Big Coil Build Log

[flyingperson23]

New coil time! This one is joining the long line of MITERS coils.

Specs:
Inverter: BSM400GA120DN2 full bridge, possibly upgrading to CM600 or FZ600 if those die   --now CM600--
MMC: 733nF 18kV, made with 15x .44uf 6kv dawn caps, rated for 300A rms but heavily air cooled just in case
Bus capacitors: 12000uF 1050v - I shouldn't need as much as for example br0din coil as I'll hopefully have almost twice the power factor
Primary: 3/8in copper tube, 3/8 in spacing, ~7in inner diameter
Secondary: 12.75"x42" 22awg
Topload: 42"x12" skeleton topload, 1/2 in aluminum tube, 13 rings, 15° spacing
Power supply: up to 800V from a boost converter, likely will run at closer to 400V with looooong on times.
Driver: UD3.1 with custom code to run the boost converter
Expected to run anywhere from 10-30kW, maybe higher for short runs

Boost converter:
Uses ff600 half bridge with top igbt shorted open, 380uH 150A inductor from a prius inverter, switching at 16kHz using PID control

Primary current: likely ~1kA with freewheeling, at least 1ms on time, 15-25% dtc max

Top view of mostly finished electronics:


Front view:



Precharge uses capacitors instead of resistors as it's AC. I used 30uF motor capacitors, they work great! much faster than resistors with no power loss

[davekni]

-cat5 for GDTs is prone to flashing over with high bus voltages. cat6 is much less so due to the plastic barrier between wire colors.

Usually each twisted pair is split, half primary and half secondary.  That minimizes parasitic leakage inductance.  Grouping wires to take advantage of plastic barrier for voltage will increase leakage inductance.  Added inductance may be OK for a low frequency coil.

BTW, what is your line voltage input?  240Vac single phase?   Or do you have access to three-phase?

[Mads Barnkob]

Nice big coil in works there! I had to google MITERS, I thought you mean "meters" as in a coil that could make sparks longer than a meter and here, was the forum, so got a little puzzled.

I like the idea of a simple boost converter, can you share more details on how its controlled?

I think you will get problems with your DC bus wiring, those long red wires are going to add an unwanted amount of inductance. Might even limit your peak current ability and will melt with those planned on-times.

What is the capacity of the MKP cap sitting directly on IGBTs?



Your safety feature with disconnecting GDT, before discharging DC bus seems like a lottery. You leave your IGBT gates in a uncertain state with possible current flowing and a full DC bus on it. I foresee a race condition problem with blown up silicon as the result.

[flyingperson23]

Usually each twisted pair is split, half primary and half secondary.  That minimizes parasitic leakage inductance.  Grouping wires to take advantage of plastic barrier for voltage will increase leakage inductance.  Added inductance may be OK for a low frequency coil.

BTW, what is your line voltage input?  240Vac single phase?   Or do you have access to three-phase?

My GDTs do probably have a lot of leakage inductance. Everything all worked fine at 100khz, actual Fres will likely be <40khz. Someone I know had a GDT flashover which killed his IGBTs, driver, and blew up bus connections and capacitors. I think a little extra inductance is fine to make sure that doesn't happen.

Input is designed for 240v 30A 3ph, but would run fine on single phase or even a dc source like an EV battery.

Nice big coil in works there! I had to google MITERS, I thought you mean "meters" as in a coil that could make sparks longer than a meter and here, was the forum, so got a little puzzled.

I like the idea of a simple boost converter, can you share more details on how its controlled?

I think you will get problems with your DC bus wiring, those long red wires are going to add an unwanted amount of inductance. Might even limit your peak current ability and will melt with those planned on-times.

What is the capacity of the MKP cap sitting directly on IGBTs?



Your safety feature with disconnecting GDT, before discharging DC bus seems like a lottery. You leave your IGBT gates in a uncertain state with possible current flowing and a full DC bus on it. I foresee a race condition problem with blown up silicon as the result.

A friend helped with the boost converter code. The way it works is 2 nested PID controllers. The outer one measures voltage deviation from the setpoint and based on that outputs a target current, something like 0-50A. This goes to the inner controller, which measures the difference of the target current and the actual current, and from that outputs a duty cycle to the IGBT.

The capacitor directly on the bus is a 560uF electronicon dc link capacitor. I'd hoped it would take most of the resonant current, but I'll probably switch the red wires to 2x 4awg twisted together for less inductance.

There are 1k resistors between G-E of each igbt, so when I turn off the GDT signal, all IGBTs should turn off. I'll probably be running within the datasheet current limit. When I added that contactor, if I turned off AC it would be a few seconds before the UD went into UVLO. I've since added fans on the same power supply that draw 2.5 amps so the UD goes into UVLO very quickly, so I think the contactor may be unnecessary. I just wanted to make sure the arc would shut off as soon as I pressed the estop.

[davekni]

Input is designed for 240v 30A 3ph, but would run fine on single phase or even a dc source like an EV battery.

That is a very unusual voltage to find in US.  Usually 3ph is either 208V phase-to-phase (120V phase-to-neutral), or 480V (277V to neutral).

The capacitor directly on the bus is a 560uF electronicon dc link capacitor. I'd hoped it would take most of the resonant current, but I'll probably switch the red wires to 2x 4awg twisted together for less inductance.

Less inductance is good.  I'd agree that 560uF will handle resonant current.  2x wires are likely unnecessary.

There are 1k resistors between G-E of each igbt, so when I turn off the GDT signal, all IGBTs should turn off. I'll probably be running within the datasheet current limit. When I added that contactor, if I turned off AC it would be a few seconds before the UD went into UVLO. I've since added fans on the same power supply that draw 2.5 amps so the UD goes into UVLO very quickly, so I think the contactor may be unnecessary. I just wanted to make sure the arc would shut off as soon as I pressed the estop.

1k is unlikely to be stiff enough (low enough resistance) to keep Vge below threshold.  Remaining GDT magnetization current will be a problem, as will gate charge caused by Vcg swings and Ccg as primary current rings down.  IGBTs might or might not survive.  I'd remove GDT contactor.

[flyingperson23]

Input is designed for 240v 30A 3ph, but would run fine on single phase or even a dc source like an EV battery.

That is a very unusual voltage to find in US.  Usually 3ph is either 208V phase-to-phase (120V phase-to-neutral), or 480V (277V to neutral).

The capacitor directly on the bus is a 560uF electronicon dc link capacitor. I'd hoped it would take most of the resonant current, but I'll probably switch the red wires to 2x 4awg twisted together for less inductance.

Less inductance is good.  I'd agree that 560uF will handle resonant current.  2x wires are likely unnecessary.

There are 1k resistors between G-E of each igbt, so when I turn off the GDT signal, all IGBTs should turn off. I'll probably be running within the datasheet current limit. When I added that contactor, if I turned off AC it would be a few seconds before the UD went into UVLO. I've since added fans on the same power supply that draw 2.5 amps so the UD goes into UVLO very quickly, so I think the contactor may be unnecessary. I just wanted to make sure the arc would shut off as soon as I pressed the estop.

1k is unlikely to be stiff enough (low enough resistance) to keep Vge below threshold.  Remaining GDT magnetization current will be a problem, as will gate charge caused by Vcg swings and Ccg as primary current rings down.  IGBTs might or might not survive.  I'd remove GDT contactor.

Oh yes sorry, I meant 208v. I know it's 208 but for some reason my brain defaults anything starting with 2 to 240

Fair enough for the contactor, it's probably a good idea to remove it.

[Mads Barnkob]

The capacitor directly on the bus is a 560uF electronicon dc link capacitor. I'd hoped it would take most of the resonant current, but I'll probably switch the red wires to 2x 4awg twisted together for less inductance.

Less inductance is good.  I'd agree that 560uF will handle resonant current.  2x wires are likely unnecessary.

Current distribution in parallel capacitors is split on capacity ratios, please tell me if I'm wrong

So let's say you pull 100A from the DC bus and you have 560uF and 12000uF, the 560uF will supply 4.5A and the 12000uF 95.5A (so the largest current part is still hindered by the "long thin" wires)

[klugesmith]

Current distribution in parallel capacitors is split on capacity ratios, please tell me if I'm wrong
So let's say you pull 100A from the DC bus and you have 560uF and 12000uF, the 560uF will supply 4.5A and the 12000uF 95.5A (so the largest current part is still hindered by the "long thin" wires)

The partition of RF current might depend a lot on the inductance of wires.   Could be modeled with a SPICE simulator -- good opportunity for newbies to advance from transient analysis to AC analysis.

In some cases a parallel combination of large and small capacitors has an impedance magnitude much _larger_ than either set alone.
For example, when a big FPGA is decoupled with dozens of MLCC's of two values, the low-C array can resonate with the mounting inductance of the high-C array.
This was familiar in PDN engineering circles when I had to fix a 5 MHz resonance about 15 years ago.
https://resources.system-analysis.cadence.com/blog/msa2021-how-pdn-loop-inductance-contributes-to-ripple-and-total-impedance

[flyingperson23]

How would I find the inductance of the wires?

Disregarding wires the larger capacitors should be a higher impedance path to the bridge due to unavoidable inductance and higher ESR as the smaller is a film capacitor. I assumed this difference was large enough that the smaller capacitor would handle most of the resonant current, but I may be wrong.

Experimentally, I've drawn 300A at 20% dtc from the bridge and saw no voltage drop along the long wires or heating anywhere.

[klugesmith]

>> How would I find the inductance of the wires?

Try an Internet search for inductance of twisted pair.
You can find formulas and calculators at, for example, https://www.allaboutcircuits.com/tools/twisted-pair-impedance-calculator/ or https://www.eeweb.com/tools/twisted-pair/

Inductance depends on log of (wire spacing / wire diameter), but the empirical formulas are less accurate when the insulation is much thinner than the conductors.   

If you conclude that the inductance is an important bottleneck, you could reduce it by using more wires, e.g. 4 wires in pattern with square cross-section, arranged so each wire's closest neighbors have current flowing the other way.

[davekni]

The partition of RF current might depend a lot on the inductance of wires.

Exactly what I was thinking, why I guessed wire size was not an issue.

Inductance depends on log of (wire spacing / wire diameter), but the empirical formulas are less accurate when the insulation is much thinner than the conductors.

Very rough estimates should be fine here.  Physical layout prevents twisting for  ~half of wire length anyway.

Just looking at existing wiring, I'd guess between 0.6uH and 1uH for wires.  At 40kHz coil, ripple current is 80kHz.  0.6uH is ~0.3 ohms reactance at 80kHz.  560uF capacitor at bridge is ~0.0036 ohms.  Thus very little 80kHz current will flow through wires.  Interconnect inductance of 560uH capacitor is likely higher than 0.0036 ohms, so may control wire current more than capacitance.

Edit:  Forgot about pulse-skipping.  That will generate lower frequency ripple current.  At 20kHz, current sharing would be closer to 5:1.  But I think RMS current component at 20kHz will be lower than 80kHz of full-on operation.  Would need to simulate or hand-calculate a bit more to be sure of this.

The biggest risk may be if pulse-skipping frequency happens to hit exactly at LC resonant frequency of 560uF and wire inductance.  That's where some form of R+C snubbing might be needed at bridge.

[flyingperson23]

Update:
Removed gate contactor
Blew up IGBTs probably from a gate TVS coming loose and my GDTs apparently being a lot worse than I thought
IGBTs replaced with CM600HA-24H and EVR gate drive boards
GDTs replaced with small commercial ones from pulse electronics that appear to work well enough
Twisted pair of 4awg from bus caps to bridge instead of loose 10awg

Coils and topload have been made:

[davekni]

GDTs replaced with small commercial ones from pulse electronics that appear to work well enough

Commercial GDTs usually have higher leakage inductance than good hand-wound ones, often much higher.  Commercial GDTs need to meet regulatory isolation voltage requirements and are also optimized for size and cost.  Might still be good enough, but I'd scope Vge waveforms again when running with reasonable bus voltage.