Quote from: afk on August 23, 2017, 05:09:29 PM

Quote from: kamelryttarn on August 23, 2017, 04:11:05 PM

That setup is basically my end goal

Wouldn't it also be possible to control the gates via some kind of logic gate circuit and mosfet driver circuit?

I think that it should be possible. The second video in petespaco's post was using a temperature controller to control the gate via a general purpose relay and a contactor. It is just I don't know if I can separate the supply for the gate and the supply for the LC circuit, with LC circuit being on voltage all time as I switch on/off the gate. Theoretically it shouldn't pose any problem if the latching resistors just work out well. There is also the problem with the gate being switched on/off too frequent, and the surge at every ON.

I have used separate supply for gates and power electronics, with shared ground, that can be done.

However I would not turn off gate drive, unless by turning off you mean shutting it down, but still powered up to keep it at zero Volt, only then would I dare having the power electronics supply on all the time, if you turn the driver supply off, you do not know the state of the switches.

Since it is only the gate drive is being turned off (B1 and B2 above being cut off from the circuit) while the rest (ground, power electronics) is still on, I believe that the 10kOhm pull-down resistors in the circuit will bleed the gate voltage down to ground quite quickly. The supply on LC circuit can stay on voltage (like 60V) while the gate signal being cut off via a general purpose relay.

Frankly, it is possible to get a contactor to turn off the power electronics as well, but a contactor is a lot more expensive than a general purpose relay. I also need a RC snubber that is more robust (hence more expensive) for a contactor, since the current surge and arc are much stronger. Instead of controlling a lot of current, just turning on/off the gate with little current would be more preferable. The only question is whether the cutoff is stable and there is no latching (or latching happens but mostly negligible before the pull-down resistors kick in). Also the question of current surge with a workpiece inside the coil when the gate signal is ON again.

It is also interesting to note that a temperature controller has a delay time of a few seconds before it can engage the general purpose relay. This delay time can serve well to stabilize my power supply once I activate the power supply.

Interesting video. I just don't know if the guy separated the gate input and the coil input? I plan to use the temperature controller for only the gate signal while keeping the coil input on (the other end of the two chokes on voltage). Since the gate signal is low, the MOSFETs are OFF, thus no current will conduct. This way I can get a general purpose relay at low voltage, low current to turn on/off the gate voltage. A general purpose relay is a lot cheaper than getting a contactor for higher current rating. I just don't know if I can leave the coil on voltage while cutting off gate signal. Theoretically there shouldn't be a problem but I can't really be sure of everything in practice.

It is also necessary to wire a RC snubber circuit in parallel with the contacts of a relay/contactor to suppress arc. The first relay in the video got welded because of that, and also due to the transient current surge I think.

Also, thanks for the banggood. It looks like the prices are a lot cheaper than what I could find here, only at around 50% without counting shipping fee.

Hello, afk.
  You said that "The machine also heated up immensely."  Tell us more about this.

-Mosfet or IGBT heatsink temperatures
-Tank capacitor temperatures, and, did the temperature vary much between capacitors
-Temperature of the cooling water.  Talk about your method of moving the cooling water through the system and what you are doing to cool the water
-temperature of anything else that got hot.
-Did any of these temperatures stabilize, or did they keep climbing?

Tell us about current flow changes as the metal heated toward the curie point.

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

My mechanical design is temporary so there are a lot of drawbacks when it comes to surveillance of the machine... Measuring temperature is quite hard. I'll get down to the details once I overhaul the design again.

On the other hand, both MOSFETs and tank caps were below 90°C in the video... On extended usage that will be quite a problem, but so far there isn't any degradation yet. Work coil, however, could reach 200°C so there is a dire need for water cooling... I'm gonna buy some distilled water next week along with some graphite crucibles. I plan to freeze a part of water so that it can keep the cooling water at good temperature.

Regarding crucibles I'm thinking about DIYing some steel crucible instead. Steel can reinforce the heating before Curie temperature but the problem is to find a steel can with correct shape. Else, graphite crucible should be good enough.

Thank you for sharing all the oscilloscope measurements, it all looks very clean, stiff and stable. I might just be wrong or my own Royer was much worse, but I seem to recall that the gate waveform was not as nice as yours.

Have you tried heating a large diameter water pipe? Just to try something with a larger diameter to get a better coupling to the work coil. I think that bolt is simply too small to be effectively heated in such a big coil.

The graph might not reflect correctly what is going on, actually. I used average sampling to get the measurements. The signals weren't this stable and were fluctuating quite a bit. The efficiency increases if I put a bigger workpiece, that is pretty much no-brainer already. I just wanted to test the machine with that bolt. I'll do with a bigger object but I have to make the machine neater first.

IIRC Mads or somebody said that 10000µF or so should be able to support 20 amps... Since I'm aiming for 60 amps I went all out for the filter. Having more doesn't hurt for me, since I can't afford having strong spike at start that cause the voltage to drop too much. I got the 1000µF/450V for $1.53 each... I can't find any high voltage cap that has higher capacitance.

Big ripples shouldn't be much of a problem I think, as long as the voltage is well above 12V (iirc the minimum working voltage, I once tested a ZVS heater at 12V before). The issue is that the starting spike can cause a big voltage drop that can latch both MOSFETs and short the drain to the ground, risking a big blow (I once got this which dropped my 20V to 8V and fried up my circuits). The transient current can damage your cap bank when you start the circuit, which is also a factor. If you are using tri-phase transformer it shouldn't be much of a problem since the initial ripple of the rectified input is fairly small compare to mine, but you should calculate a bit.

I planned to use 470 Ohm for the gate resistor but I couldn't find any at 5W at the store, so I decided to get 630. Apparently it works at 20V and obviously it still works at 60V, which is my application. The gate resistor is to limit the current charging the gate, so having it small can overload the gate. It depends on your MOSFETs IMO, but I played it safe and chose 630.

I'm disassembling the circuit to tweak my coil. I'll try to take gate signal when I finish.

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.

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.

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.

Quote

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.

"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.

Since I'm still looking for filter cap and better toroid for the two chokes, I'm posting the schematics of the driver circuit:



I'm using double MOSFETs schematics so that it can support higher current consumption. I have tested its functionality before with 12V and IRF250N before and it did work, but just for confirmation whether it needs some tweaks.