Some questions about my first drsstc.

[Mathieu thm]

Hello,
First of all sorry for my bad english, I am French
My name is Mathieu I'm 18 and I make my first drsstc, i wish i could tune my tesla coil better
and I have some questions.

First the setup:
     -controller : UD 2.7 rev c
     -IGBT : HGTG30N60A4D
     -OCD : 350 A (can I set it higher ?) 300 A
     -MMC : 10kv 60nf
     -CT : n87 (400:1 with two 20:1)
     -Secondary frequency : 210khz (10x30cm ; 0,25mm wire)
     -Primary frequency : 204khz 190khz
     -interrupter : up to 450hz and up to 100us (I ordered and am waiting for the famous Midistick!)


Here are some pictures :








During my first lightning test I obtained 50-60 cm of lightning (primary at 213 khz and secondary at 210.3khz) but the ocd at 350a was triggered at 100 Vac
I lower the frequency of the primary to 204 khz and I get 60-70cm of lightning and the ocd is now triggered at 150Vac.
I also noticed that the tesla coil draw less power at the input.
Do I have to lower the frequency of the primary much more? Are there any factors to consider?


In the next part the bus voltage is 30v which comes from a lab power supply because I don't have a differential probe

I tried to adjust phase lead as best as I can (7m3 153 inductor), here is the signal, in yellow the output of the bridge and purple the primary current:


What bothers me is that we can see that the beginning of the signal is quite good but towards the end there is a huge spike,
here is another screenshot:



And finally, here is a photo of the wires that go from the gdt to the igbt (around 20-30cm), are they a little too long?

[Mads Barnkob]

Hi Mathieu  and welcome to HVF! Your English is fine, do not worry about that!

Did you run your numbers through JavaTC for tuning point? I usually takes what it suggests and detune the primary to a 5% lower frequency. That gives for the loaded secondary change in frequency.

Your ZCS looks fine and the huge spike is at your turn-off, you can see the phase reversal as the energy flows back into the DC bus.

Your GDT wires are not longer than what I have used before, so I do not see a problem there.

350A OCD is suitable for a full-bridge of TO-247 switches. I blew mine up at 500A

[Mathieu thm]

Hi Mathieu  and welcome to HVF! Your English is fine, do not worry about that!

Thanks !

Thank you for all this information, and thank you for making me understand about the spike, by the way the huge spike is not too dangerous?

Did you run your numbers through JavaTC for tuning point? I usually takes what it suggests and detune the primary to a 5% lower frequency. That gives for the loaded secondary change in frequency.

This is what javaTC says :


I detune by 5% compared to the primary frequency given by javaTC or compared to the secondary frequency given by javaTC?

What was your igbt that you exploded?

[davekni]

Looks like a great clean build.  Congratulations on getting it working.

OCD : 350 A (can I set it higher ?)

Looks like a good value for those IGBTs.  Does depend on Vge in your build (UD2.7 supply voltage) and on your tolerance for risk of fried IGBTs.  HGTG30N60A4D datasheet short-circuit current graph is continuing up at it's 15Vge endpoint.  So it might be possible to push beyond 350A for short bursts with 20 or 24 Vge.

Do I have to lower the frequency of the primary much more? Are there any factors to consider?

Test lower primary frequency to see how it works.  Scoping primary current is possible at full power, as is scoping secondary current.  That can show what's happening even w/o being able to scope H-bridge outputs.

What bothers me is that we can see that the beginning of the signal is quite good but towards the end there is a huge spike,
here is another screenshot:

Hard to say what the spike will look like at full H-bridge voltage.  However, I'd explore it's cause a bit more at 30V.  It is a very short spike.  Zoom in to see what it looks like in detail.  Also scope gate voltages.  HGTG30N60A4D are fast IGBTs (good), so may respond to short Vge glitches.

Spikes also depend on H-bridge layout and parasitic inductances.  Since your normal switching looks good, parasitic inductance is likely low enough.

And finally, here is a photo of the wires that go from the gdt to the igbt (around 20-30cm), are they a little too long?

Do you happen to have a photo of GDT itself?  Leads are very nicely twisted together.  I presume GDT primary leads are also twisted together.

[JCF]

Belle réalisation Mathieu ! bienvenue sur le forum

[Mathieu thm]

Looks like a great clean build.  Congratulations on getting it working.

Thanks !

Belle réalisation Mathieu ! bienvenue sur le forum

Merci ! 

I tried to reduce the resonance frequency of the primary and I get more and more impressive results.
I will continue to reduce the frequency until I have the best possible results.
We can note that the better the drsstc is tune, the less the OCD is triggered, now I can go up to 190-200Vac without the ocd being triggered (with the same 350a OCD).
I still hadn't put a video so here is a video of the drsstc in operation ! :

/>And Some pictures:

lightning reaches 80 cm, maybe more

Does depend on Vge in your build (UD2.7 supply voltage) and on your tolerance for risk of fried IGBTs.

The us 2.7 is supplied with a 24V switching power supply.
What do you mean by "risk of fried IGBTs." ? I don't understand 

Zoom in to see what it looks like in detail.

I'll tell you all about it soon

Otherwise I noticed that the more I put a big dummy load, the more the spike shrinks, so I assume that with the secondary, and well adjusted, maybe the spike shrinks ?

[davekni]

What do you mean by "risk of fried IGBTs." ? I don't understand 

If you try 400A OCD, you may get higher performance.  Or IGBTs may fry (may be destroyed due to the high current).  At 24Vge, you may get away with 400A.  However, if you do not want to take the chance on needing to buy more IGBTs and replace damaged ones, then say at 350A.

Otherwise I noticed that the more I put a big dummy load, the more the spike shrinks, so I assume that with the secondary, and well adjusted, maybe the spike shrinks?

That sounds encouraging.  If the cause of the spike can be found, it may be possible to predict if it is likely to be an issue at full bus voltage or not.  I wouldn't count on loading fixing the spike.  Arc loading can come and go as arcs form and end.

Nice performance!  Clearly the spike isn't too problematic.  No IGBTs have died yet.

I will continue to reduce the frequency until I have the best possible results.

At some point the primary frequency will get too low to start an arc.  OCD will trip before arcs start.  Just above that frequency is likely to produce longest arcs.  Best music playing (MIDI) will be at higher primary frequency (less detuning) than for longest arcs.  Experiment to see what works best for your coil.

[Mathieu thm]

Zoom in to see what it looks like in detail.

So I zoomed in, and here's what it looks like (at 30v)

 [ You are not allowed to view attachments ]
 [ You are not allowed to view attachments ]
 [ You are not allowed to view attachments ]
 
I thought of a solution for the spike, maybe increase / replace the snubber capacitors, I currently have two 941c 1200v 0.47uf in parallel, so 0.94uf in all, is that enough?

Experiment to see what works best for your coil.

I experimented with a little more detune but the arcs get too big and there are flashovers, so I went back to the previous setting which is the final setting.
I prefer the drsstc to be more reliable  than produce more arcs, I think this setting will be suitable for midi reading, I will tell you all that when I receive my Midistick

[davekni]

I thought of a solution for the spike, maybe increase / replace the snubber capacitors, I currently have two 941c 1200v 0.47uf in parallel, so 0.94uf in all, is that enough?

For a 20ns spike, 0.94uF is plenty, a tiny fraction of an ohm.  The spike is across snubber lead wires and other H-bridge parasitic interconnect inductance.  Amplitude could be reduced by reducing parasitic inductances.  However, there is a limit to that reduction given minimum of IGBT lead inductance.  Vge scope traces might show some glitch or ringing that is causing rapid current switching (hard switching).  Removing the cause of rapid current switching is better (if possible) than mitigating voltage amplitude by just reduced interconnect inductance.

In particular, I'm wondering why there's any H-bridge output voltage transition 500ns after the normal transition.  UD2.7 should disable within ~20ns of last normal transition, the delay of 74HC74.  That 20ns glitch (from transition to disable) may or may not get through gate driver and to gates.  Even if it does get through, doesn't usually cause issues.  All four IGBT gates should be off well before 500ns.  Thus I'm suspicious of some issue with either your UD2.7 or with GDT or R//D connections from GDT to IGBTs.  Would help to see scope traces of GDT inputs (UD2.7 outputs) and IGBT gates zoomed-in around the last normal cycle and first post-disable cycle.

Of course, since no IGBTs have failed yet, you don't necessarily need to find the cause of glitches.

BTW, Americans often call damaged or failed or dead electronic parts "fried".  It's another synonym within electronics field.  I'm guessing that was the cause of confusion with my initial post.

I experimented with a little more detune but the arcs get too big and there are flashovers, so I went back to the previous setting which is the final setting.
I prefer the drsstc to be more reliable  than produce more arcs, I think this setting will be suitable for midi reading, I will tell you all that when I receive my Midistick

Yes, flashovers or racing sparks or other such issues can limit arc length.  Sounds like a good plan to leave it tuned for likely-good MIDI performance.

[Mathieu thm]

Thus I'm suspicious of some issue with either your UD2.7 or with GDT or R//D connections from GDT to IGBTs.

the gdt I use is a Chinese 18/18 gdt  but when i tested it with my function generator the signal was perfect.
maybe shortening the wires from the gdt to the IGBTs would help.
 [ You are not allowed to view attachments ]
I tried to make a new gdt with n87 ferrite, and here is the result:
 [ You are not allowed to view attachments ]
And here is the output of the gdt (with 50ohm load) with my generator connected to the input at 20v and 190khz:
 [ You are not allowed to view attachments ]
 [ You are not allowed to view attachments ]
I don't think replacing it will make a big difference.

Would help to see scope traces of GDT inputs (UD2.7 outputs) and IGBT gates zoomed-in around the last normal cycle and first post-disable cycle.

I will test all that and I will tell you again, I have already tested to scope the outputs of the gdt in operation on the drsstc and I remember that the signals were clean, but I will tell you again to be sure.

Edit:
 this is an output of the gdt scope on an igbt:

and that, an output of the ud2.7 (from ground to an output pin of the ud2.7):

[davekni]

Your new GDT looks great!  Old one may be fine, but not sure.

this is an output of the gdt scope on an igbt:

Looks a bit questionable in two respects.  First, Vge is ~+5V after burst ends.  Should be ~0V.  This may be the cause of the H-bridge output glitch.  The IGBT may not be completely off.  I'd suggest scoping that GDT output.  If GDT output is 0V but Vge isn't, then diode from GDT to gate must be open or reversed.  (Is the scope probe ground connected to IGBT emitter when scoping gate?  If grounded farther away, ~5V might be difference in ground to emitter voltage.)
Second, Vge is +-18-19V.  UD2.7 output is about 0 to 22V, so should make +-22V between the two UD2.7 outputs, and therefore should make +-22Vge.  Could be a GDT issue or an issue with R//D between GDT outputs and gates or an issue with the other UD2.7 output that isn't shown in a scope trace.

BTW, the 20ns glitch may never cause a real failure or problem.  Don't feel obligated to keep chasing the cause.  However, I'd suggest understanding the cause.  It could be something that eventually does cause a failure.

[Mads Barnkob]

Congratulations on getting it to run and make sparks!

The glitch is a "feature" of the UD2.x designs. Steve Ward tried to get it removed, but I seem to recall that it was just overcomplication for something that does not pose a problem.

About testing at 30VDC on the bus, you will see much large spikes compared to the input voltage. This is due to the capacitances in the IGBT are all set in stone, no matter the voltages it sees. Try to up the DC bus voltage and take close note if the spikes get smaller in ratio to the DC bus or if it keeps growing as well.

[davekni]

The glitch is a "feature" of the UD2.x designs. Steve Ward tried to get it removed, but I seem to recall that it was just overcomplication for something that does not pose a problem.

Agree, UD2.X generates an output glitch due to 74HC74 delay time.  It shows up on mthome4's scope capture of one UD2.7 output.
I don't think this UD2.7 output glitch is the cause of the negative glitch 500ns later on H-bridge output.  The one Vge scope capture shows Vge isn't being held close to 0V after burst ends.  It doesn't look like the normal ring-down of GDT inductance with UD2.7 output coupling capacitance.  Vge of other three IGBTs may look different.  I suspect an issue with R//D connections from GDT to IGBTs, or possibly some issue with GDT and/or UD2.7 output coupling caps.

Also agree that H-bridge output glitch may not grow with DC bus voltage.  Clearly it hasn't been a serious issue yet given great performance.  The glitch cause may be benign.  I do tend to obsess with any anomalies that aren't understood.  However, if Vge isn't being held close to 0 after bursts, higher temperature or other changes at full power might lead to thermal runaway.

One possibility just came to mind:  Are there 18V TVS or zener diodes on gates?  That could explain some of the symptoms.  If feeding UD2.7 with 24V, gate clamp diodes need to be 24V or higher.

[Mathieu thm]

Congratulations on getting it to run and make sparks!

Thanks !

About testing at 30VDC on the bus, you will see much large spikes compared to the input voltage. This is due to the capacitances in the IGBT are all set in stone, no matter the voltages it sees. Try to up the DC bus voltage and take close note if the spikes get smaller in ratio to the DC bus or if it keeps growing as well.

The glitch is a "feature" of the UD2.x designs. Steve Ward tried to get it removed, but I seem to recall that it was just overcomplication for something that does not pose a problem.

I will try and thanks for this valuable information on the ud2.7.

(Is the scope probe ground connected to IGBT emitter when scoping gate?

Yes I scoped on the gate and emitter of the IGBT.

One possibility just came to mind:  Are there 18V TVS or zener diodes on gates?  That could explain some of the symptoms.  If feeding UD2.7 with 24V, gate clamp diodes need to be 24V or higher.

Oh , it’s true that I use 18V TVS on IGBT gates....
Can you explain to me how that a problem?
If this is the problem I order 24v tvs to replace them.

[davekni]

Oh , it’s true that I use 18V TVS on IGBT gates....
Can you explain to me how that a problem?
If this is the problem I order 24v tvs to replace them.
« Last Edit: August 26, 2022, 11:14:32 PM by mthome4 »

The problem is that UD2.7 is attempting to drive Vge to +-24V.  (Actually slightly lower due to IGBT gate capacitance, perhaps actually +-23V.  The 1V drop is mostly across UD2.7 output coupling capacitors.)  The 18V TVS diodes are clamping actual Vge to +-18V.  Draws large current from UD2.7 outputs and leaves excess voltage on UD2.7 output coupling capacitors, which is likely the cause of the remaining ~5Vge after enable ends.
Your options are to either reduce UD2.7 supply voltage to 19V or under, or to change TVS diodes.  Most DRSSTC coils use 27V-33V TVS diodes.  As Mads has posted before, actual IGBT gate oxide punch-through failure occurs around 80V.  TVS diodes gate-emitter are intended to clamp occasional spikes such as might be caused by flash-over or other unintended events.  That's why they are normally several volts above normal operating Vge.

Hopefully this will fix both your Vge waveforms and remove the short negative H-bridge output spike.  Needing to change something is a nuisance, but I'd suggest thinking of it as a good situation.  Now you know what the (hopefully only) issue is.  Your coil should be more reliable after fixing this.

[Mathieu thm]

Thank you very much for explaining to me the problem with 18V tvs

I tested several things and the results are interesting:

Most DRSSTC coils use 27V-33V TVS diodes.

I tested to put 27v zener diodes instead of 18v TVS:


I also put my new gdt:


The results are interesting, first of all I no longer have a problem with the 5v voltage at the end of the cycle, the result is the same with the old gdt (so I will order 27v TVS because my zener are of poor quality, thanks for noticing that!)


And the second good news is that the spike seems to be getting shorter:



I still have a question about my new gdt, so I'm using an n87 ferrite core with a cat 5 cable, I did 10 turns.
However my cable has a metal shield:


The performance of the gdt can be better by removing this shielding?

[Mads Barnkob]

If you put the shielding in parallel with the primary wires of the GDT, you get one of the lowest possible inductance GDTs and with the highest possible coupling. Compared to delta, sectional, circumferential and trifilar wound GDTs.

[davekni]

The results are interesting, first of all I no longer have a problem with the 5v voltage at the end of the cycle, the result is the same with the old gdt (so I will order 27v TVS because my zener are of poor quality, thanks for noticing that!)

Yes, Vge waveform looks more normal now.  The undershoot on falling edges is normal due to diodes from GDT to gate.  Rising edges have some ring.  That is within normal range too.  Higher-value gate series resistors would reduce that ring, but might also require readjusting phase lead to be a bit more.

And the second good news is that the spike seems to be getting shorter:

That 330ns positive pulse is normal, the time from burst end until primary current reversal.  It is the very-short negative spike at the end of this positive pulse that might be of concern.  As Mads mentioned, that short ~20ns spike might not increase with higher bus voltage.  As long as the sum of spike voltage and DC bus voltage together doesn't exceed IGBT's 600Vce rating, there is no issue.  If that short negative spike does grow with bus voltage, then it might eventually lead to IGBT failure due to excess Vce.

Just realized that you are probably connecting scope probe ground lead to one H-bridge output and probing the other output.  Even with an isolated bench supply, connecting scope ground to an active signal can cause problems.  The bench supply has significant capacitance to ground.  I'd suggest scoping one H-bridge output at a time with scope probe ground connected to negative bus supply.  Scoping the two outputs individually also makes it clear which output has the spike.  Scoping both outputs at the same time using two probes can be helpful.  Also, two-channel captures showing both one output and one gate simultaneously is helpful.

Your decision on whether to continue scoping signals to figure out the 20ns negative spike, or just have fun enjoying your nice new coil.  If you do continue scoping, a couple more suggestions on scope use:  First, I'd suggest changing vertical channel mode to DC rather than AC.  Including DC information is often useful.  Second, use the lowest volts/division setting possible without spikes going off-screen.  Makes seeing subtle detail easier.  Vge is difficult to read at 50V/div.  10V/div would be more appropriate.  For chasing this specific burst-end spike, most useful horizontal scale would be about 1us/division, showing one cycle before and one cycle after burst end.

Do you happen to have any Y-caps around (capacitors rated for line to ground use)?  If you have one, or some ~2-10nF capacitor of sufficient voltage rating, it is possible to get useful scope captures even when running at full voltage.  Connect scope ground lead through capacitor to Vbus-.  That will be sufficient high-frequency grounding to allow scoping H-bridge output spikes to see if they grow to a problematic amplitude.  Use low burst repetition rate when scoping since many probes have low RMS voltage rating at high frequency.  Of course, use caution when working with line power.

I still have a question about my new gdt, so I'm using an n87 ferrite core with a cat 5 cable, I did 10 turns.
However my cable has a metal shield:

I haven't seen a GDT use shielded cable before.  Mads has a good idea of paralleling with primary wires.  Only disadvantage that comes to mind is increased capacitance between windings due to shield.  With your fast IGBTs and resulting fast output slew rates, lower inter-winding capacitance could also be advantageous.  Another thing that can help prevent issues resulting from fast output slew rates is to add ferrite beads around the two GDT outputs going to high-side IGBTs.  So, perhaps use Mads suggestion for shield and then add ferrite beads to high-side outputs to mitigate possible capacitive coupling issues.

Interesting project!  But, don't let my personal curiosity drive what you want for your project.  Proceed as you prefer.

[Mads Barnkob]

I think the idea originally comes from using coaxial cables in GDTs, so not to see the shield as a shield, but solely as a primary winding. I got some measurements somewhere that should go in the not-yet-finished GDT part of the DRSSTC guide. Maybe some day its done

I still have a question about my new gdt, so I'm using an n87 ferrite core with a cat 5 cable, I did 10 turns.
However my cable has a metal shield:

I haven't seen a GDT use shielded cable before.  Mads has a good idea of paralleling with primary wires.  Only disadvantage that comes to mind is increased capacitance between windings due to shield.  With your fast IGBTs and resulting fast output slew rates, lower inter-winding capacitance could also be advantageous.  Another thing that can help prevent issues resulting from fast output slew rates is to add ferrite beads around the two GDT outputs going to high-side IGBTs.  So, perhaps use Mads suggestion for shield and then add ferrite beads to high-side outputs to mitigate possible capacitive coupling issues.

Interesting project!  But, don't let my personal curiosity drive what you want for your project.  Proceed as you prefer.

[Mathieu thm]

Bad news :
I tested the drsstc with the midistick, everything worked perfectly, I moved the coil outside to be able to test better.
when I turn it on, nothing happens, I look at the fuse and it burnt.
Infact two igbt are dead, I don't know why they burned, the phase lead was well adjusted.
I did the test at 100-120Vac max
small note: the heatsink where the igbts are was warmer than in fixed mode

Buying new IGBTs is not a problem, it's more the fact that why it burned?

otherwise this is how i installed the new gdt:



I will let you know when I have replaced the IGBTs, the TVS and maybe put higher gate resistors (10ohms)

EDIT:
I may have a hypothesis, to test I used midi files with a lot of notes, I saw the files of bytecode77 and they have a lot less notes, so I think with all these notes, on times overlapped and maybe it made hard switching.
By the way, I have 60N60SMD lying around, are they as good as the HGTG30N604AD?

Anyway I will still buy HGTG30N60A4D.