When you're performing an arc flash hazard analysis, I'd like to know what your practice is in determining the arc flash energy for step down transformers. Do you rely on the primary OCP to determine the arc flash energy or do you use the secondary energy?

It's not clear what you're asking. The primary overcurrent protection device is just that...the overcurrent protection. Depending on the transformer design, you may need to treat the incident energy for the primary and secondary sides separately or take the worst case. As an example if you have a dry transformer in a housing or a pad mount style where the bushings for both the primary and secondary sides are in the same enclosure then run the calculation both ways and take the worst case incident energy (generally the secondary side). However if you have say a network transformer design with a separate termination compartment for primary and secondary sides then you can treat them separately. Generally speaking at 480/600 V class equipment with a 1500 kVA or larger transformer, the incident energy on the secondary side using primary protection only will be at or above 40 cal/cm^2. At 4160 V it takes a transformer about 10 times bigger to produce the same amount of incident energy. There isn't a "rule" here...just something that after doing hundreds of arc flash studies and mitigation efforts the pattern becomes very clear.

Since the impedance drives the fault current down quite a bit and drastically increases the opening time, generally the secondary side faults have the highest incident energy. However there are some ways to combat this. The first is to use differential relaying which is common on larger (>10 MVA) transformer protection schemes. Differential relaying usually has very fast (1 cycle for the relay and 3 for the interrupter) trip times and produces very low incident energy. A second scheme is to use cable protectors bolted directly to the lugs or with elbow connectors, a connector that includes a fuse inside it, or fuses on the secondary side mounted inside the transformer in oil-filled arrangements. These all provide secondary protection with little or no "unprotected" zone between the transformer itself and the outgoing terminations. Finally another scheme is to place bushing transformers on the secondary lugs. These can reduce the unprotected/underprotected zone to zero and can be used either in a differential (87) relaying scheme tripping a breaker on the primary side or simply feeding a second relay that trips the primary side breaker in a so-called "virtual breaker" arrangement giving the effect of a secondary breaker when in fact it doesn't exist. Another alternative that I've used is to mount a large molded case breaker mounted within a dry transformer enclosure and then provide a second traditional breaker or disconnect downstream at the distribution equipment. The first molded case breaker is never used for say LOTO purposes or shutting power off to the distribution equipment. It is serviced along with the transformer and it's only purpose in life is to catch arcing faults either in the feeder cable from the transformer secondary or at the distribution main overcurrent breaker. Thus in normal circumstances it is almost never needed or used. It doesn't have to be a breaker. I've also used Class L fuses and similar fast fusing protection with the goal being simply to minimize the zone of high incident energy.

Using the higher secondary current with the long primary clearing time should give the most conservative result. This would represent the incident energy between the secondary terminals and the first downstream protective device.

A transformer is just that. The primary side overcurrent protection will trip if the secondary side current is sufficiently large. A secondary side fault is seen as a dead short on the primary side AFTER the impedance (expressed in %Z) of the transformer is taken into account...it shows up as an impedance device. This gives you the "opening time" of a fault on the secondary busihings. This is done automatically by power system analysis software.

It is also not uncommon for the transformer impedance and/or cabling to have enough impedance to drive the opening time of the primary side protection to the point that it "never" (as in beyond 2 seconds) trips.

I've also had situations where the primary side fuses had to be increased beyond the "standard" rules (typically 250%). If there is a single large motor sized very close to the capacity of the transformer, the resulting starting current will usually trip the primary side overcurrent device if "standard" design rules of thumb are used....this is where using power system analysis software is useful for far more than just arc flash analysis.

Why do you need to do arc flash study for a transformer? What kind of interactive operation is required for transformer normally?

Really? Are you kidding?

Lots of reasons:
1. The lowest cost (high volume...what utilities buy) oil filled padmounts are typically designed for exposure to "the public" so the soil ample port and gauges are mounted inside the compartment(s). Depending on the size and age it could be live front air terminations, or dead front elbow style terminations and the former turns this into energized work even if it's just opening the door to read the highest temperature on the gauge.
2. Dry transformers require periodic cleaning/inspection of the fans if they have any, and IR inspections. Can't really check to see if a fan is working or get temperature readings with no power. At some point even if the practice is to take power off before starting work the test for absence of voltage and/or grounding is still energized work. In one case I'm very familiar with due to some incorrect labeling and confusing wiring electricians disconnected and tested externally at the disconnect. However the transformer was very much alive and testing a second time at the transformer itself would have avoided a near-miss situation altogether. I could go on to extend this example to virtually any system and dozens of examples of where failing to test for absence of voltage has resulted in a near miss or a serious shock but it should be intuitive why testing for absence of voltage is always energized work and is always required.
3. Basic troubleshooting occasionally occurs. On a sealed unit with elbow connectors and capacitive taps for medium voltage applications obviously this is totally unnecessary but that's not the norm for at least a lot of existing transformers.
4. Ducted network style transformers still occasionally have leaks and partial discharge issues in the bus bars at the throats. Same issues apply here as well.

I could go on for a long time on this but suffice to say that sooner or later you will have to access a transformer and second that you really can't rely on disconnecting means alone to ensure that absence of voltage exists every time.

And finally even if the equipment is "infrequently accessed" and thus doesn't require a label according to NEC on the day it does require access, technicians and engineers need to be able to find the answer if the equipment isn't labeled.

Paul, I wanted to know how everyone else is treating the arc flash hazard. Generally, I used the secondary (as it's almost always higher than the primary arc flash hazard) for smaller step down transformers. I wanted to hear what others are doing. It seems that relying on the primary to clear a secondary short is not a good practice.
Thank you all for your replies.

The secondary side of a transformer is always troublesome when it comes to incident energy. These are the viable solutions to it that I've found and actually put into service:
1. Keep transformers small enough. For instance with ANSI standard impedance with a 480 V transformer stay at or below about 1500 kVA and use more, smaller transformers. Alternatively raise the impedance although again we're trading off time/current and it is not always beneficial to do this. I haven't investigated the reverse case (lower impedance to trip faster on the primary side).
2. Use the primary side disconnect only if the equipment arrangement is such that there is no possibility of confusion. Where there is a single transformer and/or the disconnect is arranged right beside it, this really can't be screwed up as long as backfeeding is also arranged so that it can't occur, although protective grounding might also be able to mitigate this (never bothered).
3. Use circuit breakers or fuses built into the transformer housing. As an example on an excavator I worked on recently we purposely installed molded case breakers directly on the secondary lugs feeding MCC's. At the MCC end we used main breakers. Thus for disconnect/service purposes only the MCC main breakers are used. The transformer secondary breakers are set so high that they only trip for arcing or short circuit faults on the feeders to the MCC's. This was also done with a panelboard in another application. The breakers on the transformer can only be serviced by using the primary side disonnection means.
4. Another alternative is that at least one vendor makes elbow connectors with a fuse built into the connector. I haven't used this method personally.
5. Put bushing CT's on the transformer secondary bushings and wire these to a relay that shunt trips a circuit breaker mounted on the primary side. This arrangement fixes both issues. It reduces the size of the "high incident energy" zone to existing only inside the transformer tank itself and provides good, fast secondary protection that can be easily adjusted to protect against arcing faults.
6. Put fuses or circuit breakers outside the enclosure but upstream of existing equipment. This is not ideal because it doesn't eliminate the hazard but when retrofitting existing applications quite often the options are limited. As there is really no need for a disconnect here (incident energy is very high) it removes temptation to operate a handle if there isn't one there so a simple enclosure with fuses only (no disconnect) is ideal for the application.
7. Raise the secondary voltage. As voltage increases incident energy decreases in this particular situation. Obviously this is again another limited application situation but if you are already faced with a large above NEMA (500+ HP) motor it may make sense overall to switch to a 4160 V application instead of attempting to run lots of 350+ MCM cables, huge breakers, large motor housing for heat dissipation, etc., just to avoid the dreaded "medium voltage". Even though there is still a significant cost advantage to avoiding medium voltage VFD's up to around 1000 HP, this disadvantage is slowly coming down.