Is there an industry accepted Arc Flash calculation method for DC Arc Flash for a PV Facility? The application I'm in search of is for PV facilities where a central inverter is used. Typically the DC voltage can be as high as 1500 V and the inverters can be anywhere from 1-4 MVA.

I have seen some developers use the paper "Circulating DC Arc Flash" but I'm curious what others are using.

Keep in mind three things:

1. There is little to no testing and what there is, is pretty limited.
2. There are two DC arc flash calculations out there. One due to Neal is basically the Lee equation adopted to DC. The other due to Ammerman is a little more complex but takes into account the shape of the V-I curve for DC arcs. It results in about a 50% reduction in calculated incident energies. For obvious reasons Ammerman's calculation is probably the better of the two but no matter what you are doing, it's still all based on theoretical calculations.
3. Of the test work I've seen pretty much shows that the Ammerman and Neal equations are off by a factor of 2 to 4, but the test work was at around 250 VDC so not really applicable to your situation. I'd expect the error to grow significantly as voltage increases though because that's what happens with AC calculations using the same type of model (due to Lee).

Thanks Paul. You have confirmed my suspicion

Just as a follow up the "off" part is that it OVERPREDICTS but a factor of 2-4, at least in that model.

There are lots of reasons for this. One is that it assumes that energy conversion to heat is 100% efficient when in reality it gets converted to other things such as kinetic energy (noise). Another is that it assumes that the arc thermal radiation is not occluded or absorbed somewhere else such as from smoke or molecular excitation and movement. If this was all there was to it, we could easily just apply a correction factor.

There are similar problems in the AC model but one of the curious things about the AC model is that we have to reduce the assumed arcing current by a significant amount to get the data to "fit" but the current reduction goes away as the voltage goes up and is basically gone above 1 kV. The clear reason for this has to do with how an arc works on an AC sine wave. As the current goes through zero twice per cycle, this causes the arc to extinguish. There is a minimum flashover voltage needed to restrike the arc though. So the arc does not necessarily instantaneously restrike. There is a delay until the voltage reaches the flashover voltage again before it restrikes. At low system voltages, the fraction of time that the arc takes before it restrikes is much longer than when the system voltage is higher. Also the air temperature is the other major factor here which is dependent on available fault current which is the major determining factor in the arc power which is responsible for heating up the air. This is what drives the "arcing current" calculation that feeds the normalized incident energy calculation in IEEE 1584.

In a DC arc this arc current reduction shouldn't exist for obvious reasons...no current zeroes. However the arc restriking characteristic is what is attributed to the lower arcing current in the calculations. That does not necessarily mean that this is all that is going on in a DC arc. So there may yet be much more to learn when it comes to DC arc flash modelling but as of today without any laboratory testing data we're kind of stuck at doing theoretical calculations. So far the laboratory results definitely indicate that something else is going on but as of yet we don't know what that is. At best if it was a little better documented perhaps we could take the much more accurate time series model Wilkins developed and apply it to DC arcs except that obviously time isn't changing but the only paper Wilkins did on this model is missing a couple key items...notably the constants, so you can't just take the formulas in his paper and use them as-is.

1584-2018 will have some pretty major updates to the DC.

There are two DC methods at 1000Vdc, both based on the many DC arc fault tests that were performed, Paukert, and Stokes and Oppenlander. Paukert is only applicable up to 200mm gap distance, the Stokes and Oppenlander method up to and beyond 1500mm.

Paukert is basically a lookup table based on tests, Stokes and Oppenlander is a formula.

Both of these will then want to have the Arc Energy used in a calculation based on the enclosure size/arrangement & Wilkins reflectivity.

Raghu Veeraraghavan of etap has a good slideshow on DC arc flash,
http://sites.ieee.org/pes-essb/files/20 ... ghavan.pdf

No.

As per IEEE 1584 Draft 6 section 4.11: "Arc-flash incident energy calculation for DC systems is not part of this model. However, publication references XXXprovide some guidance for incident energy calculation."

Other than providing references there is nothing in the model discussing DC at all just as 1584-2002 was silent on the subject.

All of the data you mentioned (Paukert, etc.) are all about determining the V-I curve but they don't go to the point of actually calculating incident energy or have anything to do with it. Neal's model as mentioned is just the DC version of Lee (maximum power transfer argument). Ammerman goes further by taking Paukert, Opplander, etc., and using it as a theoretical DC arc model by basically modelling the arcing voltage as nearly constant (slowly increasing) which is what the V-I curves show for a power arc, then using this as the basis for the arc power and assuming perfect heat transfer and no arc "shape". So the basic model is different from the maximum power transfer and it's about 65% of the incident energy predicted by the maximum power transfer model but despite the improvements, it goes nowhere. And this is once again a purely theoretical model, NOT based on measurements.

As to the data, the Paukert and Stokes and Oppelander data is actually very old, not new, and not part of the joint IEEE/NFPA arc flash testing. There is no incident energy calculated. It's just V-I curves. The same data you mentioned and the Etap presentation are right out of the presentations that Ammerman did, as opposed to this paper that deals with the paltry amount of real world data out there and compares it to both the maximum power transfer and the Ammerman models and shows that even at low voltage, they are WAY off. No way you can say much, if anything, about application at >1 kV other than it's all theoretical so we're all guessing.

http://www.battcon.com/PapersFinal2012/ ... 0Flash.pdf