How to Read a Fault Record: A Step by Step Method

Most engineers learn to read fault records by watching somebody else do it, which means most of us carry a private method that has never been written down. The method below is the one that survives review: a fixed order of operations that establishes each fact before the fact that depends on it. It works on a relay disturbance record, a dedicated fault recorder file, or any COMTRADE file from any vendor.

The short answer

Read a fault record in a fixed order, and never look for the anomaly first. Verify the file and its scaling, establish the prefault state, find the inception sample, classify which phases and whether earth is involved, measure phasors in a window clear of both inception and clearance, check direction and impedance against the settings that were in service, then read the clearance from the digital channels. Each step supplies the baseline the next one is measured against.

Before you open the file

Write down what you already believe happened, in one sentence, before you look at a waveform. Something like “the line tripped on distance zone 1 and reclosed successfully”. This is not a formality. A fault record contains enough signal that a reader who arrives with no prior belief will find support for whichever explanation they test first, and a reader who arrives with an unstated belief will find support for that one. Stating it up front turns it into something the record can contradict.

Then collect the context the record itself does not carry:

• The settings that were in service. Not the current settings file, the one that was applied on the day. Reach, time grading, directional polarisation and CT and VT ratios all change what the record means.
• The system configuration at the time. Which sources were connected, whether the line was on a parallel circuit, whether the transformer neutral was earthed.
• What the relay reported. Trip targets, distance zone, measured fault location and reported fault type. You are going to check these, not assume them.

Step 1: Verify the record before you trust it

The configuration header is the first thing to read and the thing most often skipped. Four values decide whether the analysis is possible at all.

• Channel scaling. Check that primary and secondary quantities are what you expect. A wrong CT ratio in the header produces currents that are internally consistent, plot beautifully, and are wrong by a fixed factor. Compare the prefault load current against the known loading of the circuit as a sanity check.
• Sample rate. 16 to 32 samples per cycle is enough for phasors, impedance and fault type. Harmonic work needs enough samples to resolve the harmonic you care about, and high frequency transient work needs a recorder built for it.
• Prefault window length. If the trigger sits two cycles into the record, there is no reliable baseline and several of the steps below degrade to estimates.
• Nominal frequency and channel identity. Confirm the record is 50 or 60 Hz as expected, and that the channel names map to the phases you think they do. Phase labelling errors are common in older recorders and they silently invert direction decisions.

If more than one device recorded the same event, note that their clocks are unlikely to agree. Records have to be aligned on a common electrical feature before they can be compared, not on their timestamps.

Step 2: Establish the prefault state

Measure the steady state before inception and write the numbers down. Load current per phase, the three phase to earth voltages, the phase to phase voltages, and the direction of real power flow.

Two things fall out of this immediately. First, whether the system was already unbalanced, which changes how you read the fault. Second, the reference angle for every direction decision you are about to make. A fault record without a usable prefault window can still be analysed, but the directional conclusions become weaker and should be reported that way.

Step 3: Find the inception point

Inception is the first sample at which any channel departs from its prefault trajectory. Find it on the current channels: they move first, they move furthest, and they are less affected by capacitive voltage transformer transients than the voltages are.

Record the inception sample number, not just the time. Everything downstream is expressed in cycles from inception, and sample numbers survive the timestamp problems that time stamps do not.

Note the point on wave as well. A fault initiated at a voltage zero crossing produces the maximum DC offset in the current, which decays at a rate set by the system X/R ratio and which is the usual precursor to current transformer saturation. Knowing the point on wave tells you in advance whether to expect the measurement to degrade later in the record.

Step 4: Classify the fault

Compare each phase current against its own prefault magnitude, not against the other two phases. On an unbalanced feeder, a faulted phase that was lightly loaded can still measure lower than a healthy phase that was heavily loaded, and comparing phases against each other gets the classification wrong.

Then confirm earth involvement:

• Residual current is the vector sum of the three phase currents. Near zero means no earth path. Substantial residual means the fault involves earth.
• Zero sequence current is the same quantity divided by three, and is the form the relay actually used. Comparing your value against the relay reported value is a direct check on whether the relay saw what you see.
• Negative sequence current is present for every unbalanced fault and absent for a balanced three phase fault, which makes it the fastest way to separate those two cases.

The voltage picture should corroborate. A phase to earth fault depresses the faulted phase voltage and raises the healthy phase voltages on an unearthed or compensated system. A three phase fault depresses all three roughly equally and leaves negative and zero sequence near zero. If currents and voltages tell different stories, go back to step 1, because a scaling or labelling error is more likely than an exotic fault.

Step 5: Measure in a clean window

This is the step that quietly ruins more analyses than any other. Protection quantities are computed with a sliding discrete Fourier transform over a window one cycle long. A window that straddles inception contains part prefault and part fault data, and returns a phasor that is neither. The same is true of a window straddling clearance.

Place the measurement window so that it satisfies both conditions:

1. It begins at least one full cycle after inception.
2. It ends at least one full cycle before the current extinguishes.

A window that violates the first condition understates fault current and therefore inflates apparent impedance, which is how a correctly operating zone 1 element gets written up as having over reached. A window that violates the second does the same thing at the other end.

Sometimes no clean window exists. A fault cleared in two and a half cycles by a modern scheme leaves nothing that is both post inception and pre clearance. That is a legitimate finding and should be reported as one. Quoting a steady state phasor from a record that never reached steady state is how a report becomes confidently wrong.

Step 6: Check direction and reach

With a clean window you can compute the quantities the relay computed, and compare rather than guess.

• Apparent impedance from the loop voltage and loop current for the fault type you classified in step 4. Use the correct loop. A phase to earth loop uses the residual compensation factor from the settings; using the phase to phase loop instead produces an impedance that is wrong by roughly the compensation factor and looks plausible.
• The directional decision, from the angle between the operating current and the polarising quantity. Memorised polarising voltage matters here: on a close in three phase fault the polarising voltage collapses, and the relay uses stored prefault voltage instead.
• Fault location, remembering that a single ended estimate assumes a homogeneous system and no infeed from the remote end. Where a parallel circuit or a remote source exists, the single ended answer is biased and the size of the bias depends on the fault resistance.

Plot the impedance trajectory rather than reading a single point. The trajectory shows whether the measurement entered the characteristic, how long it stayed, and whether it was still moving when the relay operated. A single point cannot distinguish a solid in zone fault from a load encroachment that grazed the boundary.

Step 7: Read the clearance

The digital channels carry the protection and control decisions, and lining them up against the analogue channels is what turns a fault record into a timing narrative.

1. Element pickup to trip assertion is the protection operating time, which you compare against the setting and the expected element speed.
2. Trip assertion to current extinction is the breaker time including arcing. This is measured from the analogue current, not from a breaker auxiliary contact, because the auxiliary contact reports mechanism travel rather than current interruption.
3. Current extinction to reclose is the dead time, which you compare against the autoreclose setting.

Currents in the three phases rarely extinguish on the same sample. Each pole clears at its own current zero, so a spread of up to a third of a cycle across the three phases is normal and not evidence of a pole discrepancy.

Step 8: Ask whether the measurement itself was wrong

Every step above assumed the record faithfully represents the primary system. Before writing a conclusion, test that assumption explicitly. The three failure modes that most often invalidate an otherwise sound analysis are:

• Current transformer saturation, which understates fault current, inflates apparent impedance, and can create a false differential current when one side of a differential zone saturates and the other does not.
• Transformer magnetising inrush, which produces a real differential current from a healthy transformer and is separated from an internal fault by waveform shape and harmonic restraint rather than by magnitude.
• Frequency estimation breaking down during an unbalanced fault, which causes rate of change of frequency elements to operate on a frequency excursion that the system never experienced.

Capacitive voltage transformer transients belong on the same list. A CVT does not reproduce a sudden voltage collapse instantly, and the resulting transient can bias distance measurement for the first cycle or two after inception, which is precisely the window a high speed zone 1 element is working in.

The order of operations, condensed

1. Read the header. Confirm scaling, sample rate, prefault length, frequency, channel identity.
2. Measure and record the prefault state.
3. Find and record the inception sample and the point on wave.
4. Classify phase involvement against per phase prefault, then confirm with sequence components.
5. Place a measurement window clear of inception and clearance, or state that none exists.
6. Compute impedance, direction and location, and compare against the settings in service.
7. Align the digital channels and derive protection, breaker and dead times.
8. Test the record for saturation, inrush and CVT transients before concluding.

Common mistakes

• Starting from the anomaly. Opening the record, spotting the odd looking cycle and working outward from it. It produces a fast answer that is unfalsifiable because no baseline was ever established.
• Reading RMS where a phasor is needed. True RMS includes DC offset and harmonics. Protection elements operate on the fundamental phasor. During the first cycles of an offset fault the two differ substantially, and comparing an RMS value against a pickup setting that the relay applied to the fundamental will make a correct operation look wrong.
• Aligning multiple records on their timestamps. Unless every device was genuinely time synchronised, align on a shared electrical feature instead.
• Comparing against today’s settings file. Settings change. The only relevant file is the one that was applied when the event occurred.
• Reporting a fault location to three decimal places. Single ended location carries real uncertainty from fault resistance, remote infeed and line parameter accuracy. Quote it with the uncertainty or quote it coarsely.

Doing this in Detego

Detego runs this method on a COMTRADE file in the browser. The viewer computes phasors, symmetrical components, harmonics and apparent impedance from a window you place yourself, so step 5 stays under your control rather than being chosen for you. Digital channels overlay the analogue traces for the step 7 timing work, and multiple records from the same event can be aligned on an electrical feature rather than on their clocks.

Detego AI works the same order of operations and shows the intermediate quantities it used, so its conclusion can be checked step by step instead of taken on trust. It has access to the same calculation modules described above and to manufacturer relay documentation, which is what lets it comment on whether an element behaved as its settings required.

You can open a record and work through steps 1 to 7 in the online COMTRADE viewer.

Frequently asked questions

How do you read a fault record?

Work in a fixed order rather than looking for the anomaly first. Verify the file and its scaling, establish the prefault state, find the inception sample, classify which phases and whether earth is involved, measure phasors in a window that is clear of both inception and clearance, check direction and apparent impedance against the settings in service, then read the clearance from the digital channels.

Where should the measurement window go?

At least one cycle after inception and at least one cycle before the current extinguishes. A one cycle DFT straddling either transition mixes prefault and fault data into a single phasor, which understates fault current and inflates apparent impedance. If the fault cleared too quickly for such a window to exist, report that rather than quoting a phasor as if it were steady state.

What sample rate do you need?

16 to 32 samples per cycle is sufficient for fundamental frequency phasors, impedance and fault classification, which is what most protection relays record. Harmonic analysis needs enough samples to resolve the harmonic order of interest. Travelling wave work needs a dedicated recorder.

How do you tell which phases were faulted?

Compare each phase current against its own prefault magnitude rather than against the other phases, because unbalanced load hides faults on lightly loaded phases. Confirm earth involvement with residual current, and use negative sequence to separate a balanced three phase fault from every unbalanced case.

What is the most common mistake?

Trusting the measurement without asking whether the instrument transformers reproduced the primary quantity faithfully. A saturating CT understates fault current, which makes overcurrent elements look slow and distance elements look under reached, and the resulting report blames the relay for a decision that was correct on the data it was given.