Help needed for Induction Heater circuit

[ramuna]

Hello friends,
I am attempting to design a 1 kW induction heater operating at 979Hz. Two IRFP250N mosfets are operating in Push-Pull. The parallel resonating capacitor of 33uF is on the primary side. The 2uH induction coil is on the secondary side. The Push-Pull transformer has a split primary and the primary to secondary turns ratio is 20:1. The power supply voltage is 60V DC. I have been simulating the circuit in LTSpice.

My simulation shows a sharp current spike on mosfet switch ON. My question is, how can I eliminate/greatly reduce this spike.

I have uploaded a zip file, which contains the LTspice *.asc file for the simulation, the *.spi file for the IRFP250N and a screendump of the simulation which shows the current spike. In practice I will be using a bank of 2 or 3 IRFP250s per split primary, but for the present I would like to know how to remove/reduce the current spike. Many thanks in advance!

NB: Please amend the spice directive giving the location of the IRFP250N spi file in the Inductheat_1.asc simulation file to match the actual location of the file on your computer.

 

[Lightium]

Add some resistance between your MOSFET's source and ground. Also, your inductors should have some series resistance.

 

[Diver300]

The Q factor of the tuned circuit is so low that resonance won't happen. The Q from the inductor is 2*π*L/R

That is 2 * π * 0.000002 / 0.5 = 0.025 which is so low that it can't be considered a tuned circuit at all.

An induction heater will be basically resonating with no object in range, and when an object is present, that will damp the resonance.

Some other points:-

I don't understand what the 0.2 H inductor L5 is supposed to simulate.

I think that the current spikes are just the charging and discharging of the capacitor. With the low Q, the transformer has effectively got a resistive load, which is the square wave and the spikes are on top of that.

 

[ramuna]

Thank you Lightium & Diver300 for your replies. I applied the comments made by Diver to my circuit, making the following changes:
(1) Eliminating the RF Choke L5
(2) Eliminating the capacitor C1
(3) Changing the load resistance R3 to 0.2R

With these changes in place I simulated the circuit in LTspice. The MOSFET current spike is now very brief and reduced. The power of the spike (current through mosfet x drain_source voltage) is a VERY brief spike of under 2kW. I have attached a screen shot of the simulation graphs to this post.

My question is, are these realistic results (see the power across the load resistance in the upper plot_plane) ?
And if I am making a mistake, please identify it.

 

[Diver300]

You don't appear to have any leakage inductance (https://en.wikipedia.org/wiki/Leakage_inductance) in the model.

I think that the inductance of the coils will need a lot of energy to magnetise, and if you don't do anything then that energy has to go somewhere. If there is nothing the heat, all of the coil's inductance is leakage inductance.

Switch mode power supplies normally have snubber circuits.

Card reader circuits are normally resonant.

I don't know what is normally done with inductive heaters but there is a lot of energy and the MOSFETs look like the most fragile.

All component models are wrong to some extent. The trick with using component models is using them in a way the the errors don't matter. I suspect that modelling the coil + item being heated with a perfect transformer, and inductor and a resistor is too simple for the complex situation you have, but I am not sure. I would absolutely not trust a model until I had been able to compare it with real components.

 

[ramuna]

Hello Diver,
"Card reader circuits are normally resonant"
So too are the transmitters in the cards - each being resonant to the same frequency increases the distance across which power is transferred. But in the case of induction heating, we are only concerned with generating eddy currents in the load. So inserting a piece of metal into the induction coil, will make the coil the primary of an air-cored transformer and the metal the secondary with one (shorted) turn. There is no real need for resonance, even though it is normally the case in many induction heaters. But at low to mid frequencies (eg the 50Hz induction furnaces in steel mills), the induction furnace is powered by simply applying mains electricity to the work coil.

"Switch mode power supplies normally have snubber circuits."

Please elaborate with reference to the circuit being simulated. What snubbers do you think are needed in this case?

"I think that the inductance of the coils will need a lot of energy to magnetise, and if you don't do anything then that energy has to go somewhere."

Yes this is true, and so is the saturation flux density of the core. But magnetising current is inversely proportional to the number of primary turns, and that is a separate issue, to the principle being investigated in the simulation. Meaning, that I can increase the primary inductance, and so long as the secondary inductance rises by the same proportion, the simulation result will be unchanged, even though the mag. current will fall.

"You don't appear to have any leakage inductance (https://en.wikipedia.org/wiki/Leakage_inductance) in the model."
I will have the transformer custom made by a transformer manufacturer, who will apply their specialised knowledge to minimise leakage inductance. For this reason I am disregarding it. 979 Hz, the operating frequency, is in the low audio range, which is a mass-market run-of-the mill region of operation for transformer makers (for audio transformers to match audio amps to speakers), albeit that my requirement demands much higher currents.

"I don't know what is normally done with inductive heaters but there is a lot of energy and the MOSFETs look like the most fragile."

I have seen Induction heating units on EBay and on Youtube which use two banks of 3 paralleled mosfets each, to handle 3kW of power

for example.

 

[Tony Stewart]

There are many designs with IGBT's using ZVS too with a low Cout advantage but higher Ron..

"But in the case of induction heating, we are only concerned with generating eddy currents in the load."

True. So the choice of f depends on skin depth <= material thickness for maximizing impedance of magnetic losses. Then minimize material thickness to raise temperature.

Raising f from 50 kHz to 500 kHz then allows you to heat Aluminum.

 

[Diver300]

"You don't appear to have any leakage inductance (https://en.wikipedia.org/wiki/Leakage_inductance) in the model."
I will have the transformer custom made by a transformer manufacturer, who will apply their specialised knowledge to minimise leakage inductance. For this reason I am disregarding it. 979 Hz, the operating frequency, is in the low audio range, which is a mass-market run-of-the mill region of operation for transformer makers (for audio transformers to match audio amps to speakers), albeit that my requirement demands much higher currents.

I was under the impression that L3 and L4 were the induction coil, and that the transformer formed by L3, L4 and L2 is the model of the coupling to item being heated.

Anyhow, L3 and L4 form a transformer and when one MOSFET turns off, any leakage inductance of L3 or L4 will lead to large voltages. The normal thing to do is to have a snubber circuit such as C1, D1 and R1 in this circuit:- https://i.pinimg.com/736x/a4/0a/6b/a40a6b3e02c5dd49832f346fb457ae7c.jpg

The circuit you linked to on YouTube doesn't have a transformer. The inductive spikes from turning off the current in L1 and L2 of that circuit are limited because the MOSFETs will only be turned off relatively slowly.

 

[Tony Stewart]

None of your design seems to be sensible. Perhaps you should define goals then specs 1st.
Is this heating a pan or a slug of iron? or what?
A self resonant ZCS oscillator would make more sense at a suitable f with real parts, not 10000 Henries and 63 Farads

 

[ramuna]

Thank you Tony & Diver for your remarks.

"None of your design seems to be sensible. Perhaps you should define goals then specs 1st.
Is this heating a pan or a slug of iron? or what?"

It is a coreless (ie crucible) furnace for melting a copper lead alloy. At 1kHz the skin depth for this material will be around 8mm. The material being melted will be a solid ingot.

"Raising f from 50 kHz to 500 kHz then allows you to heat Aluminum."

At 500KHz, the skin depth for aluminium is 0.115 mm. Such a depth would be EXTREMELY INEFFICIENT for melting the metal, since the melting would rely on thermally conducting the heat produced in this very thin surface layer (thinner than a human hair- nominal thickness of 0.18 mm), into the interior of the solid metal.It is for this reason that commercial induction furnaces for melting aluminium operate at medium frequencies (ie in the audible range of 1 - 4 KHz). Watch this video and listen to the sound that the furnace produces

"A self resonant ZCS oscillator would make more sense at a suitable f with real parts, not 10000 Henries and 63 Farads"

Kindly examine the circuit's component values again Tony. You above given values are very much in error.

Also, why would a self-resonant ZCS oscillator make more sense, bearing in mind the low inductance of the induction coil L1 and the need for a very large capacitance (40,000 uF), which can withstand both high voltage AND high current, to resonate with L1 at 1 KHz?

 

[ramuna]

Hello friends,
I have continued playing with this Non-Resonant induction heater design, making two major changes: (1) I changed the mosfet (2) I added a snubber to each mosfet to eliminate the ringing caused by the leakage inductance of the transformer primary and the output capacitance (Coss) of each mosfet.

I used the following reference for designng my snubber

https://e2e.ti.com/cfs-file/__key/communityserver-discussions-components-files/196/slup100.pdf

To this post I have attached a number of attachments, which are:
LTSpice circuit files for the induction heater without snubber
LTSpice circuit files for the induction heater with snubber
Datasheet for the IPP60R060P7 mosfet
Spice library file for the IPP60R060P7 mosfet (as IFX_P7_600V.txt file)
Screendumps made during running the LTspice simulations of the heater, both with snubber & without.

These screendumps are made up of
drain-source voltage across one of the mosfets (M1) operating WITHOUT snubber, showing ringing
expanded view of this Drain-source voltage ringing, so as to estimate ringing frequency
the drain-source current through mosfet (M1) operating without snubber, showing negative spike

the drain-source voltage across mosfet (M1) operating WITH snubber, showing reduced ringing
expanded view of this Drain-source voltage ringing with snubber, showing reduction in duration
the drain-source current through mosfet (M1) operating with snubber, showing negative spike

Using the display captured in the second of the screendumps listed above, I estimated the ringing frequency as 1.4MHz. The datasheet of the IPP60R060P7 gives its output capacitance Coss as 48pF. Using this and the ringing frequency (equivalent to omega^2 = 7.73777x10^13), the breakeven leakage inductance is 269.2uH. Using the expression for characteristic impedance given on the LHS of page 4 of the above linked snubber reference we have char. impedance = Sqroot(L/C) = Sqroot(269.2uH/48pf) = 2368 ohms (I used 2k7). The snubber capacitor is set at approx 4 x Coss or 200pF.

With this snubber design the ringing is markedly reduced, as can be seen in the screendumps.

PROBLEM: while the above snubber to a great extent solves the voltage ringing, I have a large negative current spike in the Drain source current, for both the snubbed and unsnubbed circuit. And despite various attempts, I am unable to eliminate/greatly reduce it. I would be grateful for ways and means to reduce this spike.

NB : PLEASE CHANGE THE SPICE DIRECTIVE IN THE LTSPICE FILES REGARDING THE LOCATION OF THE IFX_P7_600V.TXT FILE TO MATCH ITS LOCATION ON YOUR OWN COMPUTER.

 

[rjenkinsgb]

It looks like a "flywheel" pulse - as the opposite device turns off, the current continues and causes a voltage reversal in the inductor.

Try adding fast diodes from either end of the centre tapped inductor to the positive power source.
(The device being switched off will be seeing a positive voltage pulse; the diodes will give that a path other than the opposite device).

You can add low value resistors in series with the diodes to speed up the current decay.

 

[Tony Stewart]

Thank you Tony & Diver for your remarks.

"None of your design seems to be sensible. Perhaps you should define goals then specs 1st.
Is this heating a pan or a slug of iron? or what?"

It is a coreless (ie crucible) furnace for melting a copper lead alloy. At 1kHz the skin depth for this material will be around 8mm. The material being melted will be a solid ingot.

"Raising f from 50 kHz to 500 kHz then allows you to heat Aluminum."

At 500KHz, the skin depth for aluminium is 0.115 mm. Such a depth would be EXTREMELY INEFFICIENT for melting the metal, since the melting would rely on thermally conducting the heat produced in this very thin surface layer (thinner than a human hair- nominal thickness of 0.18 mm), into the interior of the solid metal.It is for this reason that commercial induction furnaces for melting aluminium operate at medium frequencies (ie in the audible range of 1 - 4 KHz). Watch this video and listen to the sound that the furnace produces

"A self resonant ZCS oscillator would make more sense at a suitable f with real parts, not 10000 Henries and 63 Farads"

Kindly examine the circuit's component values again Tony. You above given values are very much in error.

Also, why would a self-resonant ZCS oscillator make more sense, bearing in mind the low inductance of the induction coil L1 and the need for a very large capacitance (40,000 uF), which can withstand both high voltage AND high current, to resonate with L1 at 1 KHz?

The Alum. example was for thin pans.

 

[ramuna]

Thank you RJenkinsGB. I will try this out and report back - need to select a suitable diode, get its spice model etc first.
Thank you Tony:
"None of your design seems to be sensible. Perhaps you should define goals then specs 1st.
Is this heating a pan or a slug of iron? or what?"
Hence my brusque reply. Yes, if you had mentioned that you were heating a thin pan, it would have all made sense at the time. But its water under the bridge now. All the best!

 

[Tony Stewart]

Your model specs are unclear. Every passive and semiconductor component, conductor (Trace/wire/tube/connector) has RLC parameters.

T=L/R, Q=ωL/R, Zo=√(L/C) ω=1/√(LC) are important factors.
BW-3dB= 0.35 / Tr (10 to 90%) covers primary then there are higher edge harmonics.

Is this for conceptual simulation? or duplication of existing concept? The simulation at best, is only as good as the assumptions.

I suggested a self-resonant driver design but your transformer details are vague.

Applying snubbers to an artificial model is somewhat academic. You have an 80 Vdc supply and a 320 Vpp AC primary with conduction on alternating tap sides. Is this your best approach, or just the one you are copying?

When I look up Cu crucible Induction heaters, I see the large resonant capacitor I expected , possibly PU type. This would likely be 3ph, 208 to 400V, 15 kW to 200 kW using water-cooled coils.

We must do our best to communicate all assumptions.


Initially I said "
Perhaps you should define goals then specs 1st.
Is this heating a pan or a slug of iron? or what?

Thicknesses of 5x skin effect is reasonable but depends on the alloy's thermal conductance/velocity, vs electrical conductance from eddy current losses for thermal characteristics such as efficiency of kW/kg and time to melt. Magnetic stirring is also a benefit.

 

[ramuna]

rjenkinsgb - I have tried out adding fast diodes at either end of the split primary as you suggested, to remove the flywheel current spike. Unfortunately, this results in high currents thru' both the mosfets & diodes, with very little power across the load resistance R3. See attached screendump of the LTspice simulation.

I used the RF2001NS3D fast recovery diode for the sim. Please comment & advise.

Tony Stewart - I am currently examining various routes via LTspice simulations, for designing an induction furnace. Yes, doing so is not perfect, as many real world factors such as transformer leakage inductance etc are ignored. But it a start, to sift thru' various possibilities and test out ideas. Then when I settle on one route, I can include the applicable real-world factors.

As I mentioned in post #10, the furnace is for melting (lead rich) copper-lead alloys. Each melt will involve around 1.5kg (this is set by the capacity of the crucibles which I will be using). For this, I am aiming for skin depths of ~5mm. For the reasons which I mentioned earlier, melting metal needs much higher skin depth than surface hardening, heating saucepans etc.

Now, since you used copper as an example: because of copper's low resistivity, to get a skin depth of 10mm we need to go down to 43 Hz (so, around mains frequency). At such low frequencies, which ever practical inductance coil we use to enclose a crucible such as the ones I will use, will have a very low inductive reactance.
And hence the resonating capacitor will be very large.
A ballpark figure for the coil inductance is 2uH and so to resonate at 43Hz I would need a capacitance of 6.85F. This is completely impractical, even if we reduced it by a factor of 400 by having a 20:1 Primary to secondary turns ratio coupling transformer, and moving the capacitor to the primary side (6.85F/400 = 17125uF. We need this to be ideally polypropylene with very low ESR. Such a capacitance is very hard to practically realise).

Hence my preference for a non-resonant induction coil. The main reason that a resonant solution is preferred is to cancel out the inductive reactance of the induction coil. But in my particular circumstances, at a frequency of 1KHz and coil inductance of 2uH, there is very little inductive reactance to cancel out (0.0126 ohms).

Additionally, if I wanted to place the resonating capacitor on the primary side, how can I do so given that I have a split primary in a push-pull configuration?

 

[rjenkinsgb]

I have tried out adding fast diodes at either end of the split primary as you suggested, to remove the flywheel current spike. Unfortunately, this results in high currents thru' both the mosfets & diodes, with very little power across the load resistance R3.

My fault, I was in a rush and had not thought it through properly. The coil ends will be driven to double supply when the opposite end FETS switch on, assuming no resonance.

 

[Tony Stewart]

Have you considered that the heat spread velocity or thermal lag may be much faster than the electrical lag, Tau=L/DCR ? This means if you be more efficient to use the highest frequency that does not cause a huge thermal gradient from surface to interior. Just as melting metals with gas torch transfers heat from the surface and diffuses thru the full thickness I don't believe you need to target frequency to skin depth for excellent thermal conductors. This ought to raise f by several decades.

 

[ramuna]

Hi Tony,
Here are the thermal conductivities of the metals which we've discussed so far (all in W/mK):
Aluminium = 237, Copper = 413, Lead = 35. As you can see, for a lead rich copper alloy, thermal conductivity is pretty low. In fact, among the pure elemental metals, only manganese and mercury have lower thermal conductivities than lead (7.2 & 29 respectively).

Thermal Conductivity of Metals and Alloys: Data Table & Reference Guide

Thermal conductivities of common metals, metallic elements and alloys.

www.engineeringtoolbox.com

Next, as the metal's temperature rises, its resistivity increases and consequently its skin depth also increases. And since both thermal and electrical conductivities are positively correlated, as temperature rises, thermal conductivity will also fall (but not by as much, see the link above).

There is NO electrical lag. We penetrate the metal down to its skin depth right from the start. We get thru' instantly. This is why induction melting is faster than any other method of melting. To summarise: we instantly penetrate the metal down to its skin depth, this depth increases with temperature, the electrical energy transferred to the metal is immediately translated into thermal energy.

This is exactly why medium and low frequency induction furnaces exist, even to melt metals with thermal conductivities as high as aluminium and copper. Moreover, the ceramic crucible prevents heat loss, unlike the case of playing a gas torch on the surface of the solid metal. Yes, the metal will conduct heat inwards, but we will also lose more heat due to radiation than in the case of a crucible enclosed metal being heated by induction.

 

[Tony Stewart]

The interplay between skin depth and thermal diffusivity matters.

Let me find some values. ( are we assuming pulverized Cu mixed with pulverized Pb or large chunks? or what? and what mix ratio?)

Thermal diffusivity, α is a measure of how quickly heat propagates through a material per unit of depth over time, [m²/s].

Copper conducts better but slower to diffuse than both Al and Pb.

I imagined a copper tubing 2m in length to make the crucible induction heater with water cooling.

Next start with stock thickness of 5x skin depth and then scale up frequency to determine if oxidation or exciter losses becomes a limiting factor from too small skin depth or driver current from too much skin depth.

The furnace video sound matches up an 1800 to 1900 Hz square or sawtooth wave.