Measured INDUCTANCE of my welding reactor

I think you are overestimating the difficulties and underestimating the value of Spice simulations. True, simulations can lie, but you will be testing your circuit at low power and you can obtain some confidence in your spice simulation by verifying that it predicts the low power operation of your circuit, then simulate full power, then build the full power version.

This would be easier with a digital or storage scope in the situation where you are producing only a single event. In the old days before the invention of digital and storage scopes we used Polaroid oscilloscope cameras to capture single shots; they mounted on the scope faceplate with an adapter that excluded room light. The shutter was opened, the event being observed fired and the scope triggered from it, camera shutter closed and then film developed. I don't know if any cheap digital cameras have a time exposure mode, but if so you could rig one up to do this.

But for low power testing you just repeat the event at some regular rate and trigger on it.

Of course a Spice simulation lets you "scope" any current or voltage in the circuit, and tweak until you like the results. No parts need to be replaced when their ratings are exceeded either.

No, I missed it, I mostly just scan subject lines and read a few messages on interesting topics, like welding. (I paid for a good part of my engineering degree by pushing puddles of metal around.) I trust you have a means for adjusting phase overlap? Why not simulate your working circuit to gain some experience and confidence in Spice?

-----------

OT - a spice story:

A few years ago, when DOS 3.2 was mickeysofts latest and greatest, I bought the "free" student version of PSpice (with the $100 manual, good luck using it without the manual). Entirely text based, no schematic capture, much harder to use than LTSpice. Right after I read the manual a Co-Op student walked into my office and complained that a customer had provided an erroneous schematic for an overtemperature shutdown board they wanted us to reproduce since the OEM was out of business and their stock system was out of them. The student had figured out what was wrong with their schematic and fixed it, but the customer refused to accept the change, insisting that they had provided a certified drawing which could not be wrong, we must have made a mistake and we should fix our mistake not change the drawing. The student wanted me to call the customer and straighten him out, but instead I handed him my copy of PSpice and told him to simulate the circuit both ways. The customer dosen't accept the evidence, provide him with more evidence. "SPICE" he complained, "I don't know SPICE! I can't do that!" Sure you can, I said, providing a

The next day he returned with an inch thick dot-matrix printout on wide paper, which exactly duplicated both his pencil and paper analysis and the actual circuit operation. After I convinced him that rolling up the printout and beating the customer severely about the head and shoulders with it could be bad for repeat business, he sent the printout to the customer, with circles, arrows and post-it notes identifying the important parts. The customer then accepted the change and agreed to pay for it.

Next time I saw the customer I asked if he had changed his drawing. "Are you KIDDING? Do you know how hard it is to change a CERTIFIED DRAWING? I don't have that kind of time!" was his unsuprising response. But the Co-Op student gained some valuable experience, and has been a Spice advocate ever since.

Try it, you'll like it :-).

Winfield, you may be interested to know that I am reading your Art of Electronics and find it to be a fantastic book.

What do you mean by testing my models?

My job description is to write software models of certain real life processes, so I am a little confused: are you asking me to test spice simulations by seeing if they appear correct across a range of conditions, or are you asking me to test my actual devices and not to rely on models alone.

Like I said, I am tending not to trust similation here, rightly or wrongly, for two reasons:

That's true... Still, I think that the model would, due to my inexperience, give me a wrong answer and I would think that I know something, whereas I do not, a dangerous condition.

Yep...

I will check it out...

I will check it out... i

Very good, there's hope for you yet. :>)

By models, I mean the specific detailed component models Spice uses to simulate a circuit. Spice is intrinsically highly accurate, *IF* the models used in the spice simulation are accurate. Probably the most common misunderstanding new Spice users make is to assume Spice analysis is Spice analysis, without paying attention to the library models used in their Spice code. It's primarily the accuracy of the models that determines the accuracy of a Spice simulation. This means you need to create little experimental bench setups to measure all the relevant parameters of your components and add them to your Spice program.

For example, for your welder's inductor you need series resistance, self capacitance, and maybe some other parameters. I don't know if your inductor saturates at full current, but if it does, you'll need to include the nonlinear parameters for that. If it suffers a high- voltage breakdown, that should be included. I like to add the extra parameters as explicit additional elements in my Spice circuits. My point is, you can take specific bench measurements on your inductor to find out what's needed in its model, and thus get the parameters to put into the model or into your finished circuit.

When it comes to MOSFETS and IGBTs, there are a host of parameters you'll need. It may be the manufacturer's models are sufficient. But the only way to find out is to set up specific relevant tests, and see if the bench measurements match the Spice predictions to an adequate degree. Once you've done that, you can have confidence in your Spice models and thus in your Spice simulation. At that point it can become an intensely valuable "what-if" tool. I agree it takes considerable experience to know how to do all this well. But that's part of what serious electronics engineering is all about. Right?

I may have some more comments, but since this discussion has veered away from just the inductor, why don't you start a new thread?

All good points, but I think for this application getting adequate correlation between simulation and hardware will not be nearly as difficult as for instance with subthreshold MOSFET operation. In fact I expect that the IR models will be entirely adequate for i to obtain a good understanding of the effects of changing phase overlap, and gate drive failure. But I will admit to being wrong on a fairly regular basis :-).

Questions for i: What is your IGBT part number again (GA100TS60SQ ?), and have you read IR AN-1045, "AC TIG Welding Output Inverter Design Basics"? It looks like this AN covers almost exactly what you want to do, except that you do not have the luxury of pulse shaping to reduce current during switching due to lack of access to the welder current control circuit (no schematic in the manual, you would need to reverse engineer it) and the slow response available from your SCR based current controller would preclude pulse shaping even if you did.

BTW AN-1045 seems to imply that commercial welder designs simply accept IGBT failure as a consequence of gate drive failure; the snubber which will handle the inductor energy in this situation is probably not feasable.

I already used a couple of ideas described in that book.

That is the issue, I have no idea if it saturates at full current. Many, many things are undefined. So, I am facing: uncertainty about actual parameters, as well as overwhelming likelihood that I would not be able to create a model that I intend to create.

Hence, I think that the value of modeling this, for me, would be negative.

Carefully testing this system, starting at low current, would be more fruitful than trying to model it based on numerous unknowns, unknowables, possibility of mistakes etc.

An excellent point. Is breakdown of the inductor at high voltages (say 1000V) harmless to it or other circuit elements?

I admit to the same thing, and thus I am a little leery of trusting my modeling process too much.

No, Toshiba MG200Q2YS40.

See

Yes, more than once. :)

I listed a few reference papers on my webpage

That's right. The design is really quite simple. I have no plans to do such pulse shaping, that opens big trouble.

Here are some of my thoughts on this. Feel free to shoot them down.

1. During regular welding operation, switching of the H bridge would happen relatively quickly and I hope that it would happen within one microsecond. Furthermore, it is possible that I could make the bridge short for a fraction of a uS, rather than be open for a uS.

Such very small duration shorts or opens can be taken care of with a relatively small capacitor. If a duration of the open condition is within 2 uS, I calculated that a capacitor to absorb the current for the duration needs to be quite small and affordable.

1. Complete turn offs of the bridge should NOT happen during welding, on a _regular_basis. I cannot be running a bead and turning the welder's commutator handle at the same time.

Possible situations when a turn off would happen are:

In any case, these are rare events. I think that I can have a snubber circuit like a RCD, and also a bunch of transorbs (varistors). I already bought some varistors. That ought to take care of such rare events.

1. I hope that arc that extinguishes, such as when I withdraw the welding torch, is not going to generate voltage spikes and dI/dt of similarly high magnitude, as it would happen more slowly. Is that correct?

i

Well, I think that high inductance of the rreactor is kind of the central point of IGBT circuit protection issues. I do not mind starting a new thread, as such, but I also think that this thread is quite good.

Feel free to start a new thread, I will be happy to hear your thoughts.

I started a new thread, with some questions for you.

Great! I answered them.

A quick check failed to turn up any Toshiba spice models.

Actually pulse shaping avoids big trouble, but you can't do it with your supply.

Right, a very small overlao is best.

Could work.

Yes, as the arc is lengthened the voltage will go up and current will gradually drop to near zero before the arc extinguishes.

That's unfortunate...

Yep...

I will try very hard to do it.

Glen, would you suggest any sensible values for the cap and resistor? I have some high power diodes lying around (150 A, IIRC. I could parallel 3 of them).

That's very nice to know. Thanks a lot. Thoughts on sensible snubber component values would be appreciated greatly.

Sensible snubber component values need to be carefully calculated based on a more complete circuit model than I have time for - one which includes R, L and C on both sides of the H-Bridge. (The welding leads will have significant inductance also.) All of those things Win said you need to know to verify a good spice model also need to be known for a good pencil and paper design analysis. I will do both pencil and paper analysis and a Spice model when/if I ever get around do building a TIG welder - it is easy to make errors with either one, possible but less likely to make the same error with both. Agreement of analysis, simulation and hardware test is what I look for in a new design.

Since you do not appear to be prepared to do a thourough design analysis either, designing per the snubber application note you previously mentioned, ramping up your supply current slowly and monitoring your snubber performance would appear to be your other option.

The high power diodes you have hanging around may not be the most suitable. "Soft Recovery" diodes as recommended in the IR Application Notes would be better and are probably not too expensive.

That's very sensible.

Yes, I think that it is the safest approach.

I see. Are you referring to AN-1048, tig welder design?

Yes, I think they discussed soft recovery diodes there, but I could be confused and recalling some other AN.

Thanks. Glen, can you do me a huge favor and explain what soft recovery means. Thank you so much.

When a diode which is conducting in the forward direction is then reverse biased it will conduct in the reverse direction for a short period of time, then recover and shut off. Diodes which shut off in a short period of time are called fast recovery diodes (or just fast); this is good because they let through relatively little energy. Diodes which, when they have started to shut off, do so relatively slowly are called soft recovery, and the slower rate of current change during shutoff is also a good thing as the lower dI/dt produces less voltage in the wiring inductances. Fast recovery and soft recovery are not inconsistent. See for instance the data sheet of one of the diodes IR recommends for your application in AN-1048:

Thanks, but I am a little confused. Are you talking about freewheeling diodes, or the diode that is a part of RCD snubber? My IGBTs already have freewheeling diodes.

If you suggest this diode for the snubber, then it would need to be paralleled, right?