low-cost 1800-amp heating source

Interesting to see this discussion migrate to the design of magnetic amplifiers, or mag-amps. In the applications where I have seen mag-amps used they were phased out a long time ago, alas due to just plain silly classification of certain programs I can't discuss those applications. But I found a manufacturers description at

which shows both the parallel and series configurations you have been discussing here as the two standard configurations. There are/were other configurations in use with for example multiple control windings for control from multiple independent inputs, and the rectifier version already mentioned.

Typical 1940's application ran the mag-amp control winding at a fairly high voltage from a tube amp, then ran the DC output from the mag-amp to the generator field of a big motor-generator set which provided power to the load. State of the art for fast, accurate positioning of large objects at the time.

[....]

It looks like my series case has enough faults to lose out tot he parallel case. You could feed the control current through an inductor so that the ripple voltage doesn't get to the control circuit.

I have been thinking about using a "light dimmer" style circuit to make the control current. Since the ripple from noe hits its maximum just at the zero crossing, it will change the timing of when the core saturates. In the current design, the cores saturate late in the alternation. This would shift it even later making the power factor worse. It would be nice if the control circuit instead made the power factor better.

Just for fun I was also thinking about placing the tranformer that drives the control circuit in parallel with the saturable reactor. Unfortunately this runs into trouble at the full blast end. It does have a nice local feedback effect that would make it easier to control the voltage at the load more accurately. Perhaps a small transformer at the load to sense the load voltage would be a better idea.

This saturable reactor idea has the nice feature of good isolation between the control side and the power side.

Ok Tony, agree that amp*turns must be satisfied. So if we had the luxury to choose the ideal core materials we would choose a material with high flux density (B) to reduce N and low coercive force (H) to reduce I with no air gap. I am going with Supermenour. Thanks, Harry

[snip]

There was a DC output mag-amp called the Ramey Amplifier. A quick google on that produced on the first page a very interesting Swedish paper, about controlling the field of a JAS-39 generator with a 3-phase Ramey amp. Quite an in-depth treatment, including the calculation of core losses and photographs of stacked toroidal inductors with the control winding wound around the stack. Early 1990's apparently.

--
Tony Williams.

I'll back away from my sweeping statement that when in paralleled primaries the control current should be from a high AC impedance.

The reason for that is that the measured AC current in the 10000uF does not exceed 0.4Arms and that only adds 128mW to the dissipation in this control winding (from 3.2W up to 3.328W at 2Adc control current).

Having the 10000uF then makes a PWM'd control current much easier because the inductor no longer has to have a high impedance at 50/60Hz.

[snip]

For 50W of control power into a low resistance load I suspect it has to be a PWM'd field coil driver type of circuit.

Vsupply ---------+----------+-----+ | | | | ) | | Rsec1) | _|_ ) |+ /_\\ | ===10000uF | ) | | Rsec2) | | ) | | | | | +-----+ | | | |) | |)Inductor | |) | | +----------+ |--+ PWM--> ------|| |--+ |

Nice paper, thanks.

Mag amps are also often used as switches for applications like SCR gate drive where isolation is required, and apparently for SMPS output regulation:

I think they might be used more if more people understood them. Hey Win, how about another addition to your next AoE edition? (Said without checking your exixting book for mag-amp info only because it seems to have wandered away from the bookshelf where it belongs. My failure to remember such is hardly conclusive. :-)

Glen

That's him, thanks. About a 4.5meg download.

That paper also discusses Harry's point about the best magnetic material to use.

--
Tony Williams.

I assume you are willing to waste 128mW. I think you may have jumped from this to the idea that it is better to waste the power. I disagree as I will go into below.

I disagree with this I will explain after the modified drawing.

Your version for reference of those reading:

My version:

---------- Vsupply ---------+----------+-----! Down ! | | ! shifting !---IL ! \\ ! circuit ! ! / ! ! ! \\ ! ! ! ! ! ! ! +-----! ! ! ! ---------- | ) | Rsec1) _|_ ) /_\\ | | ) | Rsec2) | ) | | | ! | | | |) --!!-- | |)Inductor ! ! | |) +-/\\/\\-+ | | ! ! +----------+ Iset-/\\/\\+-!+\\ ! |--+ ! >-+---------|| IL--!-/ |--+ | GND

The real inductor blocks the bulk of the AC from getting to the control windings.

I loose more power in the sense resistor but it makes the control circuit much easier. By removing the capacitor on the windings we end up controlling the actual current in the control windings and not the voltage on them. This mostly takes the pole from the inductance of the windings out of the servo loop. The PWM shows a very high impedance to the control windings.

I assume you have a feedback from the temperature of the heater to the Iset. This will have a large phase lag. The mass will integrate and the distance from the heater to the thermistor may add a transport delay. The last thing you will want is another couple of poles in the control circuit of the saturable reactor.

Yes. Both your and my designs are basically regulators on their own.

The reason I backed away from a high impedance drive was the realisation that the bottom end of that inductor is always at a low impedance point, either 0V or Vsupply.

This means that, irrespective of the PWM frequency, the inductor has to have a high enough value such that it presents a high impedance to the 100Hz ripple voltage. That's 'high' relative to the source impedance of the ripple voltage.

I did a crude attempt at measurement of the source impedance, measuring the change in the pk-pk volts as a result of changes in resistive loading of the control winding.

At 0.5A, 1A, and 1.5A DC control current, the output impedance of the ripple was around 1.5 to 1.6 ohms.

The inductor's impedance has to be much higher than 1.5 ohms. The waveshape of the ripple is not a sinewave it is saturation difference spikes at 100Hz so the inductance need not be as high as first thought, but it will probably need to be in the mH region, (at 2A dc polarising current).

BTW: The source resistance seen by those two primaries is the 590 ohm load and I am attracted by the coincidence that 590 ohm transformed by two 240:18V transformers in parallel would present 1.65 ohms on the secondaries. That's probably yet another bizarre TW-ism though.

BTW2: The voltage across each secondary is of the order of 55V pk-pk, and the resultant difference ripple is around 2.5V pk-pk maximum. So they are not doing too bad a job of looking at each other whilst one of them is going into a region of low relative permeability.

I had thought that the transformers effectively shorted each other out so the impedance looking back into the control winding would be mostly resistive. Worth a quick check then.

The input impedance was measured at 50Hz and 5KHz with a signal generator and scope. Calcs suggest the control input looks like 1.4 ohms and 92uH in series.

1.4 ohms is near the sum of the secondary and transformed primary resistances (1.3R), and I suspect that 92uH is something like two leakage inductances in series.

That's getting far too complicated for me atm. :)

--
Tony Williams.

I think this is an error. Observing my PWM system: If the current in the induct attempts to change from the desired set point, the comparitor turns the MOSFET on or off. This means that for low frequencies and small signals, the impedance is extremely high.

You didn't show what was hooked to your current sense so It may or may not also be true of your design.

The big question is whether the situation is "small signal" or not.

[....]

I am slightly surprised that the impedance of it is so high. I had figured it would be mostly the resistances of the transformers.

[.....]

That may be it. The 100Hz depends on the AC current in the primary causing saturation. Imagining putting a zero AC impedance between the control windings, I see that changing the current in the load.

This means that there is a path for the load's impedance to get translated into something we see at the control side.

That is low enough that this may really be "low signal".

I think this sounds right. That is quite a small time constant so it is not a problem for your servo design.

"Press on and the light will shine"

Sorry Ken this has been me being thick and I now see what you have been saying.

For ref: I had a quick look at the FFT of the spiky ripple at the control terminals a few days ago. The harmonics extended out to 5KHz, although they are generally 40dB down by 2.5KHz. That gives some idea of the required bandwidth of an inner current servo loop.

No I was locked in field coil mode, running the PWM stage in low gain open loop, controlled by the outer loop. I had not even thought of having a current-demand inner loop that was fast enough to take care of the ripple.

snip rest.

--
Tony Williams.

The simple minded comparitor sort of design will work if the ripple never causes us to get outside the "small signal" case where the pulse width is varying continuously.

The biggest problem in this area is that the loop can't stop the current. We could add a second power MOSFET and diode and make it so that we can tranfer power back onto the supply rail if we want to slow the current down. This would require a small dead band so that we aren't constantly engaging that part of the circuit.

If we don't need the extra circuits, the design of the control part is too simple to be very interesting. Even with it things aren't very complex. There are two comparitors with slightly different thresholds and a little positive feedback.

If I was doing it, I would have the positive feedback of each comparitor also go to the input of the other and have one extra path in the feedback I will new describe:

With the "increase the current" comparitor goes to the off condition, a series RC from its output reaches over to the "decrease the current" comparitor's inputs. This differentiated signal prevents the "decrease the current" comparitor from acting for a brief time. This gives the resistance of the control circuit a little time to reduce the current before the extra MOSFET is activated for the purpose.

This extra MOSFET for the "decrease the current" requires a level shifting driver. We can use an N channel device if we have the needed higher supply voltage. This higher voltage only needs to supply several mA so it wouldn't be hard to come up with.

Higher ----/\\/\\-----+----+----+----------------- ! ! ! ! /-/ --- ! !/ ^ --- --/\\/\\----+----! NPN ! ! ! !\\e ! ! ! ! !!---Vcc ! ! ! +-------!! ! ! ! ! !!---+-- Inductor ! ! ! !/e ! ! ! +----! PNP ! ! ! ! !\\ ! ! ! ! ! ! ----+----------------------+------------+--!