I wanted a silk purse but all I could afford was this sow's ear. YA spot welder

[BradC]

G'day all,

Having used a cheap $30 spot welder for a while with a set of large SLA batteries, I was lusting after a KWeld with all the fruit.
After consulting with the ministry of finance I ended up with another Aliexpress spot welder, some new cables and an Aliexpress 16V 500F SuperCap pack.

This board as 12 MOSFETS, can do dual pulse and has no manual switch input. I put an 100A shunt on some long (1M) 6AWG cables and hooked it up to the caps to try and measure switching times and inductive pulses at switch off. The rise time was ~180uS and the fall time was ~1400uS. No inductive pulses to be seen because the turn off was so damn slow. On the other hand, I was seeing ~900A at 8V.

The gate control circuit is an opto-coupler as a pull up to the +ve rail, and a 4.7K resistor to bleed it back down. Slow to rise, much slower to fall.

So, mods. Firstly I wanted a location for a flyback diode, and a TVS to prevent avalanching the fets with longer cables. These fets are rated to 1418mJ avalanche, and I have 12 of them, but if I can avoid it entirely that'd be nice. I also wanted to see what I could modify to improve the current handling a bit.

I started with a socket for a footswitch. Easy. Just interrupt the signal from the -ve probe to the uC and run it via a socket. If i have a plug in the socket it's manual, and if not then it's auto. From there to replacing the brass bus bars on the back with some 4.13.1mm dia copper, routed to give me a point to terminate the diodes.

The last one I wanted to tackle was the MOSFET driver. Probing around showed the uC was driving the optocoupler by pulling the cathode down, anode to +5V. I settled on an MCP1406 because it handles the inverted input. I also ordered some low ESR 330uF poly caps. That was tacked onto part of the opto footprint with the +ve rail and caps dead-bugged on the top. Input and output were straight from the adjacent pins on the opto footprint and power to the +ve rail via a 1N1418 and series 39 ohm resistor to be a bit kind to the diode on power up.

Measurements using a current limited PSU (rather than the caps) indicated over 2 concurrent 100ms pulses, the gate drive voltage dived about a volt. Removing the 4.7K pull down resistor solved that, so the gate drive is a solid VCC-~0.3V at maximum pulse length (2x100ms) even when the input voltage has dropped below 1V. The uC keeps running because as soon as it wants to fire it turns off the display and beeper to conserve the logic rail.

I hooked the caps back up and did some 2ms pulses into the shunt. Saw ~1000A at 8V and ~1700A at 14V. I could improve that significantly by tweaking the cables (in theory ~2100A in the config I'm looking at), but that's enough to blow holes in things so it'll do for now.

Rise time is now ~2-3uS and fall time is ~5uS. There is a decent turn off spike to ~28V and that is nicely managed/clamped by the BZW50-15 TVS and 100BGQ045 flyback diode.
Not surprisingly it welds a bit more consistently now, and those cheap aliexpress caps can seriously supply some current.
Some simple measuring and modeling shows resistances in the order of :

• 10AWG input cables : 1.8mOhm
• 6AWG welding cables : 2.3mOhm
• Board and terminations : 1mOhm
• MOSFETS : 0.1mOhm
• Battery : 2.1mOhm
• Shunt : 1mOhm

Total cost about $260. $223 for the caps and welder board and the rest on parts & sundries (including new probes and cables).

Board front & back. All the bus-bar work was done using a hot air pre-heater at about 150C on the other side of the board. It was impossible to solder otherwise.



Shunt for testing. Pulses > 20ms warm it up very quickly.


PS. Yes I know the soldering looks like snot. Unfortunately as I've aged I seem to have picked up a pretty determined shake. So much so I can almost solder both ends of an 0805 simultaneously while trying to hold the iron still.

[BradC]

Note to future Brad :

Removing the 4700 Ohm gate pull down is all well and good, but with an inverting driver when you remove power from the board the gate gets pulled high. It stays high for a minute or so while the 330uF cap supplying the driver slowly discharges.

So if you happen to have the probes bolted to a shunt and the contact bounces while plugging the board into the capacitor bank (which you've just charged to 14V, up from the 8V used for testing), the second time the power is applied the MOSFETS are already conducting and don't stop until the micro boots up and turns the output on.

If this does occur, then the 100A shunt changes colour, and as soon as you pull the power the inductive kickback that isn't shunted to the input caps (because you've just disconnected them) generates just enough power to short circuit the TVS diode which then explodes and blows a lead in half.

Slightly messy. Thankfully I bought a spare TVS.

Of course when in use the probes wouldn't be bolted together when applying power to the welder, but I'll put a label on it just in case I do something stupid again.

On the other hand, I'm now getting 1300A at 8.2V. A 20ms pulse makes the leads jump significantly. It also warms the shunt nicely. After the "incident" I'm not sure the shunt will be as "calibrated" as it was before the roasting.

[BradC]

As a mate of mine tells his kids "Play stupid games, win stupid prizes".

After replacing the TVS and getting it all buttoned up I was getting nice 2KA pulses at about 12V. I thought "I know, I'll wind it up to 16V (limit of the caps and welder) just to see what will happen".

What happens is at 2300A (not quite 14V) the inductance of the stupidly long 650mm weld cables is enough to destroy the TVS. So at < 2300A it's all nice, reliable and with pleasant waveforms with no excursion past rated voltages. At > 2300A the TVS lead turns into a fuse. I've made a note to keep the voltage < 12V.

Whoops! I'm onto my last TVS, so I'll have to try and avoid blowing the leads off this one.

[p.larner]

i have one of the cheap chinese 5 fet jobs,i am on my second one,the first set onfire after 3 fets shorted on its first weld!,on the next unit i fitted a doide+470uf cap as shown in a yt vid,its worked fine since,total cost was about £20.

[myrddin669]

Hi I have managed to implement your gate driving circuit modification but the freewheeling Schottky diode gives me headaches.
Without The Schottky everything runs as it should, but as I'm trying to weld nickel 0.15l/ copper 0.2 sandwich  with rather aggressive 30/70-80ms settings, MOSFETs reach uncomfortably >50'C temps after 10-15 welds, so I suppose they are avalanching, but after I connected the Schottky diode the uC either freezes on the first pulse setting or  goes into endless pulse shooting hiccup. First symptom is with smaller Schottky MBR6045, second with the diode you used vs100bgq045, also with rather big 201cnq045.
I assume it's linked with the Schottky leakage current messing with the uC contact probes sensing.
How did you manage to overcome this behavior? I can see some 1K resistors on the photos you attached, one parallel with the tvs and the other I'm not sure in parallel with positive rail?
Could You or maybe someone else elaborate on this matter?

For the reference, I'm using the same board as Yours - KEKK K100 , but with 6x G006N04 D2PAK-7 for now (another six waiting in the bin), busbars replaced with solid copper bars 6mm2. MIC4451 for the driver, 10uf ceramic and 820uf solid electrolytic, power for the driver through 33R and 1n4007.
Power source - 2 lipos in parallel - 3s 120C 7200mah, 2,5mohm IR combined.

Regards
Damian

[BradC]

Hi I have managed to implement your gate driving circuit modification but the freewheeling Schottky diode gives me headaches.
Without The Schottky everything runs as it should, but as I'm trying to weld nickel 0.15l/ copper 0.2 sandwich  with rather aggressive 30/70-80ms settings, MOSFETs reach uncomfortably >50'C temps after 10-15 welds, so I suppose they are avalanching, but after I connected the Schottky diode the uC either freezes on the first pulse setting or  goes into endless pulse shooting hiccup. First symptom is with smaller Schottky MBR6045, second with the diode you used vs100bgq045, also with rather big 201cnq045.
I assume it's linked with the Schottky leakage current messing with the uC contact probes sensing.
How did you manage to overcome this behavior? I can see some 1K resistors on the photos you attached, one parallel with the tvs and the other I'm not sure in parallel with positive rail?

I can't explain the freezing I'm afraid. Do you have the TVS across the output? I can imagine the ringing generated with the freewheeling diode and no clamp could be quite extreme.

The resistors were added to facilitate the external trigger. Leakage doesn't appear to be a problem on mine. The welder is triggered when the device senses voltage on the -ve terminal. When I broke open the sensing circuit to add the socket the pull-down was on the wrong side of the socket, so I had to add the pull down resistor in parallel with the diode.

The other resistor is supplying the socket and is simply a current limiter in case something gets shorted externally (any external contact or short with those Ultra-caps makes lots of sparks/smoke).

[Doctorandus_P]

The thick red wire attached to your shunt resistor is only inserted halfway in the crimp connector.

[BradC]

The thick red wire attached to your shunt resistor is only inserted halfway in the crimp connector.

Yes.

[myrddin669]

Thanks so much for your answer. I have tried with the TVS also, but only in tandem with 100BGQ, and ive had the other issue - after connecting the batteries, welder beeps with short intervals ca 1s and reboots constantly, so i may have uncovered some undocumented disco use with it :F
I probably borked something in the process of modification, i will be playing with it in the future, maybe i will find out.
Thanks for documenting the mods, i think it's worth more research and interest, as many people have success with these boards, the board layout must be somehow very good, but weak driver and inadequate protection for the mosfets (which you successfully addressed) makes it it very limited in its use.
I will be getting 2 more boards in the  near future, so maybe I'll be smarter soon.

Thanks for your interest
Regards
Damian