CM400 Induction Heater

[markus]

Hi all,

by chance I got the opportunity to take over a half-finished induction heating project from a nearby school. Their main goal was to melt metal and pour some stuff, for example metal parts for other projects. They could not find anyone to continue the project however, and so it has been sitting around for some time now.
Since I already experimented with induction heating - a self-built heater based on Marko's royer design managing about 800-1000W - and always wanted to build something bigger, this seemed like the perfect opportunity.

From the school project I got a bunch of mostly separate parts, mainly the following:

• Mitsubishi CM400DU-12F full bridge
• 3 IXYS IXIDM1401_1505_O isolated integrated driver modules. Each contains two isolated gate drivers with 10 A gate current, 15 V positive and –5 V negative gate voltage. They have some nice features like short-circuit protection, active clamping, UVLO and built-in dead time delay.
• 4x B43586-S4338-Q4 Epcos caps rated at 3300uF / 350V (should be good for at least 25Arms each)
• 2x 1uF 1kV MKP snubbers (apparently they got those as a freebie with some other parts - nobody knew what they were supposed to be used for, so they did not find their way onto the project so far)
• 90x Wima FPK1 0.22uF 400V DC / 250V AC ±5% caps in 3 banks with 3D-printed enclosure
• 3 6mm work coils: 1x 7 turns 5cm ID, 2x 5 turns 9-10cm ID
• A graphite crucible setup with isolation
• radiator etc for water cooling
• 3 12V 123A server PSUs
• some copper / bus bar pieces
• also a chinese ZVS induction heater board from an earlier stage of experimenting

If needed, I can also order more parts via the school (it still somewhat counts as a school project).




The project did not include a driver or anything similar yet. The plan was apparently a simple series LCR that would be driven at a fixed frequency, without feedback or power control.

It seems like various changes are needed to arrive at a nice high-powered result, and I have a rough idea of where to go after going through the forums and looking at other projects.

For now, I did a bit of a cleanup on the setup and added the snubbers and some bleeder resistors. Then I connected everything as series LCR and ran some low power tests using my UD+ with GDTs etc, together with a simple interrupter with no duty cycle limit. The MMC currently consists of 60 Wima caps for 13.2uF total.


At 15% duty cycle it pulls 20A / 12V, and slowly heats up the object. The resonant frequency is 31kHz. After a few minutes the heat sink and the black wires were slightly warm, the rest stayed cold to the touch.

Here scope shots from the test run:


Unfortunately it seems like my gate drive cannot do CW operation - at 40% duty cycle the gate resistors get pretty hot already, even though right now I have two in parallel (2W 5.6Ohm each). I tried removing the TVS since some people reported similar issues resulting from their added capacitance, but could not perceive any improvement.



Here my gate waveforms (could be slightly faster, but currently I do not have smaller gate resistors available):



What do you think is the best way to use these parts? Here some of my thoughts and questions so far:

• Any tips for how I can achieve CW operation without melting my gate resistors? So far this is the most pressing issue for me. Is it normal that they get that hot?
  Could the long wires be an issue (I did not cut them since they are supposed to go into my DRSSTC project), would lowering the gate resistance be enough, or maybe lowering the voltage to something like 15V?
  If nothing helps, I might switch to the IXYS gate driver modules. I think I should be able to route the FPGA output signals to the driver modules instead of the onboard driver stage. What seems annoying though is that the driver modules expects 15V, which means I have to make some changes on the UD+. Also it has a derived 3.3V supply for a controller which is supposed to make things simpler, but since the UD+ has its own 3.3V supply, I apparently have to deactivate that and feed the driver with my own supply instead (observing some startup and shutdown sequences).
• As for power control / current limiting, I was thinking about using the DRSSTC driver features and go with the UD+ builtin freewheeling / pulse skipping. Do you see any issues with that?
• How far do you think I can push the Wima MMC? Right now with 60p = 13.2uF the MMC is rated for 168Arms. I will keep within the 250V AC rating as long as I do not cross 600A, and frequency-dependent datasheet AC voltage rating (80V-ish at 30kHz) while staying below 200A. Downside of the large capacity is that it makes current regulation more difficult, I am also aware that I might have to rebuild it for better cooling once things start to become hot.
  Also, would a different MMC configuration be better (lower capacity / higher voltage)? I can source more Wima caps from the school if needed.
• To achieve real power I will have to add some method of impedance matching, though for now I want to get the gate drive working first. I have been going through the forums reading up on LCLR / transformer coupled topologies, and right now I think of going for transformer coupled. Not only does it provide galvanic isolation, if my understanding is correct the UD+ is also not suited to drive LCLR, since I would have to stay slightly above the resonant frequency with that topology.

I hope I can build something nice and have lots of fun with this project.

Thanks a lot for you feedback!

Edit: Btw is it possible to resize images in my post so that they take less space?

[davekni]

I tried removing the TVS since some people reported similar issues resulting from their added capacitance, but could not perceive any improvement.

These ~600W TVS diodes typically have ~1nF capacitance, insignificant compared to Cge.

Any tips for how I can achieve CW operation without melting my gate resistors? So far this is the most pressing issue for me. Is it normal that they get that hot?
Could the long wires be an issue (I did not cut them since they are supposed to go into my DRSSTC project), would lowering the gate resistance be enough, or maybe lowering the voltage to something like 15V?

Driving a brick CW requires a lot of gate power.  Yes, hot resistors is normal.
Lowering to +-15V will help by (15/20)^2, cutting power almost in half.  Even at +-15V, there is ~4.6W to dissipate (5uC * 30V * 31kHz).  Lower resistance does not change that 4.6W total.  If low enough, will push some of that power back into UD+ output FETs.  And 2.8 ohms is already under spec'ed minimum external resistance of 3.1 ohms.  Likely doesn't matter for low bus voltage.
Long wires (high GDT leakage inductance) increases gate power only if high enough to cause Vge overshoot.  Scope waveforms show no issue there.

I'd suggest lowering to +-15V and accepting 4.6W (using more or larger resistors).  Another option is to reduce to +15/-0V (or +20/-0V) with something like this circuit on GDT secondaries:
    https://highvoltageforum.net/index.php?topic=2389.msg17547#msg17547

How far do you think I can push the Wima MMC? Right now with 60p = 13.2uF the MMC is rated for 168Arms.

Since this isn't going to run 24/7, I'd push caps.  Might get away with 2x RMS rating if caps are spaced out a bit for air flow and fans added.

Downside of the large capacity is that it makes current regulation more difficult

Why is current regulation more difficult?

[markus]

Thanks for your reply!

I will mod my UD+ so it can run from 15V and get some beefier gate resistors.

Your GDT buffer circuit looks nice, but if needed I will try to get the IXYS gate drivers to work since I already have them. The nice thing with these is that they run from a single 15V supply and derive their isolated supplies themselves. I will keep it in mind though for future projects!

Why is current regulation more difficult?

I meant that with large tank capacitor current ringup will be very fast, making it harder to precisely set your desired current limit with an OCD (kinda like a super low impedance DRSSTC). It might not matter that much since we do not go for extreme currents anyways.

[davekni]

Your GDT buffer circuit looks nice, but if needed I will try to get the IXYS gate drivers to work since I already have them. The nice thing with these is that they run from a single 15V supply and derive their isolated supplies themselves. I will keep it in mind though for future projects!

The IXYS gate drivers should work well.  My GDT buffer does not require any external power.  Vge rising edge is provided by GDT secondary.  No buffering on rising edge.  Falling edge is pulled to 0V with local FET.  Requires no extra power supply.  Makes falling edge faster and limits Vge to 0V rather than negative.  Latter change reduces required gate drive power due to 0V minimum.  Theoretically to 25% (down by 75%), but about 30% in practice.  Even though unbuffered, rising edges are a bit faster because other GDT secondary windings (ones turning off) are only lightly loaded.

I meant that with large tank capacitor current ringup will be very fast, making it harder to precisely set your desired current limit with an OCD (kinda like a super low impedance DRSSTC). It might not matter that much since we do not go for extreme currents anyways.

Thank you for the explanation.  Makes sense.

Good luck with your project.

[Mads Barnkob]

Not a bad project to restart, good parts all around and not a huge rat nest

Water reservoir is however too small, even with a radiator, a good old floor bucket can sustain some run times of like 15 minutes and its up to bathing temperatures. Reuse of cooling water is then possible in the shower

There certainly is something "off" with the high capacity. I think you are looking at some unfavorable Q factors.

That MMC will also burn to the ground. You do mention that you need to cool it better, but I call for a total reconstruction.

You should have space between each individual capacitor, in both directions. Best way to achieve a MMC that can be cooled is to use small cuts of copper or brass plate/band to solder them together with. Film capacitors get rid of their heat from 2/3 through the leads and 1/3 through the package. The small copper/brass plates will work as heat sinks and with space between all caps, forced air cooling can flow between them.



I would also not trust that distance between the high voltage side (between L and C in resonant circuit), even worse if you go up in current with a small capacitance MMC.

[markus]

Good point about the isolation distance! I bent the leads apart from each other just in case.
The current setup is mostly just to get a feel for how everything works anyways and to experiment with stuff like current limiting, or different MMC capacity.
For proper operation I plan to completely reconstruct the MMC anyway to something water-cooled (what I meant with "better cooling").
But I plan to do that at the same time as I am setting up a proper water-cooled tank circuit, together with a transformer for impedance matching.

There certainly is something "off" with the high capacity. I think you are looking at some unfavorable Q factors.

Yeah, current rises pretty fast even at 12V.  Actually the original project had an even higher capacity (19.8uF), I already removed one of the cap bank modules.
So does this mean I would be better of with a lower capacity, for example half? A quick test shows that I end up around 40kHz with that.
Downside is that my Arms will also be cut in half...  The unfortunate thing is that even though they are free, this specific choice of caps makes building a properly dimensioned MMC a bit awkward. Either I end up with high Arms but high capacity, or with more reasonable (?) capacity and low Arms. For smaller MMCs the voltage rating also starts to become a problem.
This means I will have to go with a series combination to keep current Arms, but that makes the build more troublesome.

Meanwhile I did more tests with one of the 123A 12V PSUs, and got some usable results already!
Depending on the load, tank current ranges from 200A to 350A, after reaching the curie point the current grows in excess of 400A. The current setup is too weak however to do much heating beyond the curie point. Right now current limiting is set to ~360A (using freewheeling), which should still be fairly safe for the IGBTs I assume.
At 50%-ish duty cycle I approach 1.5kW input power, which is the limit of my variac (I plug the PSU in there because it has nice voltage / current meters).

MMC stays pretty much cold to the touch so far, but the IGBT heatsink started heating up so I added a bunch of fans for cooling, which should also provide a bit of airflow to the MMC. I also added a water bucket for cooling the work coil.
The bus caps are only rated for 30A at most, so I added the remaining two to get 13.2mF at 120A rating. At 1.5kW input power I should have some >100A current (my 10mm2 wiring to the PSU gets pretty warm already), and since I do power control via PWM (using the interrupter), I assume the caps should see a fair bit of stress.
Btw - is it better to connect one of the supply wires to the other end of the "U" instead for better current sharing between the bus caps? Or is this a bad idea since I will get a huge loop = adding lots of inductance? Would it be an issue for the PSU if I completely removed the bus caps (since I am using an interrupter)?

Here the current setup:


And here some waveforms. First with medium load and the second shows current limiting with freewheeling:


Next I plan on doing some tests with higher duty cycles, some 24V tests, and different MMC sizes.
Then I will try to convince the school I need material for upgrades to tank circuit / MMC, and impedance matching.


edit: I forgot to mention that switching to 15V really improved the temperature of the gate resistors. They still get kinda hot with higher duty cycles, but the larger resistors I ordered will help with that hopefully.

[davekni]

So does this mean I would be better of with a lower capacity, for example half? A quick test shows that I end up around 40kHz with that.
Downside is that my Arms will also be cut in half..

Higher frequency generally requires less current to deliver same power to load being heated.  However, that depends on load.  And you can easily increase current by supplying higher bus voltage.  Especially for experimenting before adding impedance matching, higher frequency will be closer to reasonable impedance.

edit: I forgot to mention that switching to 15V really improved the temperature of the gate resistors. They still get kinda hot with higher duty cycles, but the larger resistors I ordered will help with that hopefully.

Higher frequency increase gate resistor power proportionately.  That is one down-side.  But gate power resistors are relatively inexpensive.  Or change to a buffered circuit.

IGBT heatsink started heating up

Expected.  Switching losses are low with ZCS and low bus voltage.  Vce forward drop is the dominant loss, tracking current, not voltage.

Btw - is it better to connect one of the supply wires to the other end of the "U" instead for better current sharing between the bus caps? Or is this a bad idea since I will get a huge loop = adding lots of inductance?

Unlikely to make any significant difference.  Most AC (ripple) current is between IGBTs and caps.  Inductance to supply is not very important.

Would it be an issue for the PSU if I completely removed the bus caps (since I am using an interrupter)?

Probably would be a problem.  Most PSU output capacitors are intended to handle ripple current generated within PSU, not huge ripple current from load.

[petespaco]

Just being a curious bystander-----

I don't understand why the tank current would rise "after the temperature reaches the curie point".  What ferrous metals are you heating at that time?
In my humble (ZVS  induction heating) experience, current DECREASES as ferrous metals in the work coil reach the curie point.

I notice a graphite crucible sitting beside the work coil in one picture, but that isn't ferrous, so "curie point" wouldn't apply.  Right?

Do I understand that your input voltage to the IGBT's is only 12 VDC for all these tests?  Why not more?

Pete Stanaitis
---------------

[davekni]

I don't understand why the tank current would rise "after the temperature reaches the curie point".  What ferrous metals are you heating at that time?
In my humble (ZVS  induction heating) experience, current DECREASES as ferrous metals in the work coil reach the curie point.

Yes, same load change at Curie point causes opposite power change.  ZVS drive (parallel drive resulting in fixed voltage across coil) consumes less power when load Q increases.  H-bridge drive in series with coil consumes more power when load Q increases.

I notice a graphite crucible sitting beside the work coil in one picture, but that isn't ferrous, so "curie point" wouldn't apply.  Right?

I was wondering the same thing.  Depends on crucible electrical conductivity and geometry and drive frequency.  Test may or may not have included crucible, at least I didn't notice that being specified.  If crucible was included, implies it's conductivity is too low to block most magnetic field from center.  Higher frequency (resulting in higher volts/turn for same magnetic field) will move more heating to crucible and away from iron in center.

[markus]

Thanks for your replies dave, they are very helpful!

Higher frequency generally requires less current to deliver same power to load being heated.  However, that depends on load.  And you can easily increase current by supplying higher bus voltage.  Especially for experimenting before adding impedance matching, higher frequency will be closer to reasonable impedance.

That is good to know, I will experiment with that a bit more. My main concern however was the current rating of the MMC, since less parallel caps = less current rating. Halfing the capacity would drop the MMC Arms rating from 168A to 84A, which seems to be kinda low - even more so with lower capacities.

I don't understand why the tank current would rise "after the temperature reaches the curie point".  What ferrous metals are you heating at that time?
In my humble (ZVS  induction heating) experience, current DECREASES as ferrous metals in the work coil reach the curie point.

My setup is pretty much the same as a DRSSTC (with super low impedance), with the work piece in place of the secondary. As long as I feed the tank with energy, I will get high current ringup. Meanwhile the workpiece acts as a load and takes energy out of the system. Higher load removes more energy = less current in the tank.

    I notice a graphite crucible sitting beside the work coil in one picture, but that isn't ferrous, so "curie point" wouldn't apply.  Right?

The mentioned tests were with some random piece of metal - a ball bearing, a fat bolt, a piece of iron pipe. The large crucible is too big to fit into this work coil, I would have to switch coils for testing. Since the point of the school project was to melt metals, the crucible will certainly see some use. It is sitting there because I wanted to try if it would not perhaps fit in by chance, but it touches the coil in multiple places.
I gave the smaller crucible a try though (empty), and it presents a very light load. At 12V / 50% duty cycle it takes ages to heat up and only gets to the point where it is glowing a bit.

Do I understand that your input voltage to the IGBT's is only 12 VDC for all these tests?  Why not more?

Yes, all tests were on 12V input. I will do 24V tests next - I already prepared two stacked 12V PSUs for that which you can see in the last picture - but I expect very fast current ringup, which might make current limiting a bit difficult. So I might have to play with lower MMC size and higher frequencies.



Meanwhile I managed to track down an issue that has been plaguing me during the last few days:
My driver would sometimes skip interrupter pulses while signalling some fault / ocd, and sometimes even lock up completely. I could track this down to cases with high load.
It seems that with high load, there is (almost) no current during the first half cycle, meaning feedback starts only in the second half cycle. The UD+ compares this with the startup oscillator and since the second half cycle is negative, it thinks the frequency is way too low. Therefore it raises a feedback error and disables the output for the rest of the interrupter pulse.
Here some scope shots, first normal operation and the second showing the issue. Pink is the OCD / fault output, blue the cleaned up feedback signal. Startup oscillator is set to 3 cycles for debugging, normally I use 1 (I probably do not need it at all):



I thought I would have to reduce the number of turns on the feedback / OCD transformers (careful not to blow up the burden resistor with CW operation), or modify the VHDL to skip the feedback check during the first half cycle.
But while writing this I figured I could also swap GDT output and feedback polarity, so the second half cycle turns out positive instead of negative. This would be the easiest solution, but probably the issue will come back to bite me once I forget about it in the future

[davekni]

Meanwhile I managed to track down an issue that has been plaguing me during the last few days:

Looks like you need a bleed resistor across H-bridge output.  That way startup will be consistent.  Without resistor, leakage current of IGBTs defines initial bridge output voltage.  If output happens to be at same level as first half-cycle drives it, no current is generated.
Yet faster startup can be achieved if two resistors are used to bias initial H-bridge output to either + or - Vbus.  However, that requires checking which polarity first half-cycle drives H-bridge, then adding resistors to bias output to opposite polarity.

I gave the smaller crucible a try though (empty), and it presents a very light load. At 12V / 50% duty cycle it takes ages to heat up and only gets to the point where it is glowing a bit.

Any non-magnetic load is likely to be less effective (higher Q), so require higher voltage (higher current and/or higher frequency).

[petespaco]

I gave the smaller crucible a try though (empty), and it presents a very light load. At 12V / 50% duty cycle it takes ages to heat up and only gets to the point where it is glowing a bit.

Pardon my butting in on this thread since I have such limited experience.
But---   
With my 2500 watt Chinese induction heater running on 48 volts (at, I'd say, 100% duty cycle)  I'd draw about 35 power supply amps with a crucible of about that size in the work coil..  Subtract about 5 amps for what I call "idle current",   and I have about 30 X 48 = 1440 watts going into the crucible.  Within a couple of minutes, it starts to glow and within 10 minutes it is glowing a very bright orange, even with a couple of hundred grams of copper scrap in it.
So----  you are at about 1/4 of that input voltage and only half the "on time", for about 180 net watts into the work,  so I am not surprised that you don't see much heating going on.  If I were you, I'd get that supply voltage up as much as you can.

What is your goal for melting metals?
-Ferrous?  Non-Ferrous?
-Weight?   100 grams?  1000 grams? 
-Melt time to pouring temp?   I minute?  1 Hour?

Are you measuring power supply input current?  I'd suggest that you do that so you can get an idea of overall circuit efficiency.

[markus]

Thanks for your replies everyone! I kinda forgot to post, but did make progress in the meantime.

What is your goal for melting metals?
-Ferrous?  Non-Ferrous?
-Weight?   100 grams?  1000 grams?
-Melt time to pouring temp?   I minute?  1 Hour?

Thanks for your input and the numbers! What we plan on melting:

• Various aluminium parts - my friend wants to try and pour metal parts for his projects. For now at most a few hundred grams I think, using a graphite crucible.
• My sister and me will be doing comparatively small stuff with high level of detail, for example jewelry and ornaments. So aluminium, silver, steel, maybe even gold. My sister does a lot modelling and pouring in a dental lab, so we have professional expertise on hand. She is pretty hyped about the project, and even managed to get a bunch of used small graphite and ceramic crucibles from the lab. Of course she also found various issues with our primitive / naive methods  (her usual requirements are of course perfect quality in all aspects, but she is willing to make some compromises).
• Since my sister works with steel a lot, being able to melt steel would definitely be nice as well.
• And of course melting some random stuff for fun.

For the melting time I think something like 10-ish from the top of my head. However this will probably be skewed by the fact that my sister wants to preheat the filled crucible, and of course the form too. Edit: Apparently only steel is preheated

Any non-magnetic load is likely to be less effective (higher Q), so require higher voltage (higher current and/or higher frequency).

I'm trying to wrap my head around why higher frequency is more effective on nonmagnetic loads where we do not have hysteresis losses - is it because higher frequency leads to a higher voltage across the work coil?

Looks like you need a bleed resistor across H-bridge output.  That way startup will be consistent.  Without resistor, leakage current of IGBTs defines initial bridge output voltage.  If output happens to be at same level as first half-cycle drives it, no current is generated.

Thanks for the tip! I added a 220R 15W resistor and with that the issue is gone. I cannot believe I forgot about that since I definitely remember reading about that kind of issue before on the forums.

As for other changes, I upped the input voltage to 24V and then 36V and changed the MMC from 13.2 uF (1s60p) to 3.3 uF (2s30p), increasing the frequency to about 66kHz with the small coil.
With this, I now get 2-2.5kW input power. I can measure input current on the 36V output with the help of my multimeter which has a current clamp.
I also replaced the gate resistors with 2x 3W and added a fan, with that gate resistors are ok even in CW mode at 66kHz.

I now get some pretty nice heating on all kinds of scrap metal, with the pieces that have good coupling reaching the melting point (pipes etc). Smaller stuff only glows orange at most.
Here's a 2.5cm iron pipe and a ball bearing:


With this I also get my crucibles to heat up nicely and can melt aluminium. However since I only have a temporary setup, MMC and IGBTs start to overheat after some time, which forces me to abort. Pretty sure the heat sink is just too small for 300-400A. Meanwhile on the MMC the new setup leaves me with just 3 wires in parallel, which is just too little copper area - so the wires heat up even though the caps themselves would remain cold.


With all the experimenting on the current setup and a bunch of reading, I feel that my understanding of the system has improved a lot. Since the temporary set up has reached its limits, it is time to start with the real setup. My goal is to reach 10kW if possible.

• For the matching transformer I found some 80/40/15 ferrite rings made from K4000 material that I plan to buy (https://www.kaschke.de/wp-content/uploads/2017/06/K4000.pdf). 8 of these should give me enough cross section to not saturate even at a worst case of 565V / 8 turns / 20kHz. Before I actually buy them, do you see any issues with these cores? For example the material being wrong, or the inner diameter being too small etc.
• For the DC blocking cap I plan on getting a number of small MKP DC-Link caps (B32676E3126K) for about 200uF and >= 100Arms rating. They are rated for 300Vdc, but that should not be an issue since they will only see a few volt at >20kHz.
• For winding the matching transformer I have 0.5mm diameter magnet wire (liberated from an old TV) which I will use to create makeshift litz wire. This should be just right given a skin depth of 0.26mm at 60kHz. I just hope I have enough...
• I plan on running the inverter from rectified 230V. Full wave rectified 3 phase would be better, but the IGBT modules are only rated for 600V, with is way too close to the 565V they would see. Perhaps I will replace them with some smaller 1200V modules I have lying around.
• Of course I will also redo the tank circuit so everything is water cooled, including MMC and the secondary of the impedance matching transformer.
  I plan to use either 12mm stiff (MMC) or 8mm soft copper pipe (matching transformer secondary - a leftover from my DRSSTC), but need to find some suitable fittings for connecting everything together.
  I will add some more caps in parallel to the MMC so I can get a higher Irms rating and reduce the voltage across the tank capacitor - from 2s30p with 3.3uF now to at least 2s60p for 6.6 uF at 800V DC / 500V AC rating. This will drop the resonant frequency with the small coil from 66kHz to less than 50kHz.
  Or I could use all the caps and go for 3s60p, resulting in 4.84uF at 1200V DC / 750V AC for a better voltage rating.

[Anders Mikkelsen]

Congratulations on getting it working. As you've seen, not having a matching transformer leads to a lot of losses in the IGBTs, given that it restricts you to using a pretty low bus voltage in order to keep the power at a reasonable level. This will be even less favorable with more difficult loads, since the series resonant tank will draw maximum current when the loading is the lightest (highest Q), while a ferromagnetic steel tube will present the lowest Q of any practical load.

Your IGBTs are also pretty oversized in terms of rated current. I would aim for some more modern 1200 V parts of lower rated current, and operate it off rectified 3-phase mains. A full bridge running from 560 V bus would only need to push some 25 - 30 A to deliver your target power, so using 400 A parts just leads to uneccessary drive power requirements.

For the matching transformer, 0.5 mm wire is technically within a few skin depths of the operating frequency, but as long as you have more than a single wire then you also need to consider proximity effect. For this frequency, I would go with 0.1 mm or finer litz, but 0.2 mm is likely also fine, especially if you use a single layer winding on your toroid.

The ferrite should work fine. At these low frequencies, any power grade of ferrite should be fine.

Your caps are rated at 2.5 A RMS a piece for 15 degree internal heating. If we aim for the best practical external cooling, we can get away with 60 degrees heating, so double the current, that's 5 A per cap. .22 uF is 15 ohms at 50 kHz, which gives you a cap voltage of 15*5 = 75 V RMS at the maximum rated cap current, or 5 A * 75 V = 375 VA per cap, for a total of 34 kVAR for the 90 caps, independently of how they are configured. Right now it also looks like you have some stray inductance in your tank circuit which will eat up some of those VARs as well, by the ratio of work coil inductance to total tank inductance.

With a steel tube in the coil, you might end up with a Q as low as 4, which allows you to transfer 34/4 = 8.5 kW while staying within the ratings of your capacitors. For solid steel well matched to the size of the coil, Q might be 5 - 8, up to a few times that for smaller work pieces. For aluminium workpieces, a Q in the 20 - 50 range is not unrealistic, and for copper it can exceed 100. It follows that the power throughput will be very limited. I would definitely look for a larger capacitor if you're aiming for 10 kW. I would aim for at least 300 kVAR for a heater of this size, to give some leeway for processing a usable fraction of the rated power across a range of loads. Where are you located? A proper induction heating cap doesn't need to be particularly expensive.

[davekni]

I'm trying to wrap my head around why higher frequency is more effective on nonmagnetic loads where we do not have hysteresis losses - is it because higher frequency leads to a higher voltage across the work coil?

Yes, higher work coil voltage, which translates to higher volts/turn at load too.

With a steel tube in the coil, you might end up with a Q as low as 4, which allows you to transfer 34/4 = 8.5 kW while staying within the ratings of your capacitors. For solid steel well matched to the size of the coil, Q might be 5 - 8, up to a few times that for smaller work pieces. For aluminium workpieces, a Q in the 20 - 50 range is not unrealistic, and for copper it can exceed 100.

Anders:  Any idea what Q range would be for graphite crucibles?  I'd expect Q to be well lower than for directly heating copper or aluminum.  If I understand correctly, Markus' plan is to use graphite crucibles for melting.

[Anders Mikkelsen]

I'm trying to wrap my head around why higher frequency is more effective on nonmagnetic loads where we do not have hysteresis losses - is it because higher frequency leads to a higher voltage across the work coil?

Yes, higher work coil voltage, which translates to higher volts/turn at load too.

With a steel tube in the coil, you might end up with a Q as low as 4, which allows you to transfer 34/4 = 8.5 kW while staying within the ratings of your capacitors. For solid steel well matched to the size of the coil, Q might be 5 - 8, up to a few times that for smaller work pieces. For aluminium workpieces, a Q in the 20 - 50 range is not unrealistic, and for copper it can exceed 100.

Anders:  Any idea what Q range would be for graphite crucibles?  I'd expect Q to be well lower than for directly heating copper or aluminum.  If I understand correctly, Markus' plan is to use graphite crucibles for melting.

I just did some quick measurements with a tight fitting coil around a large crucible, the best practical case, and it goes from around 3.5 at 30 kHz to 5.5 at 100 kHz, so not too bad. This is practically with no clearance between the coil and crucible, so I would target maybe 5 - 8 Q at 50 kHz to give more flexibility in the crucible selection in practice. For a project of this cost and magnitude, I would not skimp on the tank VARs, and the numbers I gave for the FKP1s are really pushing the limit. It's also nice to be able to heat aluminum directly in ceramic crucibles, and steel above the curie point, since my experience is that graphite crucibles oxidize away with time if no protective atmosphere is used. I would still stand by my recommendation to use a purpose-made induction heating cap, given that a 300 kVAR part costs less than 100 euros from normal distributors, and 500 kVAR can be had for less than 150. This gives a lot more VARs per dollar (or euro) compared to FKP1s, with easier mounting, cooling and lower stray inductance to boot. I have  a good stock of new Celem 500 kVAR caps that I got for 25 dollars each, and I would be fine with donating one to the cause if Markus is in Europe, given the cost of shipping here.

I'm trying to wrap my head around why higher frequency is more effective on nonmagnetic loads where we do not have hysteresis losses - is it because higher frequency leads to a higher voltage across the work coil?

Yes, higher work coil voltage, which translates to higher volts/turn at load too.

Hysteresis loss is not usually very significant, and the large change in Q when steel goes above the curie point is mainly from the drop in permeability. Reflected resistance is proportional to the root of the ratio of resistivity to permeability. Hysteresis loss might represent some 15 - 30 % of losses, but the permeability can easily change from 1000 to 1, giving a loss ratio of sqrt(1000/1) = 30 between being below and above the curie point.

Inductive reactance of the coil, and therefore VARs per amp, rises linearly with frequency. Workpiece deposited power grows with the square root of frequency for a given coil and current, due to the skin depth decreasing with sqrt(F), so you get more heating per amp at higher frequency with a given coil and workpiece, but less heating per VAR. More coil turns gives more heating per amp, while preserving VARs, if the geometry is kept the same, but there's a practical upper limit to how many turns you can have. Since most film caps come with a 500+ V rating, it makes sense to dimension the tank to benefit from that.

[markus]

Hello Anders,
thank you very much for your replies, that is some very useful information there!

You are right about the parts - however, for now I worked with these because I got them for free from this school project (they also got them basically for free from somewhere else).

Your IGBTs are also pretty oversized in terms of rated current. I would aim for some more modern 1200 V parts of lower rated current, and operate it off rectified 3-phase mains. A full bridge running from 560 V bus would only need to push some 25 - 30 A to deliver your target power, so using 400 A parts just leads to uneccessary drive power requirements.

You are right about the IGBT modules - while drive power is not that much of an issue since I already solved that part, not being able to run from rectified 3-phase mains is not that great. I have quite a number of 150A 1200V BSM150GB120DL IGBT modules sitting around, switching to these would be simple for me.

Your caps are rated at 2.5 A RMS a piece for 15 degree internal heating. If we aim for the best practical external cooling, we can get away with 60 degrees heating, so double the current, that's 5 A per cap. .22 uF is 15 ohms at 50 kHz, which gives you a cap voltage of 15*5 = 75 V RMS at the maximum rated cap current, or 5 A * 75 V = 375 VA per cap, for a total of 34 kVAR for the 90 caps, independently of how they are configured.

Similar to the IGBTs, I have been using the caps since they were already there from the school project.
That kVAR calculation and the Q numbers were very informative, thank you! Up to now I always found it difficult to figure out just how much power I can / will transfer with a given setup.
Actually, there are 180 caps in total (half of these are still at the school), so I could get 68 kVAR total instead of 34. However that is still rather far away from 300 kVAR unfortunately, so it would probably be best to see if I can get a proper capacitor instead of building a MMC that would be too small anyways. Not to mention the other benefits.

the numbers I gave for the FKP1s are really pushing the limit.

I have to admit that I already exceed the Irms rating by a lot and pushed the current to 350A for the current 2s30p MMC, even though the caps are only rated for 2.5Ax30 = 75A. This makes for about 90kVAR at 66kHz / 3.3uF. Even so, I did not have the impression that the caps were getting dangerously hot.
Edit: Actually those are peak values, not RMS - so in RMS I this is "only" 250Arms, and only half the VARs for 45kVAR.

But I think you are right about the induction heating caps, and it seems there are some that actually come at an acceptable price (I always had imagined them to be more expensive).

300 kVAR part costs less than 100 euros from normal distributors, and 500 kVAR can be had for less than 150

What kind of "normal distributors" do you use, or what kind of parts should I be looking for? I only find more expensive prices using mouser / digikey / a search on octopart, with the CDE 300kVAR LC3 series (https://www.cde.com/resources/catalogs/LC3.pdf) at 120-130 EUR/pcs net, and the 500kVAR LC5 series at 270EUR. Only the Dawncap 500kVAR DDC caps (http://www.dawncap.cn/upload/20220423151314.pdf) - a dropin replacement for the Celem C500T - on aliexpress are cheaper at ca. 100EUR/pcs. For celem I could not find a price at all, seems like I would have to request a quote.
However it seems that actually getting one of these is a bit difficult, since most of the CDE are not in stock anywhere and the Dawncap aliexpress store (https://www.aliexpress.com/store/1102804025) seems to sell to every country except for Germany where I live, which is a bummer. No luck on ebay either.

I have a good stock of new Celem 500 kVAR caps that I got for 25 dollars each, and I would be fine with donating one to the cause if Markus is in Europe, given the cost of shipping here.

Wow, 25 dollars is a great price! Thank you for your offer, I would be very happy about that! I live in germany, I will send you a PN.

For the matching transformer, 0.5 mm wire is technically within a few skin depths of the operating frequency, but as long as you have more than a single wire then you also need to consider proximity effect. For this frequency, I would go with 0.1 mm or finer litz, but 0.2 mm is likely also fine, especially if you use a single layer winding on your toroid.

I had assumed 0.5mm was fine since this guy in his 60kW induction heater (https://forum.mosfetkiller.de/viewtopic.php?t=64870) used 0.5mm wire (20mm2) and was fine with it. But I can see if I can find something else - I have a one pound reel of 0.12mm (AWG36) and more than a kilogram of 0.3mm wire from my tesla coils, as well as some misc salvaged stuff so there should be something suitable.

If I understand correctly, Markus' plan is to use graphite crucibles for melting.

It's also nice to be able to heat aluminum directly in ceramic crucibles, and steel above the curie point, since my experience is that graphite crucibles oxidize away with time if no protective atmosphere is used.

Yes, I will use a graphite crucible, but I also have some small-ish ceramic crucibles I want to use. Being able to melt steel would definitely be nice.
Btw the graphite crucible slowly oxidizing away apparantly also serves as a feature, since this produces a layer of protective gas covering the metal. This seems to be the reason why in the dental lab they use graphite crucibles for stuff like gold instead of ceramic. For materials like iron / steel you cannot use graphite crucibles however, since the carbon will poison the metal.

Meanwhile I also ordered the ferrite cores and caps for DC blocking.
At this rate there will be not much left of the original project when I'm done, just the bus caps, bus bars and work coils Oh and of course the extra stuff like the crucible, modelling sand etc.

[markus]

My setup got some improvements and now runs from 230V AC (single phase)!
I switched the IGBT modules to 150A 1200V BSM150GB120L and added a quickly built matching transformer and a 192uF 300V DC blocking cap. (This means the threat title is now wrong lol)
Matching transformer is 3x R80/40/15 cores with 7 turns of 2p 2.5mm2 I had lying around. At 330V bus voltage and 7 turns the cores will not saturate as long as I stay above 40kHz.
DC blocking cap is 16x B32676E3126K MKP film caps. I noticed later on that these caps are DC link which are made for filtering DC ripple, will I run into problems using these for DC blocking?
Bus caps got reduced back to 2p 3300uF / 350V. For 3 phase operation I will put these in series to get 1650uF / 700V.


IGBTs run much cooler now, matching transformer also remains cool so far. Since the tank cap has not changed, I still stay below 400Apk, but I added more cooling to the MMC.

I think I need to tweak phase lead a bit more, but I need to change the inductor for that.
What is also missing is an inrush current limiter, for now I use the variac for startup.