need to be rather massive (expensiveness) to absorbe the inductive energy at turn off.
okashira
10 kW
Pguk, just looked at your new shots.
Add another TVS is parallel and it will take care of the rest. Also the diode I reccomended with as short as possible wires.
Youve gone from 70us at 29V to 7us at 24V. That should help the fets last a couple order of magnitude longer.
I was seeing 31V for 110uS :-O
Add another TVS is parallel and it will take care of the rest. Also the diode I reccomended with as short as possible wires.
Youve gone from 70us at 29V to 7us at 24V. That should help the fets last a couple order of magnitude longer.
I was seeing 31V for 110uS :-O
okashira
10 kW
fechter said:
The problem is kinda the size of the leads.
If you made them out of two laminated busbars you could really make a difference.
Twisting them together might help by 50% or so.
Btw I can make a paper clip jump if it's near the lead when welding :-O pretty impressive for only 10ms pulse. I bet the field is over 1T
okashira said:
DEAD wrong!
Anything will help from something that costs $1 to something that cost $100+ but I would expect to spend $5-$15 for very good results.
The cap is to absorb the initial spike the battery is to far away. No amount of laminating or twisting battery wires will solve that as well the battery is not nearly as good at absorbing fast spikes as a properly selected and placed cap.
I have massive testing in this area and I have seem some amazing things with some simple design rules and proper component selection.
okashira
10 kW
You have no idea what you are talking about. Read the thread and stop coming in here shooting from the hip confusing people. This is not a half bridge topology you can't just slap on a small dc link cap and call it a day.
okashira said:
Not at all. When mosfets (or any transistors) that are conducting current are turned off the inductance in the circuit keeps the current flowing for a length of time.
That current needs to find its way back to the battery or a cap or both.
The amount of inductance in the circuit and the current flowing will determine the amount of energy the free-wheeling diodes and the cap/battery have to deal with.
The inductance from the free-wheeling diodes to the device used to absorb the energy and the device's ESL will determine the size of the voltage spike across the free-wheeling diodes.
Again. I have turned mosfets on then off and studied this in great detail. When you put a cap as close to the free-wheeling diodes as close as you can with the lowest ESL you can get it will make a huge difference in the spike. Added diodes to help the free-wheeling currents also help. But in short component selection and circuit layout is key!
parabellum
1 MW
Yup, okashira is right, you need to go back and see what is discussed here. The circuit is actually clamped on the battery terminal, and free wheeling occurs on opposite to battery side I would say.Arlo1 said:
parabellum said:
That's how it works with a phase leg or a H bridge as well.
When the transistors on 1 side turn off the diodes in the opposite side have to conduct the energy back to the cap.
The boards I have work on the negative side. By the end of this I will have designed new boards which will have added diodes and caps to make it work buttery smooth.
parabellum said:
No, okashira is wrong, and there are *two* inductances that need to be dealt with. Energy stored in the inductance of the welding cables circulates through the freewheeling diode (if present), but what about the inductance of the battery cable (assuming the other connection to the battery is made directly by bolting the board to the battery terminal)? A capacitor across the input rails on the MOSFET board will absorb energy stored in the battery cable inductance when the MOSFETs turn off, with a brief bump in voltage that is inversely proportional to the amount of capacitance (and directly proportional to the battery cable inductance and peak current).
The amount of capacitance need not be obscene, either, but it does need to consist of some very low inductance/ESR ceramic or film capacitors in parallel with some electrolytics. A total capacitance around 4700uF should be sufficient to limit voltage overshoot to 20V with ~30cm of battery cable and 2000A of peak current. Personally, I'd go with a few uF of film capacitance (Sprague "orange drop" or the like) and 4-6 1000uF/16V elkos. [EDIT] Yes, 16V elkos are fine as the voltage will only exceed 16V very briefly and elkos will tolerate this abuse just fine.
However, there is not nearly as much energy stored in the battery side inductance - assuming there is one cable and it is as short as possible - so a TVS diode (at least 15KPA or SMDJ size) or just letting the energy avalanche the MOSFETs (and freewheeling diode, if present) is probably acceptable. It offends my sense of good engineering, but, then again, so does blindly dumping current from a battery into a spot weld, and that seems to work pretty well for most people here.
I am an outsider when it comes to discussing the inner workings of electronics, however...I get the "it's working fine on the original design" vibe...simple, cheap...if it blows, it's easy to diagnose and cheap to fix.
But...these components are really cheap! $2 capacitors, other components as well. All I want to know is...if I add $20 worth of Shottkey Diodes, low-ESR capacitors, or "X"...and it will make the device run much more reliably? why not?
I think these haven't been put into serious production simply because there is a view from the outside that some small change is about to happen, which will make the device MUCH better, and at a tiny increase in cost. I just want to know...
What improvements help?
How much do they help, and in what way?
What is the point of diminishing returns? (that point where it "could" be better, but any further improvements beyond "X" are expensive for a small bump)
You guys all know more than me about this stuff. I think we can get these into production very fast, and at a low price. Maybe a basic model, and the upscale deluxe model, If we can only get some kind of consensus. Help me...to help you.
But...these components are really cheap! $2 capacitors, other components as well. All I want to know is...if I add $20 worth of Shottkey Diodes, low-ESR capacitors, or "X"...and it will make the device run much more reliably? why not?
I think these haven't been put into serious production simply because there is a view from the outside that some small change is about to happen, which will make the device MUCH better, and at a tiny increase in cost. I just want to know...
What improvements help?
How much do they help, and in what way?
What is the point of diminishing returns? (that point where it "could" be better, but any further improvements beyond "X" are expensive for a small bump)
You guys all know more than me about this stuff. I think we can get these into production very fast, and at a low price. Maybe a basic model, and the upscale deluxe model, If we can only get some kind of consensus. Help me...to help you.
Alan B
100 GW
Good to see a little deeper look into the transients in these circuits.
There's more than one way to protect the FETs, and it needs to be "good enough", not necessarily perfect, but it is hard to know what is "good enough" with the changing resistance, voltage and currents. Adding the right component at the right place can fix the problem, or adding a different component elsewhere can be a "bandaid" that improves the problem in a different way.
There's more than one way to protect the FETs, and it needs to be "good enough", not necessarily perfect, but it is hard to know what is "good enough" with the changing resistance, voltage and currents. Adding the right component at the right place can fix the problem, or adding a different component elsewhere can be a "bandaid" that improves the problem in a different way.
flangefrog
1 kW
Why is it that brushless controllers (at least the little ones like Kunteng) use a film cap on the mosfet drain for protection instead of a flyback schottky diode (and/or a TVS diode)?
Edit: I was mistaken, the caps were on the source.
Edit: I was mistaken, the caps were on the source.
spinningmagnets said:
Yes, this is something I've made at least a passing effort to address, which is that circuit seems to work "good enough" for most people and that it is rather cheap, so it might not make much sense to spend any money on protecting it from damage if it already lasts, say, 35k welds. I, personally, wouldn't be satisfied with that kind of lifespan even if I was just a casual user, but that may just be me.
So we want to concentrate on changes/improvements that will reduce the importance of selecting the right size battery, for example. Too small a battery and you'll be limited to welding very thin metal, or not at all, while too big a battery and you might exceed the peak current rating of the MOSFETs, or their avalanche energy rating (if no freewheeling diode is present on the MOSFET board - not dangling in the middle of the welding cables somewhere... ahem).
In that spirit, then, I would say that adding a freewheeling diode is mandatory, but the input capacitors can probably be dispensed with, especially if the suggestion I'll give below is implemented.
For the freewheeling diode, part number 100BGQ045 part number that okashira suggested a few posts back is superior to the MBR8020R I suggested, both in peak current rating and cost.
The original spec'ed MOSFETs - (AU)IRF1324 [the AU prefix means Automotive rated and is optional] is a really good choice, though I'd probably bump that number up to (8) in parallel so there isn't such a worry with the size/ampacity of the battery.
I would design the board so that it bolted onto the negative post of the battery and used a short-as-possible jumper to the positive post comprised of 5-7 #14AWG insulated wires wire-tied together. Doing this, rather than using a single larger diameter wire, will put the stray inductance of each wire in parallel, which will reduce it by the inverse of the number of wires paralleled (ie - down to 1/5th or 1/7th), and the lower this inductance the less need - perhaps no need at all - for decoupling capacitance (though as I mentioned before I consider this bad engineering regardless). Also, 5-7 #14AWG wires will be cheaper and easier to obtain than a single #4-#6AWG wire, and they will be easier to solder (in individual holes) to the PCB.
I would use a stronger gate driver IC - I'd like to see at least 4A peak here - and lower the value of the gate resistors accordingly. For example, MIC4420, a 6A driver for $0.95US, and 22 ohm resistors for each MOSFET (whether 6 or 8 are used).
One of the minor changes I would make is to get power to operate this circuit from the 12V battery by using a blocking diode and a bulk storage capacitor so the circuit still function while the battery is delivering welding current. There are undoubtedly a few other things that I would do differently, but this is already turning into a proverbial "Homer's catalog of ships" (re - The Iliad) so I'll leave off for now.
flangefrog said:
Dunno - I'm neither familiar with "little brushless controllers" nor do I have any special insight into what any particular designer might have been thinking when they designed something. It could be you are mistaken on the location of the capacitor, or it could be a means to achieve zero voltage switching, or it could be bad engineering.
pguk
100 W
Arlo said:
Arlo, you are as tactful as a brick!
I reckon the 2 diode solution which okashira has documented is good enough for most of us. I do have questions in my own mind about the advantage of using a Schottky over a home brew FET body diode (forward voltage of schottky's so far mentioned seem as high as my home brew). I also wonder if there are higher voltage rated FETs than the 1324 which have equal avalanche current handling capability, even in a TO-247 package, but these are details and he reports good longevity of his FETs after this mod.
However, I'm sure we all would love to see what a buttery smooth nicely engineered solution looks like
. So if it would help you I will send you one of my driver units? It allows programming of both pulses and the delay between them, or just a single pulse if desired. Looking forward to your results. PEdit
Tesseract said:
I think most are using the MCP1407 which is a 6A device. Smart suggestion about using 12v battery to power the control side also.
jmz
100 W
Tesseract said:
I agree heartily with this; freewheeling diodes can reduce the requirement to dissipate inductive energy in the leads, but not in the battery/capacitor path. Hence, you can avoid a freewheeling circuit by making a more robust snubber.
In my welder, each of the six switching cells has one conduction FET as well as an antiparallel shottky-capacitor snubber with 300nF of 100V C0G SMT capacitors and a single 3900uF 50V electrolytic (and 10 ohms in parallel with the diode for dissipation of snubber energy between pulses). All you then have to do is get the di/dt to a manageable level that can be absorbed fast enough by the ESL of the electrolytic capacitor. This is a tradeoff between switching speed (and hence energy dissipation during turn-off) and the amount of transient voltage rise while the electrolytic cap 'commutates' into the snubber path.
I was blowing up FETs at a 600 ohm gate drive resistance during 100 us (IIRC?) turnoffs, while allowing for a very low voltage rise, until I went to 100 ohms. I will report on what the peak voltages are at this switching speed.
Tesseract, thank you for the thoughtful reply, I think that answers my question.
I'm sure many garage enthusiasts want to use whatever 12V battery is collecting dust in the garage corner, but it is starting to sound like a specific battery size is best, in order to get consistent results. Many of these batteries are $80 new, so, that isn't a horrible price to have a spot-welding kit that works well. Worst-case scenario, you decide pack welding isn't for you, and you have a spare 12V battery for your car.
Could the "improved" version be done for $50 in parts? (assuming a circuit board run of 20 units to get the price down to a reasonable level?) with the $80 car battery of a specific type being sourced by the end-user?
Maybe the base model can be the original version, with no housing, but publish the code for 3D printing one up in the maker-verse?
Deluxe version has the upgrades, and comes with a plastic housing, on/off switch, LED to indicate power is on, etc?...
I'm sure many garage enthusiasts want to use whatever 12V battery is collecting dust in the garage corner, but it is starting to sound like a specific battery size is best, in order to get consistent results. Many of these batteries are $80 new, so, that isn't a horrible price to have a spot-welding kit that works well. Worst-case scenario, you decide pack welding isn't for you, and you have a spare 12V battery for your car.
Could the "improved" version be done for $50 in parts? (assuming a circuit board run of 20 units to get the price down to a reasonable level?) with the $80 car battery of a specific type being sourced by the end-user?
Maybe the base model can be the original version, with no housing, but publish the code for 3D printing one up in the maker-verse?
Deluxe version has the upgrades, and comes with a plastic housing, on/off switch, LED to indicate power is on, etc?...
okashira
10 kW
Again, thanks to Tesseract and Arlo for confusing everyone again. :-(
flyback + TVS diode solves the problem. period.
This is not a half bridge attached to a motor controller.
The fets are sitting right on the battery.
A teeny tiny amount of avalanche is just fine and dandy. They are rated for it.
Flyback takes care of 90%
TVS takes care of 9% more.
The rest doesn't matter.
But now this whole thread has been reset because of the confusion you guys created. Great job! (sarcasm)
I am going to update the OP soon with a clarified explanation.
flyback + TVS diode solves the problem. period.
This is not a half bridge attached to a motor controller.
The fets are sitting right on the battery.
A teeny tiny amount of avalanche is just fine and dandy. They are rated for it.
Flyback takes care of 90%
TVS takes care of 9% more.
The rest doesn't matter.
But now this whole thread has been reset because of the confusion you guys created. Great job! (sarcasm)
I am going to update the OP soon with a clarified explanation.
okashira
10 kW
flangefrog said:
I'll answer the question: A brushless controller uses half bridges, and the bottom fet diode already serves the purpose of a flyback diode for the load.
But this is a thread about a spot welder that is just a simple on off switch. So, we need to add a diode to flyback the load, and the rest is taken care of by a TVS diode. Furthermore, the battery inductance is minimal compared to an EV.
Again, read the thread and stop coming in here shooting from the hip confusing people. I am talking to two of you!!
okashira
10 kW
Tesseract said:
No, I am right. Why? Because TVS diode just works to take care of the last bit. 10s of thousands of welds. It's not even needed... the tiny avalanche with the flyback only is just fine with the 1324's. They are rated for it.
Sure, you could 100% stop avalanche if you put a cap right there with 10nH of inductance and intentionally slowed the switching speed with a larger gate resistor. But it doesn't matter and you're really over-complicating things at this point.
