I have finally begun pack construction in earnest. Offroader's Q76R pack pictures gave me some solid starting points on how to best achieve a moderately sized (198 cell) Q76R pack, and so I eventually determined that a hexagonally packed layout of 22 flat 9P strips would fit straight in, giving a simple and robust pack construction with a minimum of interconnection distance. Each 9P strip would have a single strip parallel interconnection, and then be connected to the adjacent strip with 9 short strips on top of that. I should mention at this point that I only ever considered the use of 0.3mm nickel, which brings me to.. the Deathwelder.
https://imgur.com/a/xi2Hz
I thought it would be a doddle to cobble together a capacitive discharge welder. Although I am a power electronics engineer by trade, this was the first time I have ever designed or worked with pulsed power applications, and there was a bit of a learning curve. I went through multiple iterations after leaving smoking messes on both the board and an unlucky cell.
The principle of the design was using higher voltage components in a quest for More Power. I have access to a large supply of scrapped 50V 3900uF/4200uF electrolytic capacitors, so that became the target point for the working voltage level. Making a 400mF capacitor bank was the easy part, as was making some huge braided welding leads and threaded copper rods for welding electrodes.
The parts I obtained were 10x (soon to be 9x) Infineon IPB010N06N, 60V, 1 mOhm, 720A pulse rated
By using an inherently non-current-limited source and relatively less robust 60V MOSFETs, the path of simply relying on the avalance capability of the devices was closed to me. I had also elected to use honking great big leads with plenty of inductance, as well as a lumped capacitor bank which distributes even more inductance, so turn-offs have to be carefully managed or else they can easily turn fatal for the MOSFETs.
I considered that simply adding flyback diodes would be insufficient to protect the switches, as the inductance between the switching stage and the capacitor bank would still be more than enough to nuke the FETs at these current levels. I concentrated then on building the best diode-capacitor snubber I could, to absorb the entirety of the inductive energy right where I needed it, directly across the FETs.
As you can see from the images, working in my favour was a large busbar comprised of thick copper soldered to a PCB. This allowed a wide surface to distribute the current evenly, and to interleave the snubbers directly with the switching FETs. Each of the four switching FETs are interleaved within the five snubber FET-diodes (if I had some other suitable diodes I could have devoted the rest of the five snubber FETs to switching devices also). Each snubber has a couple of high voltage 22nF MLCC caps, as well as one of the same large electrolytics.
I quickly discovered that it was critical to slow the turnoff of the main switches devices to ludicrously slow levels (~4 us turnoff, through 800 Ohm gate drive resistors to a set of beefy paralleled MOSFET drivers), or else the inductance in the snubber electrolytics would not be able to keep up with the rate of current rise and they might as well not even be in the circuit (the 22nF MLCCs are only good for suppressing the most short and vicious of transients).
This got me almost to my goal of having a reliable multi-kA welder, with double/triple-pulse commutations not having the slightest impact on the main FETs, which stay below their avalanche voltage at all times and thus stay icy cold. I was getting good weld penetration into 0.3mm nickel at 25V, however there was a disturbing tendency to occasionally cause small explosions and ablate the entire layer of nickel in a shower of molten metal and with a deafening retort. This caused the loss of one cell (pictured) when this effect punched right through the nickel on the cathode.
I tried many things, from variations in pulse length and timing (even attempting an extended repetitive pulsetrain of 100us pulses), surface prep, and electrode finish, but nothing seemed to work. Eventually I got a clue from the oscilloscope probe I had set up to monitor the bus voltage, that even during the initial 'softening/cleaning' pulses the contact resistance was implied to be rapidly being raised, and that it was the very same stored inductive energy that by being dumped into the elevated workpiece resistance generated a runaway electrical/thermal/mechanical process that ultimately leads to the workpiece being turned into plasma.
No possible clamp type mechanism would have a hope of dealing with this, but with the aid of some simulation I hit upon the idea of stabilising the workpiece resistance with a parallel load, essentially a permanent near-short circuit placed as close as possible to the welding probes. This load was provided by two parallel strips of nickel, soldered to fat terminals and bolted directly to the probes, aiming to provide a similar load resistance to the minimum encountered during a pulse sequence.
The effect was to greatly stabilise the operation of the welder; no more explosions and almost no sparking, at the cost of requiring more stored energy and raising the commutation current and FET voltages a bit. I think that switched lead-acid type welders could benefit from this arrangement as well at high power levels (mitigated by lower inductance and lower peak currents because of inherent current limiting though), and something like this is almost certain to be required if attempting that holy grail of spot welding copper.
3 9P sections down, 19 to go. Should go much faster now. It can do a weld every two seconds (limited by control logic for snubber discharge time), indefinitely. Eventually those parallel nickel strips will get quite warm though!
Arduino welder control code here if anyone is interested:
http://paste.ofcode.org/mJaGWxyuKDgeGennDNjYEN
In another thread, I presented some data to the effect that the stated kM (resistive torque production efficiency) is notably higher in the higher turn count QS motors. I have gone with a 19" rear to match the 26" MTB front wheel diameter, to give good rolling performance.
