How we measure
Every figure on this site comes from DFA recordings at the substation, rebuilt from the live analysis. Here is what we record, how a recording becomes a fault sequence, how we decide what happened, and how we check our work.
One monitor per feeder
Each DFA device sits in the substation, one per circuit, on the feeder CTs and bus PTs already there.1 It samples with more resolution than a relay or a fault recorder and triggers on changes of a few amperes, so it records events that never trip a breaker - including incipient events that may cause faults and failures in the future.
Most recordings are routine: motor starts, load swings. Classification algorithms embedded in the DFA device attempt to give every recording a human-readable label. The fault study uses only the recordings classified as faults on the monitored feeder.
From recordings to fault sequences
Protection works in sequences: fault, trip, reclose, and perhaps again. This study looks at entire sequences, rather than individual faults. This includes looking forward and backward in time for related events so we have the whole picture of what happened.
Human experts reviewed 9.8% of the fault sequences in the study.
Which device acted, and what was lost
The monitor sees all of the feeder's current. When every phase goes dead, the substation breaker opened. When the feeder stays energized but load disappears, a downstream device acted. A clear faster than any recloser can open, or a fuse curve that fits both the fault current and the lost load, points to a fuse. A reclosing pattern, or a loss too large for any fuse, points to a recloser. Only fuses the utility actually deploys count.
The fault study tries to answer the following questions of every fault sequence, in order, to decide the outcome:
- 1Was the whole feeder (every phase) left dead - for how long?
- 2Was load lost at the end of the record?
- 3No load loss confirmed at the substation - what does the protective device's own record show?
- 4Nothing operated - how did the burst end?
Each reclose attempt ends with one outcome:
- Likely reclose success
- Nothing lost, or the lost load back within minutes.
- Reclose failure, high confidence
- Confirmed load loss after fault.
- Uncertain
- Unconfirmed load loss or no resolvable outcome.
A sequence with no reclose attempt is one where no device closed back in: mostly a fuse clearing the fault, or a fault that put itself out.
Loads move, so we print ranges
Steady state is a useful fiction. Within a recording, per-cycle current wanders 0.23 A per phase at the median (0.77 A at the 90th percentile). Five-minute interval data moves 2.0 A (4.8 A) in the hour before a fault.
A loss smaller than that swing cannot be told apart from normal load. So every loss figure is a range: the low end counts only losses the recording shows; the high end adds load that went missing without proof it stayed off. Load that comes back within minutes counts as a likely success.
The monitor sits at the feeder head. It sees all of the current, but never a downstream device directly: which device acted is read from the waveform.
What the fleet covers
64,382 fault sequences from 676 feeders at 22 US utilities, Dec 2015 to Oct 2026: 2,244 monitor-years of recordings.
The fleet is not a random sample of US feeders. DFA circuits tend to be longer, more overhead, more rural and more exposed to wildfire risk than average. 38% of monitor-years come from 25 kV circuits, and 16% from three-wire circuits. Circuits outside the US (9% of fault sequences) are excluded.
Weather comes from hourly reanalysis at the substation, not at the fault.
The site rebuilds from the live analysis. Numbers move as new recordings arrive and as the analysis improves.
Sources
The fault study numbers its own sources at the foot of its page. The library lists every reference in the deck, plus the working bibliography of our upcoming history of automatic reclosing.