Flash-over in drive. Second opinion needed. 5

[Skogsgurra]

I have this problem:

A customer has four drives installed in one section of his plant. They are made by a well-known manufacturer and identical drives perform flawlessly in other parts of the plant.

The power range is 390, 545, 615 and 765 kW and they are fed from a 690 V, 50 Hz IT grid with a 3.15 MVA 5.73 % DY11 transformer. Input rectifiers are thyristor controlled and keep the DC link at 930 V constant.

The problem: The fuses of the two larger drives blow infrequently. The "MTBB" seems to be about one month, but has been as short as two days.

Observations:
*The two drives never blow their fuses at the same time.
*There are no visible transients on the mains voltage (recorded with 16 kS/s and 12 bit resolution, no filter).
*Current rises from normal load current (around 300 A) to 5400 A in 3 milliseconds (same recorder) when the Silized fuses blow and current drops to zero in about 2 milliseconds.
*The arc guard (ABB) reacts about 4 milliseconds after the current starts to rise.
*What I said about "no transients" is not quite true; there is a 6 kHz ringing started by the thyristors in the input rectifier of the inverters. The amplitude of this ringing is about 100 Vp-p. But there are no "killer transients" on the three phases.
*The fault current has a 100 000 kA/s maximum rate of rise and that corresponds well to what can be expected with 3% line reactors.
*Traces of flash-overs can be seen on the thyristor heat sinks (disc thyristors, marks on bolt and heat sink). The flash length is about 8 mm. Which (in my thinking) corresponds to more than 10 kV.
*Other equipment on the same grid are not influenced.

My analysis:
I have excluded false triggering because the DC link voltage is already high and premature triggering cannot produce this kind of overcurrent.
There is no overvoltage on the line. Not recorded and not influencing other equipment.
The high voltage must be generated by something. What?

I would like to bounce this idea in the forum:
The holding current (Ih) of a normal thyristor in this size is something between 100 and 500 mA and the snubbers are sized to take care of the "snap-off voltage" when the current goes below Ih.

Now, if the 6 kHz ringing snaps the thyristor off prematurely - remember that these thyristors work against the DC link, which is a counter-EMF. What will happen? Say that it is turned off while line current is at 5 A. The magnetic energy in the line reactor will then be one hundred times the normal energy at turn off and that will take the voltage on the snubbers up to more than ten times the normal voltage (energy in capacitor plus voltage drop in series resistor).

If the thyristors can take this high voltage without damage, then there would certainly be a flash-over. Wouldn't it? The question is: Can a thyristor withstand 10 kV or more for the time needed to ignite an arc?* I have not been able to find any data on this.

I need an informed second opinion. Or even better: Have you had this problem? Is it described anywhere in literature?



* The snubbers are connected with rather long wires and I do not think that the inductance is optimal. The fast snap-off voltage is probably not absorbed by the snubbers.

 

[light1]

The flashover is related to a creepage distance. The peak flashover voltage will be higher than the SCR peak reverse voltage of your SCRs. Analyze the attached SCRs data sheet and contact the SCR manufacturer,

http://www.eupec.com/gb/2_PRODUCTS/2_1_ProductRange/pdf/ti_t_2871n.pdf

 

[Skogsgurra]

Thanks light!

Is this thyristor intended for 690 V mains voltage? That would explain that a high voltage can exist across the thyristor without "force-firing" it. The 8+ kV seems to fit the circumstances and it is possible that the short term withstand voltage is even higher.

It has not been possible to contact the manufacturer (which, BTW, is the correct one) because of the week-end, but I will do that first thing today, Monday.

Do you have any idea about the premature turn-off of the thyristors? Is it something that you have heard about? Is it at all possible in this standard configuration?

 

[ScottyUK]

Hi skogsgurra,

In a previous (R&D) role, my group had some damaged SCRs from a large traction rectifier analysed by a local university. They used a scanning electron microscope to examine the failure area, and they were able to give a good estimate of the cause of failure. In our case it was a combination of weak gate drive and very rapid rise of load current leading to current crowding in a small area of the die, resulting in localised burning of the silicon. We found that the cost was very reasonable: perhaps one of your universisties has similar facilities?

To produce your multi-kV voltage, your snubbers seem to be under suspicion. Am I right in thinking that the snubber is a standard R-C type, and the reactor they might be interacting with is the line reactor on the drive input? The Q factor of that circuit should be quite low, not given to producing large voltages.

When the fault occurs, how is the fault distributed: does it appear in two lines simultaneously? Have you got a transducer on the DC output of the rectifier, before the capacitor bank? If so, what does the current in this path do during the fault?

I apologise if any of the above makes no sense - it is 0650 hrs, and I am almost awake!




------------------------------

If we learn from our mistakes,
I'm getting a great education!

 

[Skogsgurra]

Hi Scotty!

I connected the recorder about ten days ago. We had one incident last thursday and I am trying to understand what we see on the strip (memory, actually).

The DC link input current is available only indirectly and at a slow 300 S/s sampling rate (six pulses at 50 Hz). So it is quite useless. A fast current transducer cannot be fitted (no room for it) we have tried Rogowski and also a Fischer high-speed transducer and both are too bulky. I was temted to use part of the DC Cu railing as a shunt (yes!) and install a fast isolation amplifier. But that has yet to be done. We do not have access to the drives because it interferes with production.

Yes, the snubbers are RC types with something like 47 ohms and 1 uF. But one thing that I do not like is that they are connected via a rather long pair of cables with a two-pole connector in between. It could actually be that the connector has come loose. Have not had a chance of checking that yet. The inverters are 180 km away from my place and I cannot open them to find out even if I go there.

My #1 question now: Is it possible that the "snap off" can happen at elevated current levels? Would it produce the high voltage that we think exists? Anything written about it?

Sorry that I woke you up at these early hours. ;-)

 

[Skogsgurra]

Correction!

I was wrong about the snubbers. They are NOT connected with long cables and connectors. They are placed close to the Thyristor. Also, the resistor is 15 ohms, not 47 ohms. Capacitor has 2400 V rated voltage.

 

[ScottyUK]

Skogsgurra,

Is it possible that the drive is running at light load, or regenerating even for a cycle or two, at the time the failure occurs? Regeneration in particular could give a shoot-through condition if the delay angle is such that the thyristor has inadequate time to recover its blocking state. These large thyristors are slow to recover blocking condition following a period of conduction.

I can't see a reasonable way that the snubber circuit could lead to an overvoltage condition as aggressive as the one you seem to be experiencing. I think you are more probably looking at commutation problems. Is the damage definitely a flashover, or could it be a surface burn from the 5.4kA fault through the thyristor?



------------------------------

If we learn from our mistakes,
I'm getting a great education!

 

[Skogsgurra]

Thanks Scotty. I really appreciate your help.

In this one, we have no regeneration. The input rectifier is a one quadrant thyristor rectifier so no cross-conduction is possible. I have done some reckoning to see if a false trigger pulse can "make big current happen" but since the DC voltage is kept at 930 V and rather constant, there is simply not enough voltage difference across the thyristors to produce an overcurrent that blows the fuses in three ms (even if the fuses are superfast).

To conclude:
Need input badly. This IS a tough one. Have established contact w/ manufacturer. They are thinking. Customer worried.

 

[Marke]

Hello skogsgurra

It is difficult to determine what is the cause and what is the consequence in this type of problem.
Flash overs can ocure as a result of ionised air due to a failure on another component. It is easy to look at the arc track and assume that this indicates the voltage present, but if there is a build up of dust and moisture, or if there is another component nearby that has ruptured, then this arc track may not be the cause, rather the consequence.

Can you give details of what the thyristors are, type, rating and package, and if there are any other components that need to be replaced as a result of this failure, and maybe one of us can come up with another scenario.

Best regards,

Mark Empson

http://www.lmphotonics.com

 

[jraef]

ScottyUK,
Could you please elaborate on why you think a light load might be cause for consideration? I have a similar circumstance happening within a particular frame size of VFD (25-40HP, much smaller than skogskurra's) and it ONLY happens when there is a very light load, i.e. a 57A rated drive pulling 10-12A. Regen is also not an issue, they have all been on centrifugal pumps. We are doing a forensic analysis right now, but I'm interested in why you mentioned light loads as a possible course of investigation. We too are stumped at the moment.

 

[Skogsgurra]

Mark,

I wish that I knew how to attach pictures here. I have one showing the flash-over traces. I will try and describe it:

The thyristor has customer's marking, no standard marking. The dimensions are 60 mm diameter and 26 mm thick (hockey puck). Double-sided cooling with two heat-sinks with bolts insulated on one side. Heat-sinks have ribs and there is a "ribless" area in the center of the extruded profile. That is where the insulated bolts are. It is a standard mounting for puck thyristors.

The flash-over is between the end of the bolt and the ridge of the nearest rib. The bolt is 8 mm and the distance to the ridge is about 10 mm. The plant was installed and commisioned a little more than a year ago and the drives are in an area with filtered and conditioned air. No dust. No dirt. Temperature between 20 and 25 centigrades. Normal humidity (nordic countries).

There are four drives in the 300 - 700 kW range in this part of the plant and only two of them (the largest ones) have this problem. It never happens to both drives at the same time and all that is needed to get going again is to change fuses and restart.

The drives are in 24/24 7/7 production and we do not have access to them just like that. I have been active in this business for decades and I have seen this happen before. It was in an old mercury arc rectifier when the temperature got too low and arc extinction produced high inductive voltage. But we had no snubbers on those, and the voltages involved were quite high per se. (10 MW steel mill). The analogy is that turning the thyristors off prematurely is electrically identical to arc extinction.

Own thinking machinery seems to be idling. Will Eng-Tips put it in gear again?

 

[Marke]

Hello skogsgurra

Yes, I kno the type of construction well. I have built many large controllers using hocky puck type SCRs. I imagine that the heatsink you describe would be similar to one that we used to use called a P8. Certainly, the same type principle of construction.

I am surprised that the flashover would be from the expsed end of the bolt through to the heatsink. As you say, that would require a high voltage and that would effectively be across the SCR itself. If that sort of voltage appeared across the SCR, it would almost certainly fail. Generally, for that type of application and that voltage range, I would expect the SCRs to be rated around 2200 volts and I can not see that flashing over. If you had a very fast high amplitude transien, it would cause the SCR to overhead trigger and unless the rate of rise of current was very limited, it would cause localised hotspotting on the die, and the SCR would fail. There is no way that I can see you getting the voltage across the SCR high enough, even with a transient, to cause that sort of flash over. I would expect the SCR to turn on/break down and drop the voltage. I believe that there must be another explanation.
As a thought, most of the hocky puck clamps available are designed for 400 volt supply systems. I wounder if the insulation on the clamp itself is partially breaking down, causing some form of corona type discharge through the insulation and this is causing ionised air to move up the bolt, breaking down the air insulation?
Have you disassembled a clamp to check the insulation sleeves? Some that I have seen just are not anywhere near what I would consider acceptable for the voltage and heat that they are subjected to.
I assume that you are using energy limiting semiconductor fuses on the input side. Thes are very good at protecting SCRs against over current situations, but not overvoltage. One of the consequences of using the energy limiting fuses is, that when there is some form of discharge, the damage is limited to the extent that it is often difficulat to see where it went. Standard fuses would allow a much better trail fo destruction and then things are definitly broken in identifiable places.

If we assume that the problem was due simply to an insulation breakdown in the air between the end of the bolt and the heatsink, then a good coating of silicon over the end of the bolt would sort that out very quickly. I would suggest you could coat the end of the bolts with a silicon based conformal coating such as Dow Corning 340 that is used on pcbs. a 1 mm thick coating would sort out any problem in that regards. It would make disassembly more difficult though. I have also used a liberal coating of a heatsink compound as a good insulator in this type of situation.
My recommendaion, would be to coat the heatsink around the area where the flashover occurs with 340 conformal coating. This will impair the thermal resistance a little, but the actual area that you would coat would be small relative to the total surface area of the heatsink and so should not be a problem.

This may provide a shor term solution, but I suspect that the insulation in the clamps will eventually give way.

Best regards,

Mark Empson

http://www.lmphotonics.com

 

[Skogsgurra]

Thanks Mark!

New thinking there. I also thought that 10+ kV would be impossible across a thyristor but Light1 (see his answer early in this thread) gave an example of a data sheet that I do not think is the actual thyristor but says that repetitive voltage can be as high as 8,2 kV - guaranteed. That takes the withstand voltage (typical or on a lucky day) up to the 10+ kV region. So, I guess that it could be possible.

I have now got a contact with competent people at the manufaturer´s HQ and we have discussed the matter for about half an hour. They have my full report and I hope to have some wise words from them tomorrow.

The bad insulation/ionisation thought is good. There is a unit that has been removed from one of the inverters. It has flash-over marks and I will have it examined for weak insulation.

To be continued...

 

[light1]

The posted 8000V thyristor data sheet indicates the maximum (peak) repetitive voltage.
The thyristor, just like a diode, can have a soft turn-off or an abrupt turn-off. The latter causes small damped oscillations after the abrupt turn-off.
The premature turn-off is not common at the properly functioning SCR. Possibly, a TRIAC can have premature turn-off due to the physical asymmetry.

 

[ScottyUK]

Hi jraef - hopefully skogsgurra won't mind if I sidetrack his thread for a moment until he can post his results. Hopefully we can all learn something new.

What is the rectifier on your drive? Is it a fully-controlled SCR bridge front end? Very long delay angles can cause commutation failure on fully-controlled rectifiers if the SCR does not have time to recover its blocking state. Adding source reactance worsens the problem because it slows down the transition of current from the off-going into the on-going legs of the bridge, thus reducing the period where the SCR is reverse biased during which time it must switch off and recover blocking capability.

It might be worth starting a new thread before we distract this one too much.



------------------------------

If we learn from our mistakes,
I'm getting a great education!

 

[Skogsgurra]

Scotty,

I think that we can have a broad discussion. Small excursions are stimulating and can start the brain in new directions and make someone think along fruitful lines, which serves the purpose of finding a solution to the problem.

I had the problem you describe in a range of resistance welders that we were asked to improve back in the last century. Had to make firing pulse width dependent on delay angle to avoid problems when running lightly loaded. A resistance welder abuses the transformer at high load and depends on voltage drop in primary and mains to avoid saturation of the core. The resulting load has cos(phi) not too far from 1. At low load, the load is almost purely inductive and that makes control critical. As you described. But, as you said, that is another story.

 

[Marke]

Hello skogsgurra

I would not expect to see SCRs rated much above 2500 in this application, high voltage SCRs are tremendously expensive.
Premature turn OFF should not be an issue if the unit is well designed. There have been problems in the past, (and I am sure there will be in the future) with SCR trigger circuits that comprise a single short gate pulse only. It is a well known and documented requirement for this type of application in other than laboratory conditions, that there must be continuous gate fire for the expected duration of the SCR conduction. This can be achieved by either a continuous gate pulse, (requires quite a bit of energy) or by a picket fence pulse stream.
SCRs are very slow at turn OFF, so any excess voltage generated by a turn OFF will tend to be swamped by the device turning ON again. You will not get quite the same affect as a contact opening.

I am certainly interested in your findings, I am sure that you will follow this up.

Best regards,

Mark Empson

http://www.lmphotonics.com

 

[jraef]

skogsgurra,
Even before I finished reading Marke's response I was coming to the same conclusion from your detailed description. Were it a nastier environment I would have suspected moisture tracking, but once you noted the bolt heads the light went on. I have run into that exact issue in some 1000V soft starters in a tunnel boring machine. They flashed over and it turned out that the bolt sleave was only rated for 600V max. (the assembly process had called for a different color sleave than what we found they used). Tolerance exceptions must have allowed it to work for a while, but eventually a high line condition would come along and they would blow. They were genset fed and high line voltages occurred when the load dropped off. Interesting coincidence don't you think?

Side track comment of a side track...
I have learned more in the last 2 years participating in this forum than I had in the previous 10.

"Venditori de oleum-vipera non vigere excordis populi"

 

[Skogsgurra]

Grrreat, jraef! I will have the failed unit in my lab tomorrow (if the post works the way it should). I will subject it to a current limited (resistor closest to thyristor) DC and take it all the way up to a flash-over. Any thoughts or recommendations?

I can certainly agree with your comment about learning. You are all very helpful. Never had this much support even when I was working at "the big company HQ". All of you deserve double stars.

 

[Skogsgurra]

No. It didn't work. No parcel in the post. No answers from the manufacturer. And a new case of blown fuses at the customer's plant this morning. Stay tuned!