Loading documentation...

Fault Detection

Automatic fault inception and clearance detection using superimposed quantities and RMS change analysis.

In Detego

See the Fault Analysis Guide for how to interpret these results in the app. This page covers the mathematical theory.

Overview

Accurate fault timing is the foundation of every COMTRADE recording analysis. The fault inception time determines when to measure fault quantities (phasors, impedance, sequence components), while the clearance time determines the total fault duration and relay operating time. Detego automatically detects both events using a two-method approach: superimposed quantities as the primary detector, with RMS change as a fallback.

The detection algorithm processes all current and voltage channels independently for inception detection, taking the earliest inception across all channels. For clearance detection, the algorithm first restricts to current channels only (voltage recovery dynamics produce artifacts), then classifies the fault type at inception + 1 cycle to identify the faulted phases, and uses only faulted-phase current channels for clearance timing. All clearance results are refined using a Stationary Wavelet Transform (SWT) to pinpoint the exact breaker-operation transient.

Fault Inception Detection

Fault inception is the instant when the power system transitions from its pre-fault steady state to the faulted condition. Two detection methods are used in priority order.

Primary Method: Interpolated Superimposed Delta

The superimposed delta quantity isolates the change component by subtracting the signal value from one cycle earlier. Under steady state, this difference is zero. At fault inception, the sudden change in current produces a spike in the delta signal. The reference value is obtained via timestamp-based linear interpolation, ensuring correct operation with multi-rate COMTRADE segments where the sample rate changes mid-recording.

Interpolated Superimposed Delta

\Delta I(t) = I(t) - I_{\text{interp}}(t - T)

Where

$\Delta I(t)$ Change in current at time t
$I_{\text{interp}}(t - T)$ Linearly interpolated sample value at exactly one cycle prior
$T$ One power frequency cycle period

The detection algorithm computes the RMS of the delta signal over a half-cycle sliding window, then applies a guard period, noise-adaptive threshold, and persistence check:

Guard period: The first $N$ cycles (default 3 = 60 ms at 50 Hz) are skipped. During this period, the delta noise floor is measured using the median of half-cycle delta-RMS values. This avoids false triggers from ADC settling artifacts at the start of the recording.
Adaptive threshold: The inception threshold is the maximum of two components — the noise floor scaled by a multiplier, and the pre-fault signal RMS scaled by the superimposed threshold factor.
Persistence: At least $P$ consecutive half-cycle windows (default 2) must exceed the threshold before inception is confirmed. This eliminates single-window noise spikes while adding only $P \times 10$ ms of detection delay at 50 Hz.
Onset walkback: Once persistence is satisfied, the algorithm walks backward from the first trigger point to find the exact sample where the delta first exceeded half the threshold.

Inception Threshold

\text{threshold} = \max\!\left(N_f \times M,\; I_{\text{RMS,pre}} \times S\right)

Where

$N_f$ Delta noise floor (median of guard-period delta-RMS values)
$M$ Noise multiplier (~30 dB margin above the noise floor)
$I_{RMS,pre}$ Pre-fault signal RMS during the guard period
$S$ Configurable superimposed threshold factor (fraction of pre-fault RMS)

The dual-component threshold ensures sensitivity on both clean and noisy recordings.

Multi-Rate Support

Unlike integer-offset methods that assume a constant sample rate, the interpolated delta uses exact timestamp lookup with linear interpolation. This correctly handles COMTRADE recordings with multiple sample rate segments (e.g., 1600 Hz pre-fault switching to 4800 Hz during the fault), which are common in modern digital fault recorders.

Fallback Method: RMS Change

For de-energized channels (pre-fault RMS below 0.01) or recordings shorter than the guard period, the algorithm falls back to detecting step changes in the RMS magnitude:

Relative change threshold: A configurable threshold for relative change in RMS between consecutive computation windows.
Step change threshold: A configurable threshold for step changes in the RMS derivative (rate of change of RMS).
De-energized detection: For channels with near-zero pre-fault current, the algorithm looks for consecutive RMS values exceeding a configurable fraction of the peak channel RMS.

The RMS fallback is inherently slower than the superimposed method because the sliding RMS window requires approximately one full cycle to transition from the pre-fault to fault-level value.

Channel Filtering

Channels with negligible pre-fault signal (below $0.01$ baseline threshold) are automatically routed to the RMS fallback method. This prevents the superimposed delta threshold from collapsing to the noise floor and producing false triggers on inactive channels such as neutral CTs with no ground current or spare VT channels that carry only noise.

Detection Visualization

The diagram below illustrates both detection methods working on the same fault event. The superimposed delta (red) produces an immediate spike at inception, while the RMS magnitude (blue) takes approximately one cycle to fully transition to the fault-level value.

Fault inception detection using two methods. The superimposed delta (red) spikes within a single sample of the fault, while the RMS magnitude (blue) takes approximately one cycle to fully transition. Superimposed detection is the primary method; RMS is the fallback.

Inception Detection Pipeline

The complete inception detection pipeline processes the raw signal through three stages, each refining the signal to isolate the fault transition from noise and steady-state oscillation:

Inception Detection Pipeline — 3-Stage Signal Processing

Stage 1: Raw current waveform — 100A pre-fault, 500A fault at 200ms

Stage 2: Superimposed delta subtracts one cycle prior — flat during steady state, spikes at transitions

Stage 3: Half-cycle RMS of delta compared against adaptive threshold — 2 consecutive windows above threshold confirms inception

Raw signal → Superimposed delta: Subtracting the signal from one cycle earlier cancels the steady-state component. Only changes survive — the delta produces a "wave packet" (burst of oscillation lasting ~1 cycle) at each transition.
Delta → Half-cycle RMS: Taking the RMS over a sliding half-cycle window converts the oscillating delta into a smooth envelope suitable for thresholding.
Delta RMS → Threshold + Persistence: The guard period establishes the noise floor. The adaptive threshold combines noise-floor scaling with signal-proportional scaling. At least $P$ consecutive half-cycle windows must exceed the threshold before inception is confirmed.

Fault Clearance Detection

Fault clearance detection identifies when the circuit breaker has interrupted the fault current and the system has returned to a stable post-fault condition. Two methods are used: superimposed delta as the primary, with RMS transition as a fallback. The result is then refined using wavelet analysis to pinpoint the exact breaker-operation transient.

Clearance detection preferentially uses current channels only. Voltage channels are excluded from clearance aggregation because voltage recovery dynamics (RLC oscillations, tap-changer adjustments, recovery inrush) produce spurious late transient events that corrupt the fault window. If the recording contains no current channels (voltage-only recording), all channels are used as a fallback.

Primary: Superimposed Delta Clearance

The clearance algorithm uses the same interpolated superimposed delta signal as inception detection. The key insight: the clearance transient produces the last significant delta spike in the recording.

Compute delta + half-cycle RMS using the same interpolated helpers as inception detection.
Find peak delta-RMS after inception, and compute the clearance threshold as $\max(N_f \times M,\; \hat{\Delta}_\text{RMS} \times C)$ where $\hat{\Delta}_\text{RMS}$ is the peak delta-RMS and $C$ is a configurable clearance threshold factor.
Scan backward from the end of the recording to find the last sample where delta-RMS exceeds the threshold (the clearance transient).
Confirm settling: Scan forward from the last spike for $P$ consecutive half-cycle windows where delta-RMS drops below the clearance threshold.

Fallback: RMS Transition + Stability

When the superimposed delta approach does not find a clearance (e.g., recordings that end during the fault), the algorithm falls back to RMS-based detection:

Find peak fault RMS after inception. Identify the maximum RMS value during the fault window.
Detect transition: Scan forward to find where the RMS drops below 50% of the peak fault value.
Confirm stability: Verify the RMS is stable for a configurable number of consecutive samples with less than a configurable change threshold between successive values.

Minimum Fault Duration

A minimum fault duration of 10 ms (half a cycle at 50 Hz) is enforced. Events shorter than this threshold are rejected as noise or switching transients rather than true faults. This prevents the algorithm from detecting brief power system events (capacitor switching, load tap changes) as faults.

Fallback Clearance (Global Inception Anchor)

When per-channel clearance detection fails on all channels (e.g., on recordings where load current dominates fault current), the algorithm re-runs clearance detection anchored at the globally-detected inception time rather than each channel's individual inception.

This handles cases where the fault signature is difficult to see on current channels but voltage channels clearly show the inception. On load-dominated faults, per-channel inception detection on current channels may latch onto the breaker opening instead of the true fault start, leaving no room to find clearance afterward. By anchoring the search at the global inception (typically derived from voltage channels), the algorithm can detect the breaker opening as a large current drop from the correct starting point.

The fallback only fires when the primary path returned no valid clearance on any channel, so it cannot cause regressions on recordings where per-channel detection already works.

Multi-Event Detection (Auto-Reclose)

Power systems often experience auto-reclose sequences: the breaker trips to clear a fault, waits a dead time, then recloses. If the fault was transient (e.g., a lightning flashover), the reclose succeeds. If the fault is permanent, a second fault occurs and the breaker trips again. This produces multiple fault events in a single COMTRADE recording.

Architecture Overview

The multi-event detector uses a two-phase architecture: the first event is detected using the proven single-event inception primitive combined with a forward-scanning clearance detector, and subsequent events use delta scanning for inception with the same forward clearance approach.

Multi-Event Detection — Algorithm Architecture

Event 1 uses the proven single-event inception detector for consistent, reliable timing across all channels.

Events 2+ use raw delta scan to find subsequent inceptions after a quiet gap, then the same forward RMS clearance.

Post-filter removes events whose peak RMS is <10% of the recording's peak — eliminates noise blips on quiet channels.

Forward RMS Clearance

The forward clearance detector scans half-cycle RMS values forward from the inception point, using a running-peak tracker and multi-phase validation to distinguish real clearances from mid-fault current dips:

Forward RMS Clearance — 4-Phase Validation

Phase 2 (Sustained drop): RMS recovers above threshold within 3 cycles — mid-fault dip rejected, scanning continues

Phase 3 (Level check): Post-transition level (100A) is within 3× pre-fault (750A) — real clearance confirmed

Running peak: Tracks maximum RMS seen during forward scan — each event gets its own adaptive threshold

50% threshold: RMS must drop below half of running peak to be considered a candidate transition

Running-peak transition: Track the maximum RMS and detect when it drops below a configurable fraction of the peak.
Sustained-drop validation: Verify the drop persists for several power cycles. Mid-fault dips that recover are rejected.
Post-transition level check: Verify the stable level is physically consistent with a real clearance — either near pre-fault level or very small relative to fault magnitude.
Stability confirmation: Verify the post-clearance state is truly stable with minimal variation between successive RMS windows.

Subsequent Event Detection

After the first event clears, the algorithm scans the superimposed delta for additional inception spikes separated by quiet gaps, then applies the same forward clearance detection for each subsequent event.

Auto-reclose scenario with two fault events. Event 1 inception uses the proven superimposed delta detector; clearance uses forward RMS with running-peak tracking (orange/green dashes). Event 2 inception is found via delta scan after the dead-time quiet gap. Each event is numbered independently.

False Event Suppression

Three guards prevent false events from noise, ring-down transients, and mid-fault current variations:

Sustained-drop validation: Rejects mid-fault dips where the current briefly drops but recovers within a few cycles (CT saturation, partial breaker opening).
Post-transition level check: Rejects sustained drops where the post-transition level is still far above pre-fault and still a significant fraction of the fault peak.
Minimum event RMS: Each event must reach a configurable fraction of the recording's peak RMS to be retained.

Annotation Numbering

When a single fault is detected, annotations are labelled as before: "Inception (IA)" and "Clearance (IA)". When multiple events are detected on the same channel, they are numbered: "Inception #1 (IA)", "Clearance #1 (IA)", "Inception #2 (IA)", "Clearance #2 (IA)", and so on. An event with no clearance (fault persisting to the end of the recording) will only have an inception marker.

Classification Scan Window

When fault classification scans across the fault window, the scan region is tightened to exclude the first cycle after inception (DC offset settling) and the last cycle before clearance (breaker transient noise). Phasor filters produce spurious negative-sequence current during breaker operation (Kasztenny, SEL, 2019), which would cause false classification changes at the edges of the fault window.

A guard ensures the end of the scan window never precedes the start, so the scan degrades gracefully on very short faults.

Fallback Behavior

Short recordings and de-energized channels fall back to single-event detection, returning at most one inception/clearance pair.

AI Analysis Integration

The AI analysis pipeline also uses multi-event detection. When calling detect_fault_timing on a channel, the tool returns an events array listing all detected fault events with their inception and clearance times. The AI uses this to identify auto-reclose sequences and analyze each fault event independently.

X/R Ratio Bounding

The X/R Ratio computed signal derives the system X/R from the exponential decay of the DC offset in fault current: $X/R = 2\pi f \tau$ , where $\tau$ is the DC decay time constant. The DC offset is measured per half-cycle using $DC = (max + min) / 2$ , and an exponential fit $DC(t) = A \cdot e^{-t/\tau}$ extracts $\tau$ .

Without bounding, the exponential fit would run from the DC peak all the way to the end of the recording. After the fault clears, the current drops to near-zero noise, which inflates the fitted time constant and causes the X/R ratio to rise artificially. This produces misleading values in the post-fault region.

X/R ratio with and without fault window bounding. The unbounded fit (red dashed) includes post-clearance noise, inflating the time constant and causing X/R to rise artificially. The bounded computation (green solid) drops to zero at clearance, preventing misleading values.

Fault Window Bounding

The X/R computation now uses multi-event fault detection internally to identify fault windows. For each detected fault event:

Find the DC peak within the event window only
Fit the exponential decay within the event window only
Compute per-half-cycle X/R values within the window

All samples outside any fault window output 0, clearly indicating that no meaningful X/R measurement is available in those regions. When multiple fault events exist (auto-reclose), each event gets its own independent X/R calculation.

No Fault Detected

If the multi-event detector finds no fault events on the channel, the X/R computation falls back to the original unbounded behavior (fitting from global peak DC to end of recording). This preserves backward compatibility for recordings where the fault detection thresholds may not trigger.

Settings Reference

The fault detection algorithm exposes several configurable parameters via the global settings system.

Parameter	Default	Description
superimposedThreshold	Configurable	Signal-proportional threshold factor (fraction of pre-fault RMS)
guardCycles	Configurable	Number of cycles to skip at recording start (noise floor measurement)
persistenceCount	Configurable	Consecutive half-cycle windows required to confirm trigger
noiseMultiplier	Configurable	Noise floor scaling factor (~30 dB margin above quantization noise)
inceptionThreshold	Configurable	RMS fallback: relative change threshold for inception detection
clearanceThreshold	Configurable	Stability threshold for clearance confirmation
transitionDropThreshold	Configurable	RMS fallback: fraction of peak below which clearance transition is detected
minFaultDurationMs	Configurable	Minimum fault duration in ms (events shorter than this are rejected)

Engineering Applications

Accurate fault inception and clearance timing enables several critical protection engineering assessments:

Relay operating time: The difference between the trip contact closure (if recorded as a digital channel) and fault inception gives the relay operating time. This is compared against the relay's expected operating time for the measured fault current to verify correct relay performance.
Total fault clearance time: The difference between fault clearance and inception gives the total fault duration, which includes relay operating time plus breaker interrupting time. This is critical for thermal damage assessment and arc flash energy calculations.
Breaker failure protection timing: If the fault persists beyond the expected clearance time, breaker failure protection must initiate backup tripping. Accurate inception timing is essential for assessing whether breaker failure timers are correctly set.
Auto-reclosing coordination: The dead time between fault clearance and reclosure must be sufficient for arc de-ionization. Accurate clearance detection enables verification of auto-reclose dead times.

Timing Accuracy

The inception time accuracy depends on the recording sample rate. At 4000 samples/second (typical for modern digital fault recorders), the time resolution is 0.25 ms. At lower sample rates, the detected inception may lag the true inception by up to one sample period. The superimposed method is inherently more accurate than the RMS fallback because it responds within a single sample.

PreviousProtection Quality NextFault Classification

Loading documentation...

Fault Detection

Automatic fault inception and clearance detection using superimposed quantities and RMS change analysis.

In Detego

See the Fault Analysis Guide for how to interpret these results in the app. This page covers the mathematical theory.

Overview

Fault Inception Detection

Fault inception is the instant when the power system transitions from its pre-fault steady state to the faulted condition. Two detection methods are used in priority order.

Primary Method: Interpolated Superimposed Delta

Interpolated Superimposed Delta

\Delta I(t) = I(t) - I_{\text{interp}}(t - T)

Where

$\Delta I(t)$ Change in current at time t
$I_{\text{interp}}(t - T)$ Linearly interpolated sample value at exactly one cycle prior
$T$ One power frequency cycle period

The detection algorithm computes the RMS of the delta signal over a half-cycle sliding window, then applies a guard period, noise-adaptive threshold, and persistence check:

Guard period: The first $N$ cycles (default 3 = 60 ms at 50 Hz) are skipped. During this period, the delta noise floor is measured using the median of half-cycle delta-RMS values. This avoids false triggers from ADC settling artifacts at the start of the recording.
Adaptive threshold: The inception threshold is the maximum of two components — the noise floor scaled by a multiplier, and the pre-fault signal RMS scaled by the superimposed threshold factor.
Persistence: At least $P$ consecutive half-cycle windows (default 2) must exceed the threshold before inception is confirmed. This eliminates single-window noise spikes while adding only $P \times 10$ ms of detection delay at 50 Hz.
Onset walkback: Once persistence is satisfied, the algorithm walks backward from the first trigger point to find the exact sample where the delta first exceeded half the threshold.

Inception Threshold

\text{threshold} = \max\!\left(N_f \times M,\; I_{\text{RMS,pre}} \times S\right)

Where

$N_f$ Delta noise floor (median of guard-period delta-RMS values)
$M$ Noise multiplier (~30 dB margin above the noise floor)
$I_{RMS,pre}$ Pre-fault signal RMS during the guard period
$S$ Configurable superimposed threshold factor (fraction of pre-fault RMS)

The dual-component threshold ensures sensitivity on both clean and noisy recordings.

Multi-Rate Support

Fallback Method: RMS Change

For de-energized channels (pre-fault RMS below 0.01) or recordings shorter than the guard period, the algorithm falls back to detecting step changes in the RMS magnitude:

Relative change threshold: A configurable threshold for relative change in RMS between consecutive computation windows.
Step change threshold: A configurable threshold for step changes in the RMS derivative (rate of change of RMS).
De-energized detection: For channels with near-zero pre-fault current, the algorithm looks for consecutive RMS values exceeding a configurable fraction of the peak channel RMS.

The RMS fallback is inherently slower than the superimposed method because the sliding RMS window requires approximately one full cycle to transition from the pre-fault to fault-level value.

Channel Filtering

Detection Visualization

Inception Detection Pipeline

The complete inception detection pipeline processes the raw signal through three stages, each refining the signal to isolate the fault transition from noise and steady-state oscillation:

Inception Detection Pipeline — 3-Stage Signal Processing

Stage 1: Raw current waveform — 100A pre-fault, 500A fault at 200ms

Stage 2: Superimposed delta subtracts one cycle prior — flat during steady state, spikes at transitions

Stage 3: Half-cycle RMS of delta compared against adaptive threshold — 2 consecutive windows above threshold confirms inception

Raw signal → Superimposed delta: Subtracting the signal from one cycle earlier cancels the steady-state component. Only changes survive — the delta produces a "wave packet" (burst of oscillation lasting ~1 cycle) at each transition.
Delta → Half-cycle RMS: Taking the RMS over a sliding half-cycle window converts the oscillating delta into a smooth envelope suitable for thresholding.
Delta RMS → Threshold + Persistence: The guard period establishes the noise floor. The adaptive threshold combines noise-floor scaling with signal-proportional scaling. At least $P$ consecutive half-cycle windows must exceed the threshold before inception is confirmed.

Fault Clearance Detection

Primary: Superimposed Delta Clearance

Compute delta + half-cycle RMS using the same interpolated helpers as inception detection.
Find peak delta-RMS after inception, and compute the clearance threshold as $\max(N_f \times M,\; \hat{\Delta}_\text{RMS} \times C)$ where $\hat{\Delta}_\text{RMS}$ is the peak delta-RMS and $C$ is a configurable clearance threshold factor.
Scan backward from the end of the recording to find the last sample where delta-RMS exceeds the threshold (the clearance transient).
Confirm settling: Scan forward from the last spike for $P$ consecutive half-cycle windows where delta-RMS drops below the clearance threshold.

Fallback: RMS Transition + Stability

When the superimposed delta approach does not find a clearance (e.g., recordings that end during the fault), the algorithm falls back to RMS-based detection:

Find peak fault RMS after inception. Identify the maximum RMS value during the fault window.
Detect transition: Scan forward to find where the RMS drops below 50% of the peak fault value.
Confirm stability: Verify the RMS is stable for a configurable number of consecutive samples with less than a configurable change threshold between successive values.

Minimum Fault Duration

Fallback Clearance (Global Inception Anchor)

The fallback only fires when the primary path returned no valid clearance on any channel, so it cannot cause regressions on recordings where per-channel detection already works.

Multi-Event Detection (Auto-Reclose)

Architecture Overview

Multi-Event Detection — Algorithm Architecture

Event 1 uses the proven single-event inception detector for consistent, reliable timing across all channels.

Events 2+ use raw delta scan to find subsequent inceptions after a quiet gap, then the same forward RMS clearance.

Post-filter removes events whose peak RMS is <10% of the recording's peak — eliminates noise blips on quiet channels.

Forward RMS Clearance

Forward RMS Clearance — 4-Phase Validation

Phase 2 (Sustained drop): RMS recovers above threshold within 3 cycles — mid-fault dip rejected, scanning continues

Phase 3 (Level check): Post-transition level (100A) is within 3× pre-fault (750A) — real clearance confirmed

Running peak: Tracks maximum RMS seen during forward scan — each event gets its own adaptive threshold

50% threshold: RMS must drop below half of running peak to be considered a candidate transition

Running-peak transition: Track the maximum RMS and detect when it drops below a configurable fraction of the peak.
Sustained-drop validation: Verify the drop persists for several power cycles. Mid-fault dips that recover are rejected.
Post-transition level check: Verify the stable level is physically consistent with a real clearance — either near pre-fault level or very small relative to fault magnitude.
Stability confirmation: Verify the post-clearance state is truly stable with minimal variation between successive RMS windows.

Subsequent Event Detection

False Event Suppression

Three guards prevent false events from noise, ring-down transients, and mid-fault current variations:

Sustained-drop validation: Rejects mid-fault dips where the current briefly drops but recovers within a few cycles (CT saturation, partial breaker opening).
Post-transition level check: Rejects sustained drops where the post-transition level is still far above pre-fault and still a significant fraction of the fault peak.
Minimum event RMS: Each event must reach a configurable fraction of the recording's peak RMS to be retained.

Annotation Numbering

Classification Scan Window

A guard ensures the end of the scan window never precedes the start, so the scan degrades gracefully on very short faults.

Fallback Behavior

Short recordings and de-energized channels fall back to single-event detection, returning at most one inception/clearance pair.

AI Analysis Integration

X/R Ratio Bounding

Fault Window Bounding

The X/R computation now uses multi-event fault detection internally to identify fault windows. For each detected fault event:

Find the DC peak within the event window only
Fit the exponential decay within the event window only
Compute per-half-cycle X/R values within the window

No Fault Detected

Settings Reference

The fault detection algorithm exposes several configurable parameters via the global settings system.

Parameter	Default	Description
superimposedThreshold	Configurable	Signal-proportional threshold factor (fraction of pre-fault RMS)
guardCycles	Configurable	Number of cycles to skip at recording start (noise floor measurement)
persistenceCount	Configurable	Consecutive half-cycle windows required to confirm trigger
noiseMultiplier	Configurable	Noise floor scaling factor (~30 dB margin above quantization noise)
inceptionThreshold	Configurable	RMS fallback: relative change threshold for inception detection
clearanceThreshold	Configurable	Stability threshold for clearance confirmation
transitionDropThreshold	Configurable	RMS fallback: fraction of peak below which clearance transition is detected
minFaultDurationMs	Configurable	Minimum fault duration in ms (events shorter than this are rejected)

Engineering Applications

Accurate fault inception and clearance timing enables several critical protection engineering assessments:

Relay operating time: The difference between the trip contact closure (if recorded as a digital channel) and fault inception gives the relay operating time. This is compared against the relay's expected operating time for the measured fault current to verify correct relay performance.
Total fault clearance time: The difference between fault clearance and inception gives the total fault duration, which includes relay operating time plus breaker interrupting time. This is critical for thermal damage assessment and arc flash energy calculations.
Breaker failure protection timing: If the fault persists beyond the expected clearance time, breaker failure protection must initiate backup tripping. Accurate inception timing is essential for assessing whether breaker failure timers are correctly set.
Auto-reclosing coordination: The dead time between fault clearance and reclosure must be sufficient for arc de-ionization. Accurate clearance detection enables verification of auto-reclose dead times.

Timing Accuracy

PreviousProtection Quality NextFault Classification

Fault Detection

Primary Method: Interpolated Superimposed Delta

Fallback Method: RMS Change

Channel Filtering

Primary: Superimposed Delta Clearance

Fallback: RMS Transition + Stability

Wavelet Refinement (SWT)

Minimum Fault Duration

Fallback Clearance (Global Inception Anchor)

Architecture Overview

Forward RMS Clearance

Subsequent Event Detection

False Event Suppression

Annotation Numbering

Classification Scan Window

Fallback Behavior

Fault Window Bounding

Fault Detection

Primary Method: Interpolated Superimposed Delta

Fallback Method: RMS Change

Channel Filtering

Primary: Superimposed Delta Clearance

Fallback: RMS Transition + Stability

Wavelet Refinement (SWT)

Minimum Fault Duration

Fallback Clearance (Global Inception Anchor)

Architecture Overview

Forward RMS Clearance

Subsequent Event Detection

False Event Suppression

Annotation Numbering

Classification Scan Window

Fallback Behavior

Fault Window Bounding