Royer induction heater

[afk]

"but I'm aiming the freq to fall between 70-200kHz. It is used to melt copper and silver for metallurgy. Since I'm working with powdered metal, that frequency should be enough.":
    This will be interesting to me, since the non ferrous metals that I have attempted to heat have done VERY POORLY so far with my Chinese 1000 watt device at anywhere from 43KHz up to about 113 Khz.  I can only get about 1 amp of NET current into a piece of 1/2" OD copper tube.  Same for a 1/2" X 5/8" solid tin bar.
So, at 48 volts input, that's only about 50 watts.
   I do see that people are melting small quantities of solder and aluminum, but they are using graphite crucibles.  In that case, I think it's the crucible that gets hot, imparting its thermal energy to the work.  There is one guy, "The Radio Mechanic", who seems to be successfully heating a small  pot of solder using a ceramic crucible.  He even has a cheap PID controller and thermocouple setup to control the induction heater's output in an on/off mode.
  From what little I have read, non ferrous metals need much higher frequencies and a LOT of power.

I understand that so I'm planning to use a metallic crucible (or any heat-resistant crucible with added metallic parts) to transfer heat instead.

Both copper and silver have melting point way below metal so this should be doable.

Still, even for metal when it passes its curie temperature the heating efficiency drops. For metal it is around 700-800°C or so, IIRC.

High frequency increases heating power, but the skin effect prevents eddy current from penetrating deeper for heating purpose. In fact, high frequency is only for heating up small objects and powdered metal.

There are also a few things to consider as well. For example, oscillating current can be calculated as I=V * Sqrt(C/L), and freq = 1 / 2pi * Sqrt(LC). To increase frequency, it is easier to drop C than L, but then, you will be reducing the oscillating current, in turn, reducing the heating effect.

[petespaco]

"oscillating current can be calculated as I=V * Sqrt(C/L)"
Thank you for this formula.  I have  seen  this phenomenon occur when I imbalance C and L from the "stock" values, to change Fres.  Now I should be better able to predict and optimize current flow.
  As you probably know, I have tested a couple of different graphite crucibles and they do heat up pretty well, as long as they are well insulated.  My main problem with them has been to match the ID of the work coil with the OD of the crucible to get enough current flow but not too much.   

[Uspring]

High frequency increases heating power, but the skin effect prevents eddy current from penetrating deeper for heating purpose.

That's true for a fixed coil inductance and current. A higher frequency will imply a higher voltage on the coil in this case, though. Often the constraints are different, e.g. the power supply will have a certain voltage and current capability or the transistors. For a given voltage and current rating, doubling the frequency e.g. implies halving the tanks inductance and capacitance. That will keep tank voltage and current unchanged.

Under this constraint, i.e. keeping f*L constant, there is a sweet spot for the frequency for the max power transfer. It's basically an impedance matching issue between the samples inductance and resistance. The frequency is lower for more conductive materials and for larger objects.

[afk]

High frequency increases heating power, but the skin effect prevents eddy current from penetrating deeper for heating purpose.

That's true for a fixed coil inductance and current. A higher frequency will imply a higher voltage on the coil in this case, though. Often the constraints are different, e.g. the power supply will have a certain voltage and current capability or the transistors. For a given voltage and current rating, doubling the frequency e.g. implies halving the tanks inductance and capacitance. That will keep tank voltage and current unchanged.

Under this constraint, i.e. keeping f*L constant, there is a sweet spot for the frequency for the max power transfer. It's basically an impedance matching issue between the samples inductance and resistance. The frequency is lower for more conductive materials and for larger objects.

Actually the oscillation/tank voltage is almost fixed. According to Mads, the peak-to-peak is pi times the source voltage, so if you use 60V, it should be around 188V peak (it is a criterion for choosing the MOSFET, since IRFP250 can deal with 200V, a safe value for our usage). Frequency will shift depending on the work coil and tank cap. The only thing that will influence the tank voltage is the voltage drop via parasitic impedances through out the circuit as the consumption current increases. Voltage drop that occurs with transformer is also another thing. However, in ideal PSU and ideal circuit, tank voltage shouldn't change. At least, I have confirmed that through a few experiments (I succeeded in making a weaker version of the driver, btw).

I believe most of us already know this, our circuit is in fact a LC oscillator with a charger that supplies the oscillator energy. In ideal situation, the LC circuit will keep oscillating without losing energy. In such case, the charger will not supply any energy at all, so the current consumption is zero. However, our situation is not ideal. There is always some loss throughout the circuit, like parasitic resistances that heats up when the oscillating current passes through them. The eddy current is also one of the losses that occurs, though this loss is intended. Energy loss of the LC oscillator will be resupplied by the charger, and there we can see the current consumption. To calculate this consumption we need to calculate the heating power of eddy current, but this is pretty hard and there is no general formula to calculate it (as I searched on the Internet). We only know that eddy current depends on a few things:

• The frequency of the changing magnetic field (which is also the oscillation frequency of the LC circuit).
• The strength of the oscillating magnetic field (which also depends on the intensity of the current in LC circuit).
• The object that needs heating - shape, material, etc. This is the biggest variable in the equation. It also influences the skin effect and stuffs.

Mostly we work around on frequency and the tank current, but it is pretty much "hit-or-miss" at least for my case as I don't work in magnetic and related subjects, so I'm pretty much a noob in eddy current.

Although it is easy to say "halving the tank inductance and capacitance", it isn't easy in practice. Work coil is the most constrained, as it has to have a certain size for the intended purpose. When you make the work coil with copper pipe for water cooling as well, it is even harder to fine-tune the coil.

I haven't thought about keeping f*L constant, as it isn't easy to modify L, but your input gives me some thought. I'll try to calculate it out first, though.

[Uspring]

I believe most of us already know this, our circuit is in fact a LC oscillator with a charger that supplies the oscillator energy. In ideal situation, the LC circuit will keep oscillating without losing energy. In such case, the charger will not supply any energy at all, so the current consumption is zero.

I'm not familiar with this particular circuit. I believe, that the transistor will still carry current if there is no dissipation. But this current does not seem directly related to the tank current but only to the max expected power supply load. So in principle one could have an arbitrarily high current in the tank without overloading the transistors. So how would you optimize? By maximizing the ratio between power in the sample and power loss in the working coil itself? That seems to favor large frequencies. The effect flattens off at some frequency.

Edit: I thought a little about this circuit. It seems, that transistor current in the unloaded case just depends on the choke inductance (and voltage and frequency). So it is also not directly related to the tank current. There is one thing I don't understand: The max voltage on the transistors seems to be pi*Supplyvoltage in the steady state case, i.e. a constant load. But in a transient state, e.g. supply voltage is applied suddenly, there seems to be a large overshoot of up to twice the steady state value.

[afk]

I believe most of us already know this, our circuit is in fact a LC oscillator with a charger that supplies the oscillator energy. In ideal situation, the LC circuit will keep oscillating without losing energy. In such case, the charger will not supply any energy at all, so the current consumption is zero.

I'm not familiar with this particular circuit. I believe, that the transistor will still carry current if there is no dissipation. But this current does not seem directly related to the tank current but only to the max expected power supply load. So in principle one could have an arbitrarily high current in the tank without overloading the transistors. So how would you optimize? By maximizing the ratio between power in the sample and power loss in the working coil itself? That seems to favor large frequencies. The effect flattens off at some frequency.

Edit: I thought a little about this circuit. It seems, that transistor current in the unloaded case just depends on the choke inductance (and voltage and frequency). So it is also not directly related to the tank current. There is one thing I don't understand: The max voltage on the transistors seems to be pi*Supplyvoltage in the steady state case, i.e. a constant load. But in a transient state, e.g. supply voltage is applied suddenly, there seems to be a large overshoot of up to twice the steady state value.

I'm not familiar with ZVS driver either, so I just generalize it to simpler circuits that I understand the most. I see that there is an oscillation with a LC circuit, so I immediately characterize it as a LC oscillator. The FETs' role is like a charger, supplying this oscillator energy that got lost from heat (both from parasitic resistances and from eddy current). In ideal case, since no energy is lost, and the LC circuit is charged at maximum, it draws no energy from the charger. The mechanics behind is simple as that.

With the same generalization I can also calculate the current flow in the circuit, by calculating the possible energy can be stored as magnetic field and electric field, both in ideal case:

Welec = CV²/2 and Wmag = LI²/2

Energy of magnetic field in the coil and energy of electric field in the tank capacitors will periodically exchange into each other (hence the oscillation), so both Wmag=Welec. Expressing this further, I = V * Sqrt(C/L).

This calculation does not take account of the parasitic resistances, however. So the actual current flow will be less than that theoretical value. To calculate it you have to solve the differential equation of second order for RLC circuit, with R as parasitic resistance being an unknown variable.

This current, obviously, only flows within the LC circuit, and will not affect the FETs. Something like 200V and 200A+ flow would destroy any FETs before it can even oscillate: The functional ZVS driver shows that this current is independent from the FETs.

The best way to optimize is to cut off the parasitic resistance as much as possible (to reduce unneeded heat loss and also current drop), and choose the frequency at the right value depending on the material and the shape of your workpiece. The second is from the book - I'm not that knowledgeable in the field so excuse me for that. According to Wikipedia:

Frequency (kHz)	Workpiece type
5–30	Thick materials (e.g. steel at 815 °C with diameter 50mm or greater).
100–400	Small workpieces or shallow penetration (e.g. steel at 815 °C with diameter of 5-10mm or steel at 25 °C with a diameter around 0.1mm)
480	Microscopic pieces

There are also a few sources about induction heating as well, but I can't seem to find them again. Sorry about that, though.

As for the transient state, you have to charge up the LC circuit. There will be a great power spike seen at the PSU, which is typical, but in the actual LC tank, the voltage should be rising up gradually.

[Mads Barnkob]

One of the highest losses are properly proximity effect, to the adjacent turns in the work coil, together with the resistive losses from a low skin depth these two are the primary reasons for water cooling, maybe to some degree also the heat radiated from the work piece

I did describe the proximity effect / Medhurst factor in my chapter on Tesla coil secondary design and this can easily be transferred to the work coil: http://kaizerpowerelectronics.dk/tesla-coils/drsstc-design-guide/secondary-coil/

I thought I had more papers on induction heating, but there should be some good data to find one materials and frequencies in these old documents: http://kaizerpowerelectronics.dk/tools/file-archive/?drawer=application_notes*induction_heating

[Uspring]

The best way to optimize is to cut off the parasitic resistance as much as possible (to reduce unneeded heat loss and also current drop), and choose the frequency at the right value depending on the material and the shape of your workpiece.

Those might be conflicting goals. Lowest resistance from skin effect in the work coil implies a low frequency. That might not be best frequency for the work piece. A single figure of merit would be nice.

As for the transient state, you have to charge up the LC circuit. There will be a great power spike seen at the PSU, which is typical, but in the actual LC tank, the voltage should be rising up gradually.

I believe the tank voltage overshoots in some situations. Let's first consider the steady state case, i.e. constant amplitudes of voltages and currents:

Say, the current in one of the chokes (it doesn't matter which one, both have the same form) is the same for every tank cycle. Then the average value of of the voltage across the choke must be zero. A non zero average would imply a steadily rising current after each cycle. In equations:

integral Vchoke * dt = 0 (integral over a cycle)

If the cycle has length T, that amounts to two components:

Vsupply * T/2   +   (Vsupply-Vtank*2/pi) * T/2

For one half of the cycle, the FET is conducting, for the other half not. Putting the above sum to 0 we get

Vtank = pi * Vsupply

Now consider the case, that the tank has no energy, is not loaded and then the power supply is turned on. Since Vtank is near 0 during the first cycle, the V*dt integral won't be 0. After some cycles Vtank will have grown, so that the integral will become 0. During all this time the average choke current (averaged over a cycle) will have risen. It will stop rising at integral = 0, but it won't be 0 at this point. There will still be power fed to the tank, so its voltage will increase. The integral will become negative and average current will decrease but it will only stop rising, when the average current has fallen to 0. All this works out to an oscillating tank voltage envelope. Does this make sense?

@Mads: Interesting pdfs, but so looong.

[Mads Barnkob]

Uspring, from the measurements I did on my Royer ( http://kaizerpowerelectronics.dk/general-electronics/royer-induction-heater/ ) the switch voltage and tank voltage was the same, is this because the Vtank calculations for the LC circuit reflects its voltage envelope onto the power source?

[Uspring]

Hi Mads,

The closed FET switch grounds one side of the tank, so the other FET sees the whole tank voltage, so tank and FET voltage are the same. Under "Considerations" in the link you say, that the switches see pi times the supply voltage. That makes perfect sense to me for steady state operation. I'm assuming here the idealised situation, that the DC choke resistance is 0.

The one thing, that worries me is the transient situation of abrupt power up. With the idealised assumption of no losses, it seems, that tank voltage can rise up to 2*pi of the supply voltage. That would kill a FET specified only to pi times the supply voltage. My results are from theoretical deliberations only. I've tried a LTSpice simulation on a simplified circuit to confirm this, but my circuits don't want to oscillate. Do you have a working simulation?

[afk]

So I have been quite busy recently to continue on the project... But here I am, with a little progress!



I'm making a better tank cap bank using the X2 caps (as I have bought them, it is a waste not to use). It is hard to solder them directly on 0.5-mm-thick copper sheet (the heat dissipates too fast even with 100W soldering gun to have a focused heating point for soldering), so I get them on a PCB first, then bolt the two copper sheets on each side of the bank. I hope that the electric current will flow well through the bolts - I'm thinking of doing the same to the filter cap bank. The copper on the PCB is too thin so I'm thinking of reinforcing the flow with additional - and thicker - copper sheet, which also helps dissipating heat (as it did with my soldering attempt).

This tank cap bank consists of 11x 0.33µF, resulting in around 3.66µF measured.

My old tank caps are here, using 16x 0.22µF WIMA caps (3.52µF measured):


I'm redoing the circuit for now. I don't know if I should allow the current to flow through the heatsink of the FETs, but through my experiments before it does improve the current flow a lot. I'm thinking of coating the heatsink with a bit of paint after bolting the FETs on the heatsink so that it will prevent accidental shorting. There will be a lot of energy here so I want to avoid sparkling a bit.

[petespaco]

re: "I'm thinking of coating the heatsink with a bit of paint":
  Isn't it the job of the heatsink to take heat away from the Mosfets?  If so, then the LAST thing I'd do is to cover the heatsink with some coating.  You will probably want to fan-cool the Mosfets anyway, so just place the fan over the heatsinks so you can't get to them to short them out.

[afk]

re: "I'm thinking of coating the heatsink with a bit of paint":
  Isn't it the job of the heatsink to take heat away from the Mosfets?  If so, then the LAST thing I'd do is to cover the heatsink with some coating.  You will probably want to fan-cool the Mosfets anyway, so just place the fan over the heatsinks so you can't get to them to short them out.

I'll be covering the whole metallic parts with a thin layer of paint after I have connected everything intact. I'm thinking of using both watercooling and fancooling for it.

It is scary to leave big heatsink in air with voltage.

[Mads Barnkob]

You have some good surface area on the MMC to take away some heat from the capacitors and also low losses due to minimized skin effect. Good that current sharing is also taken care of by insuring a almost even current path from connection points to all capacitors.

It is better for the heat transfer from MOSFET to heat sink that it is mounted directly onto the heat sink without insulating pads, so I just mounted the heat sinks with acryllic or other plastic material to keep them floating with each MOSFET. I would also advise against painting the heat sinks, unless its with a special heat conducting paint, I have seen that used in some motor drives, but I doubt it has any insulating properties.

[afk]

You have some good surface area on the MMC to take away some heat from the capacitors and also low losses due to minimized skin effect. Good that current sharing is also taken care of by insuring a almost even current path from connection points to all capacitors.

It is better for the heat transfer from MOSFET to heat sink that it is mounted directly onto the heat sink without insulating pads, so I just mounted the heat sinks with acryllic or other plastic material to keep them floating with each MOSFET. I would also advise against painting the heat sinks, unless its with a special heat conducting paint, I have seen that used in some motor drives, but I doubt it has any insulating properties.

Is that so? Hmm... I'm torn, really. I also prefer mounting the MOSFET directly onto the heatsink without insulating pads, firstly to ensure good heat dissipation, and secondly, to use the large surface area of the thermal pad on the MOSFET to conduct electric directly to the LC circuit, bypassing the aluminium heatsink which has large cross-section for conducting electric. The issue is that the heatsink will be exposed to the air which is quite dangerous if there is an accidental shorting - and as a zealot of Murphy's Laws I prefer to avoid such risks as much as possible.

[afk]

So I'm back again!

After one month and a half that is way too busy to spend more time on this project, I finally finish this. The induction heater is below, heating up a piece of door hinge. The image next to it to remind you of my circuit, using double MOSFETs like the last time.



This time I am using the following capacitor bank as tank. 11x0.33µF for 3.63µF measured capacitance. Surprisingly, the MKP-X2 caps work fairly well without heating much (I'll talk about the consumption later on) despite my abuse for a few dozens minutes. I soldered the caps on a PCB and complement them with thick copper sheets (0.5mm thick).



I replaced the old two chokes with bigger chokes and obviously, thicker wire (3 mm in diameter). I forgot about their inductance but they are all above 200µH (the previous two are 1mH, so I was worried if these new chokes couldn't do. Well, they work).



Big improvement this time, I attached and soldered all components directly onto each other without using a PCB. I also used 12V/5W Zener (the black component soldered between the gate and source) instead of the old 2W to ensure their robustness. FR107 is still working fine as fast diode - there is no problem with it being a bit slow. Gate resistors are also 630Ohm/5W (the big white component I leave hanging in the air). The big thing with glassfiber tape and electrical adhesion tape is the ground line - I use 10x2.5mm² square wire to minimalize parasitic resistance and to increase heatsink.



Since the last time the filter cap bank was insufficient, I more than doubled the amount of capacitors. I also removed the 50mH filter choke as advised by Mads - mainly I think the choke will not be working well at high current and to reduce wasted heat. This capacitor bank has 35 capacitors of 400V/1mF. Since the capacitors are of low quality, I'm not having 35mF but it should be at least 30mF.



And voilà! I have a good result on my oscilloscope. Yellow and Blue are the two ends of the tank cap/work coil, and the Red/MATH is the difference of potential a.k.a tank voltage. Btw I can't seem to measure the tank current despite having used two coils in cascade for a stepdown current transformer.



We have f = 45.77kHz and tank cap is 3.63µF, so it is easy to calculate the inductance of work coil. I calculated it to be 3.33µH, which is quite a big improvement from the old coil (only 1.14µH). This work coil has 6 turns with 10cm in diameter, using 7mm-diameter copper tube (previous is 5 turns, 5cm diameter, same tube). I ran out of copper tubes else I had made more turns.



Measurement time for no workpiece. I used 23V input and the input voltage only drops to 21.92V. The input current is 2-2.13 A (approx 43-45W). A great reduction from the previous 16A (in the OP). It seems that direct soldering on components and using bigger wire as ground greatly improve efficiency. Tank voltage peak is 70V (or 148V on the oscilloscope) which is roughly pi*Vin. At this time I felt like a fool for expanding the filter cap so much. But well, the spike at start should be quite powerful so being careful isn't so bad.

When I put the door hinge into the coil, the consumption increased to 6.93A. The input voltage dropped to 21.06V for approx 146W consumption. Every measurement on the oscilloscope is still fine.



I also took time to measure the heating of the work coil when without any workpiece. The graph below shows the evolution of temperature based on time. I haven't flowed water in the coil tube yet, but after using the induction heater for more than 10 minutes to heat 5 bolts to more than 300°C (the maximum that my multimeter can measure) the temperature of the work coil is still under 70°C.



I'm still looking for method to increase the output. It can reach 10A but it is still low. I'll test the machine at higher voltage later to see if it can fare well.

[Mads Barnkob]

Great update on the project!

I love the absolutely over-engineered parts like 25mm^2 ground connection, its better to make things sturdy and never have failures

While you trippled the inductance of the new work coil, compared to the old, you would also limit the current to 1/3 of before. I also think that the diameter is possibly too big compared to the work piece, so there is simply not enough coupling to transfer energy. You need to have work coils that fit pretty close to the work piece.

Higher input voltage will of course help on the performance.

Regarding your LC circuit current transformer that does not work, did you add a load resistor across it? A current transformer merely transforms current in a ratio, f.ex. 1:50, but you need a burden resistor to have a measurable voltage across it, figured out by Ohms law to the voltage that you need for your control system/meter.

[afk]

I already tried to put a 0.5 Ohm resistor (actually 2x 1Ohm in parallel) as shunt but there is barely a signal outside. Actually, I don't know where I should latch the ground of the oscilloscope on. The current transformer is pretty much on air.

As for increasing the size of the coil, I think that it will be fine. My intended workpiece will be big enough. I leave more space in order to avoid direct heat loss from the workpiece to be transferred to the work coil. Tripling the inductance only reduces the oscillation current by sqrt(1/3) according to my calculation, as impedance of inductor is Lω where L is increased by 3 and ω is reduced by sqrt(1/3) from inductance increase. I will tweak around its size later when I get my crucible.

I am thinking of wrapping alum foil to reflect heat from the workpiece. It should help reducing loss and overheating the work coil. Since the surface of the alum foil will be in parallel with the magnetic field inside the coil there shouldn't be a problem of the foil heating up via eddy current.

A bit off-topic, I'm also buying a thermocouple to measure temperature. I found a K-type for less than $10, but there is a risk of green rot damaging the thermocouple.

[afk]

So I tested the circuit at 60V and... darn the heat was immense.

Without putting a workpiece the work coil heated up quite fast. I measured it to be 8-9.3A consumption (supply at 62.15V) so the loss is already at 500-600W, which is huge. This is rather troubling as I expected the loss to be increased nine-fold (from around 43-45W 22V). I hope that when I add water cooling it can help mitigating the loss from increased resistance due to heat.

Other than that, other components heated up quite fast as well, especially the MOSFETs. The choke also got a bit warm. I will need something to dissipate the heat from the choke as the enamel of the wire could get degraded.

This is what I got from the oscilloscope, though. As usual, yellow and blue are the two sides of the coil and the red one is the tank voltage:



Heating up 1 bolt, the current raised to 11.07A. Heating up the door hinge again, it was 21.73A (supply dropped to 59.22V).

The video below shows the heating of 8 steel bolts. Supply current read at 33.33A and supply voltage dropped to 57.33V. It took around 30s to get red hot. 1900W is already enough for me to use, but at least 600W got wasted (actually more since there are more current flowing through parasitic resistance), which is not really desirable.

By the way, I have a question concerning water cooling. Do I have to distill water for cooling? It is a bit troublesome to distill water. Somebody said that tap water contains some sodium or calcium minerals that can cause electrolysis within the copper tube and corrode it. Since it is an alternative current at a few dozens kHz I doubt electrolysis might happen, but I just want to ask just in case.

[RocketScienceSmurf]

Very impressive induction heater! When I see your cap bank I am a bit concerned if my bank will not have enough capacitance. I couldn't afford and fit more than 1900uF so I am expecting a bit of ripple but I can't see how that would be a big problem. Just like you I plan to use beefier zener diodes, but isn't 630 Ohm gate resistor a bit much? I had planned to use 100 Ohm instead. Have you looked at the gate signal with your oscilloscope?