An awful lot can happen in 10 cycles

Here is a high level analysis of a fault which occurred in our system a few months back. The interesting part of it, to me, was how many different configurations the fault went through in 10 cycles. This fault occurred on a 12 kV bus which feeds a load bus as well as a shunt capacitor bank for power factor correction. There is a current limiting reactor applied to reduce fault levels, and the fault started in the vicinity of this reactor (probably due to an animal climbing up the reactor).

The reactor is a three phase unit, with all three phases stacked on top of each other.

The oscillography above is a "RAW" event record from an SEL-351 relay. It is sampled at 16 samples/cycle and does not have the 60 Hz filtering that is applied before it is sent to the protection elements. The oscillography below is from a normal "Filtered" event record, sampled at 4 samples/cycle. Notice that the Filtered record is devoid of the DC offset visible on the current waveforms, as well as harmonics visible on the voltage waveforms. As a general rule, if you want to know what the relay is doing, look at the filtered event records, but if you want to know what the power system is doing, look at unfiltered, high resolution events.

We will now shift from looking at the waveforms to looking at current magnitudes, and break the event into 6 separate segments.

Segment A is our prefault. If we switch to phasor view, we can see that we have balanced currents and a good power factor.

Segment B is our original fault. We have current in two phases, voltage drop in the same two phases where the voltages move closer together, and we have no zero sequence current, so we tag this one as a phase to phase fault. The fault current is around 14 kA, which is intriguing because our fault current limiting reactors are supposed to limit 3 fault current to somewhere under 8 kA. This is where the physical layout of the system in question becomes invaluable. If we look at the drawing of the reactor, we can see that there is a location where the input of one phase and the output of the adjacent phase are not very far apart, so we hypothesize that perhaps this phase to phase fault was not on either the source or the load side of the reactor, but from the source side of one phase to the load side of another. By doing this, the phase to phase fault which would normally have a max current of VLL/(2 * Xr) (ignoring source impedance and arc resistance) can now have a max current of VLL/Xr, which works out to be around root 3 times the current you would get for a 3 phase fault after the reactor. 8 kA * sqrt(3) = 13.8 kA, so this hypothesis seems reasonable.

Segment C sees the fault type remain the same, but the fault current suddenly rise to around 36 kA. This is near the bus fault level before the reactor, so we can only assume that the fireball from the arc has now expanded enough to engulf the source terminal on both faulted phases.

Segment D is very interesting, although quite short-lived. Current in C phase suddenly rises, but not to bus fault levels, but around 12 kA. The explanation here is that the fireball has now engulfed C phase load side terminal, so we have a phase to phase fault on A and B phases on the source side of the reactor, and that is connected to the load side of the reactor on C phase. When you have a bolted phase to phase fault, the phase voltages collapse towards each other, and in the limit, each faulted phase voltage is of 0.5 pu magnitude and of an angle opposite to the unfaulted phase. So from the point of view of this A phase to fireball fault, we have 1.5 pu voltage across the current limiting reactor, which should result in around 1.5 * 8 kA = 12 kA of current. That makes sense. On to the next segment...

Disregard the image below. The image below that is the correct one.

In segment E, we see neutral current, which up until now has remained solidly at zero, start to come up. This is our indication that the fireball has now expanded enough to intercept something grounded. The configuration of the system is starting to get pretty complicated here. We have A and B phases connected together directly to the fireball, C phase connected to the fireball through a current limiting reactor, and the whole throbbing flaming mass connected to ground. Coming up with an estimate of ground current for this fault is difficult since we have different impedances in each leg, but we can attempt using superposition to calculate it. Let's consider this to be the combination of an A-B-G fault with no current limiting reactors and a C-G fault with a current limiting reactor. We know the C-G fault should be around 8 kA, and Aspen OneLiner says 27 kA of 3I0 for the A-B-G fault. Since these currents are mostly antiphase to each other, we subtract them and get an estimate of 19 kA in the neutral, which doesn't quite agree with the measured 14 kA, but is maybe close enough to let us say that this is what happened.

Segment F sees all currents drop. This is due to ONE of the high side breakers clearing. One of the high side breakers (the one whose 52A contact is connected to IN102) interrupts and disconnects half of the fault current sources from the fault. A little over 4 cycles later, the breaker whose 52A contact is connected to IN101 follows and the fault is cleared. I don't know why the breaker was late tripping, but it has been looked into.

Going back to the layout drawing of the current limiting reactor, we can visualize the progression of the fault.

Now, let's look at a couple of tinier details from the event record.

Note how during the phase to phase fault on the source side of the current limiting reactor, that the A and B phase voltages are flat topped. This is an indication that there is a very small impedance between our VT and an arcing fault. Arcs have a negative incremental resistance characteristic (as the current increases, the resistance decreases), which results in arcs having a fairly consistent voltage drop regardless of the current through them. The voltage on C phase remains sinusoidal because the voltage drop is dominated by the current limiting reactor.

Here I have overlaid the Raw and the Filtered event record currents on the same chart. The Raw currents are the darker traces. Note how at the beginning, the Filtered record nearly leaves out the first half cycle, and at the end, how the digital filters ring for nearly an additional cycle after fault current has been interrupted.