On 2/17/2015 5:03 AM, Chris Jones wrote:


You could also disconnect some of the caps.
I was hoping to stay within the current rating of the caps which is
1900A each, so I would want several in parallel to get enough welding
current. The caps are capable of developing nearly 10kA each according
to the datasheet but that supposedly damages them and they are somewhat
expensive.

The inductance in the path limits the rate of rise.
The resistance in the path limits the peak current.
Depending on your construction methods and means to equalize
current sharing, you might find that removing some caps
doesn't much affect the stress on the others.
Doesn't take many of those extra milliohms in wires and
connections to put a serious dent in how much current you can deliver.

Or maybe not fire
all the FETs...although that may be easier said than done.
Again, I was intending to stay within the current rating of the FETs so
I would want all of them on to share the current as evenly as I can,
even if I have to do something else to reduce the current. I could
reduce the current by adding some cable resistance if there is not
already enough. I would still want to have the FETs there as a way of
switching it off at the end of the weld. Maybe PWMing it to regulate the
current would be worthwhile, but it probably means twice as many FETs.
Still I might need catch diodes if I didn't put the second lot of FETs
so maybe that is no extra problem.

It's all about your objectives. For me, I decided that I needed to weld
battery tabs. I have stick and wire feed and Acetylene and Miller
handheld spot welders for big stuff with long weld times.
Building something more universal would be a lot of effort with little
benefit.


I had great difficulty measuring output current because the huge
magnetic fields generated coupled into
everything.
Yes, that sounds difficult. I guess you could use that as the way of
measuring the current, e.g. a Rogowski coil or whatever it is called.

and it would also be very useful to be able to adjust the current by PWM
during the weld.
Closed-loop PWM sounds great in theory. But, for short weld times, you
need a wide
bandwidth control loop. But you want to filter out all those
transients. And there are lots of internet discussions about
blowing up all their FETSs and being unable to adequately protect them.
You're gonna have a lot of really big parts in a design where resistance
and inductance are your enemy. Something as simple as a foot-long
buss bar connecting the some might not behave quite like you'd expect
by looking at the schematic. Thick conductors aren't much help if
the skin effect confines the current near the surface.
Would be interesting to see a simulation
of the current distribution when you short the array of cap busses with
a weld. I don't think I have any tools that model skin effect.
Might be interesting to weave some "litz-type" wire out of smaller
conductors.

See the link below
Therefore a lot of MOSFETs would be required. It seems
that the best current rating per dollar occurs for individual MOSFETs
rated at about 100A, so about 100 of these in parallel would be required
for 10kA. I think a totem-pole style half-bridge topology might work,
using the output cables as an inductor to smooth the output current. A
multi-phase PWM arrangement with multiple output inductors could make
better use of the current rating of the caps. It would be an interesting
project but I don't have time to do it yet. I am somewhat concerned
about what would happen in the event of one failed MOSFET, and I would
like to think of a way to mitigate that. Perhaps the bondwire or package
pin would be an adequate fuse.
Each FET has to be able to handle all the current it can deliver with
the output shorted at the weld. Limit the charge current to the caps
and a short won't cause much problem.
I think it could be a problem... Hehe these caps are pretty amazing, one
of them will easily melt a nail well after the charging current is
disconnected. Then another nail, and another. 3000 Farads is more like a
small capacity battery (but with less resistance than a battery).
If you have 100 parallel paths, you should be able to do some good with
fusible links. Just watch those milliohms. Doesn't take too many to
ruin your ability to deliver that current.

I wonder if you couldn't run a wire from each FET thru a common toroid
before the junction point to help equalize the peak currents??
I was more thinking putting the wire from each FET stage through its own
toroid, and sensing the currents in each wire and regulating it with
individual PWM for each stage. That sounds like too many parts so maybe
just having separate longish wires from each FET stage would be enough
sharing resistance even with common PWM for all stages. Anyway I won't
build this for ages so I will have forgotten by then.

As for failures, anything that would "blow" under short conditions would
seriously undermine your ability to deliver current to the "short"
at the weld.
Well I was not intending to exceed the ratings of the FETs, so if there
are 100 FETs, each with wires that fuse at say 200A, but normally each
conducts 100A, then if one FET fails short and starts carrying a few kA
it will be good if its wires fuse cleanly rather than keeping conducting
and starting a fire. As long as I don't exceed each FET's ratings it
would not be a problem for each FET to have a fuse action at some higher
current. I am reminded of the photos of inside a Tesla's battery pack
that seems to have a lot of cells in parallel, each cell with an
individual thin fuse wire in series with it.

The advantage of CD is that you get a defined energy pulse.
The energy delivered to the weld point is relatively insensitive
to the resistance of the contact. Far less so than with a MOT.

I've watched a lot of youtube videos of successful welds using
racks of low voltage caps and arrays of FETs.

My Unitek uses about 400V discharged into a pulse transformer.
That V^2 factor for energy storage adds up rapidly.
The magic is in the pulse transformer. It's smaller than my
fist and claims to put 7KA into .001 ohms.
Would be interesting to learn how to build one of those.
From this:
http://i.imgur.com/ZeZerGx.png
it appears that they might pass current backwards thru the transformer
to "reset" it.
Interesting, but I hope I don't need to build one, partly becuase it
restricts pulse length etc.

Yep. That wasn't part of my plan because I want the shortest reasonable
pulse width. You can get your weld fixture up close and personal to the
secondary of the transformer. All your drive circuitry gets easier
by the transformer ratio. One SCR is far easier to manage than 100
FETs. It's easy to get seduced into making something far more complex
than is required, "just because you can." ;-)

My thinking was that, if I have so many joules to deliver,
if I deliver them fast,
more goes into the weld and less is conducted away by the weldment.
In that scenario, you should be able to control the cap voltage
over some range and range-switch by disconnecting some caps.

Twice, I welded tabs onto a 2032 coin cell for a laptop.
Both times, the open circuit voltage dropped from 3.2V to 3.0V.
Even a short, single pulse did some damage.

I have one of those handheld spot welders that welds "nails" onto
car bodies for pulling dents. Might be interesting to examine the
characteristics of that transformer. I expect its inductance precludes
use as a short pulse transformer, but might be interesting as a replacement
for the MOT.