AI Analysis

Detego uses Claude, Anthropic's large language model, to perform comprehensive fault analysis. The AI acts as a virtual protection engineer — it examines your recording, takes measurements, and synthesizes a structured fault report.

All computations are performed locally in your browser against the actual COMTRADE data. No raw waveform data is sent to the AI — only computed results (phasor values, impedances, harmonic spectra, etc.). The AI receives these results and produces a structured report covering fault identification, fault characteristics, protection assessment, and recommendations.

The AI engine has access to a comprehensive suite of analytical tools covering phasors, impedance, harmonics, fault detection, and protection assessment. It automatically selects the appropriate measurements based on your recording content and available channels.

Fundamental Measurements

Advanced Analysis

Fault and Protection

When multiple fault events are detected (e.g., auto-reclose sequences), each event is analyzed independently with its own inception and clearance times.

The AI report is organized into sections, each covering a specific aspect of the fault analysis:

Report Header

• Station and device identification
• Recording timestamp
• Fault type classification (e.g., "Single Line-to-Ground A-G")
• Severity rating (Low, High, or Critical)
• Verdict — the overall protection assessment

Verdict Categories

Typical Sections

• Fault Identification — What happened, when, and which phases were involved
• Fault Characteristics — Impedance, sequence components, X/R ratio, fault current magnitude
• Protection Assessment — Trip/no-trip determination, relay operating time, directional assessment
• Power Quality — Voltage sag depth and duration
• Recommendations — Suggested follow-up actions

Embedded Charts

Reports may include waveform charts showing key moments in the recording (fault inception, relay operation, clearance) with annotations highlighting important events. These charts are rendered from the actual COMTRADE data.

After receiving an AI report, a Chat button appears in the report toolbar. Clicking it opens a floating chat interface where you can ask follow-up questions about the analysis findings.

The AI retains full context from the original analysis — including the report content, recording metadata, and any relay settings or notes you provided. This means you can ask targeted questions without re-explaining the scenario.

Follow-up questions are useful for:

• Clarifying findings — "Why did you classify this as a B-G fault?"
• Exploring alternatives — "What would change if the CT ratio was different?"
• Requesting detail — "Can you explain the impedance measurement in more detail?"
• Investigating specific sections — "What does the negative-sequence current magnitude tell us?"

Conversations are persisted per analysis, so you can close the chat and return to it later. The message count badge on the Chat button shows how many exchanges have occurred.

When you re-analyze a recording that has been analyzed before, Detego preserves all prior analyses. A History picker appears in the report toolbar showing a numbered list of analyses with timestamps, fault types, and assessment verdicts.

Click any entry to view that historical report. This is useful for:

• Comparing analyses — See how a report changes after providing relay settings or additional context via user notes.
• Tracking changes — If a recording is re-analyzed after software updates, you can compare the previous and current findings.
• Preserving work — Each analysis is timestamped and permanently stored, so nothing is lost when you run a new analysis.

The most recent analysis is shown by default. The history picker indicates which report is currently displayed.

When the recording contains sufficient data, the AI performs specialized protection assessments:

Directional Assessment — Phase Faults

For phase faults, the AI uses the 90-degree quadrature connection to determine fault direction. Standard polarizing pairs are used (e.g., Phase A current with BC voltage). Results include:

• Direction: Forward, Reverse, or Low Confidence (when the margin from the operate/restrain boundary is below a minimum threshold)
• Directional angle and P₀ (directional power quantity)
• Margin percentage (distance from the operate/restrain boundary)
• A polar diagram showing the operating point

Directional Assessment — Earth Faults

For earth faults, the AI uses residual (zero-sequence) quantities with an RCA appropriate to the network earthing type. Four earthing configurations are supported:

• Solidly earthed distribution (RCA = -45°)
• Solidly earthed transmission (RCA = -60°)
• Insulated (RCA = +90°)
• Petersen coil / wattmetric (RCA = 0°) — uses the wattmetric method, where the sign of P₀ determines direction rather than angular margin

The earthing type is detected automatically from the zero-sequence fault measurements. When a dedicated zero-sequence current channel is available (e.g., CBCT / core balance CT), it is used in preference to the calculated residual for improved measurement accuracy.

Differential Assessment

When the AI identifies differential relay channels (matching HV/LV winding currents or pre-computed DIFF/BIAS channels), it performs an operate/restraint analysis at multiple time points through the fault, plotting operating points against the percentage slope characteristic.

Understanding the AI's limitations helps you interpret results correctly:

• Relay settings optional — Relay protection settings can now be provided via XRIO file import through the Device Management workflow. When settings are available, the AI uses actual pickup thresholds and time delays for more specific assessments. Without relay settings, the AI relies on heuristic thresholds and the verdict may be INCONCLUSIVE for marginal cases.

• Voltage-dependent analysis — Recordings with only current channels (no voltage) cannot support directional, impedance, or voltage sag analysis. The AI automatically skips these tools.

• Channel naming — The AI identifies channels by their names in the COMTRADE file. Non-standard or ambiguous names may lead to incorrect channel pairing for three-phase calculations.

• Model-based reasoning — The AI applies protection engineering principles but does not simulate relay behavior. Its assessment is based on measured quantities and heuristic thresholds, not a relay model.

• CT saturation — Heavily saturated CT waveforms can distort all downstream calculations (phasors, RMS, impedance). The AI attempts to detect and flag CT saturation but cannot fully compensate for it.

Full transparency is provided through the analysis report. Every report includes a Methodology section at the bottom that documents the measurements taken during analysis. This allows you to verify the AI's findings independently using the Compute tab.