The data problem the 2002 edition had

The 2002 edition's empirical model was built from a real but comparatively limited test program — approximately 300 tests, per clause G.1 of the 2018 edition. That data set covered a real but bounded range of system voltages, gap distances, and enclosure sizes. Outside that range, the 2002 equations were an extrapolation, not a measurement — which matters a great deal for equipment at the edges of what utilities and industrial facilities actually run.

What the 2018 test program did differently

The 2018 revision was preceded by a much larger, multi-year joint IEEE/NFPA testing program — over 1,860 tests, per the same clause, across a wider gap range. That's not just "more data" in the abstract — it's data on electrode configurations and enclosure sizes the 2002 model did not take as inputs, which is why the 2018 model treats them explicitly. The 2018 edition also checked its new model against the 2002 test results.

Structural changes to the model itself

A few changes are structural, not just "different coefficients":

Electrode configuration became explicit. The 2018 edition formally defines five configurations — VCB, VCBB, HCB, VOA, HOA — and the equations depend on which one applies. The 2002 model described equipment more coarsely: open or box, plus an equipment class and its typical gap. See our companion piece, Arc Flash Electrode Configurations, Explained, for what each one means and how much difference it actually makes.

Arcing current variation. Both editions run a second case at a reduced arcing current, because a lower arcing current can mean a protective device clears the fault more slowly — which can produce higher incident energy than the average arcing current would suggest. The 2002 edition did this below 1000 V with a fixed 85% of the predicted arcing current (clause 5.2, as amended by IEEE 1584a-2004). The 2018 edition replaces the fixed 85% with a variation correction factor that depends on electrode configuration and voltage, and applies it from 208 V to 15 kV (clause 4.5). A study that skipped the reduced-current case is worth a closer look under either edition.

Enclosure size correction. The 2002 model used one typical enclosure for each equipment class. The 2018 model adds an enclosure size correction factor based on height, width, and depth (clause 4.8).

System grounding. The 2002 model took grounding as an input, giving ungrounded and high-resistance grounded systems higher results than grounded ones (clause 5.3). The 2018 model does not use grounding, because the IEEE/NFPA collaboration tests did not show a significant grounding effect (clause G.7.11).

What this means if you have a 2002-vintage study

A study built on the 2002 model isn't automatically wrong, but it was built on a narrower tested envelope than the current standard — which matters most for equipment sitting near the edges of common voltage classes or enclosure sizes, exactly where the 2002 model was extrapolating rather than measuring. NFPA 70E already calls for arc flash risk assessments to be reviewed on a regular cycle and whenever the electrical distribution system is modified; a study still on the 2002 basis is a reasonable thing to flag for that review specifically, rather than assuming a study is current just because no equipment has changed since it was performed.

Deciding whether a specific study needs revisiting is a different question from what the model changed, and it is answerable from the study document itself — which edition and model it states, whether the equipment falls inside that model's applicability limits, whether the reduced-arcing-current case was run, and whether the modeled system still matches the plant. Does your arc flash study need to be redone? walks through those four checks. (Ohmics is also a PEFG, PLLC product.)