Anybody has any thought on the appropriate settings for the pickup current and time delay for optical relays:

My first thought would be to set the pickup above at least two times the nominal circuit current and the time delay to the minimum allowed by the settings.

Any other thoughts?

First, the way you phrased your question seems puzzling to me.

Generally speaking for a given configuration of equipment and a given time delay, we can calculate an incident energy with reasonable accuracy. If the time delay is long enough, this includes an arcing current. If we then adjust the relaying to trip at a point below the arcing fault current, this controls the arcing time and linearly, the incident energy.

This last point is key. If we have a target incident energy (say 8 cal/cm^2) and a known incident energy, since incident energy is linear with time, we can simply calculate a ratio to determine a new required opening time. Subtracting the opening time of the breaker gives us the required tripping time.

If the tripping time is over 1 cycle (0.0166667 seconds), we will be able to achieve this with a conventional instantaneous relay. And that's where the fun begins. Using instantantaneous overcurrent will of course cause coordination issues with any overcurrent devices downstream of the breaker. So if we have at least 2+ cycles available the coordination issue can generally be solved using conventional approaches...that is, adjust the devices based on a time-current curve so that we provide adequate coordination between devices.

However once we drop below that threshold, things get much more tricky. If you don't mind the manual approach, simple "maintenance switches" (purposely ignoring coordination for a period of time) does the trick and is the least expensive approach in most cases. For a 2 cycle tripping time, we can use zone selective relaying.

At 1 cycle tripping, the options are limited. This will include "arc flash relays" and bus differential tripping schemes which trip essentially instantaneously. The "arc flash" relays can get down to as low as about 1/4 cycle (4 ms) at best. Why this low? That's just long enough to see the first half of a sine wave and make a call that it is time to trip.

Keep in mind that up until this point, I never needed an arc flash relay. The only time they might be a consideration is when coordination becomes problematic and/or the trip time becomes very short. And even if its very short, we're already capable of doing 1 cycle "instantaneous" tripping...the arc flash relay just sometimes makes coordination a bit less onerous to achieve. In most cases though it is often acceptable to allow some miscoordionation in niche cases anyways as long as everyone understands what is going on.

If the time required is negative, there are only two practical options. Either we need to introduce some sort of current limiting into the system, we have to divide up the power sources (more, smaller transformers), or we need to get a faster tripping device. The current "speed kings" for larger breakers are various "arc terminators" which can trip as fast as 1/4 cycle, making 1/2 cycle tripping a possibility.

And this points out a major problem with "dual" tripping (arc flash AND current). An optical relay by itself does everything required. The sole purpose for also monitoring current is to avoid nuisance tripping if for instance an electrician takes out a cell phone and snaps a picture of something in the cabinet while it is energized. With traditional CT's, CT saturation has to be considered because otherwise tripping can be significantly delayed just because the CT signal takes a while to peak. So no matter what the relay settings are, you can't overcome the physics of the CT being used to detect high current.

An alternative (and the arc flash relay companies sell these) is to use a Rogowski coil. This is a special high frequency CT that does not saturate so it will output a current regardless of the rise time (dI/dt). Since an arcing fault appears as a very fast rise time, some arc flash relays on the market actually use the high dI/dt signal itself as a trip signal. When combined with the optical sensor, a high dI/dt by itself is likely just a switching transient. An optical signal by itself is most likely just a camera flash. But seeing both signals indicates presence of an arcing fault for tripping purposes.

So the reason the question is puzzling is because if you are doing standard 50/51 relaying, you don't need an optical relay in the first place. If you need an optical relay in the first place then the 50/51 relay should be instantaneous tripping and the trip setting is based on triggering just below the minimum calculated arcing fault current, and most likely a 50/51 relay isn't the best choice unless the only reason for using the optical relay is to overcome a coordination issue, one that most sites can live with anyways.

Depending on the load characteristics you might need to set your current threshold higher than 2 x rated. Consider motor starting currents or transformer inrush. I agree that generally speaking the time delay should be kept as low as possible. Motor starting could be up to 6x the rated motor current. Transformer inrush could be 12x at 0.1 seconds and more at less than 0.1s, which might require a longer delay on your optical trip.

Long term trip settings for motors are generally 6X. But instantaneous can be much higher. With NEMA B it was traditionally 13X. With energy efficienct motors it can be as high as 22-23X. NEC has increased the limit now up to 17X (from 13X about 15 years ago) but with IEC style motors it can be much higher.