AI Electrical Fault Finding — How It Works and When to Use It

Electrical fault finding is one of the most challenging aspects of electrical work. Unlike new installation — where you follow a design and build from scratch — fault finding requires you to work backwards from symptoms to causes in an installation you did not design and may never have seen before. The possible causes for any given symptom are numerous, and the most obvious explanation is not always correct.

Consider a common scenario: an RCD trips repeatedly. The possible causes include a failing appliance with leakage to earth, a deteriorated cable with damaged insulation, moisture ingress at a junction box or socket outlet, a neutral-earth fault somewhere on the protected circuits, a faulty RCD mechanism, cumulative leakage from multiple circuits just exceeding the trip threshold, or even an incorrectly wired circuit where neutral and earth have been transposed. Each cause requires a different diagnostic approach, and systematically working through them all takes time.

Experienced electricians develop an intuition for fault diagnosis. They recognise symptom patterns, they know which causes are most common for each type of installation, and they have a mental library of faults they have encountered before. This experience is invaluable, but it has limits — even the most experienced electrician has not seen every possible fault, and unusual or intermittent faults can be genuinely difficult to diagnose.

AI fault finding does not replace this experience. What it does is extend it — by drawing on a fault database that is larger than any individual electrician's experience, and by applying systematic pattern matching to identify the most likely causes for a given set of symptoms.

Elec-Mate's AI fault diagnosis tool uses a combination of natural language understanding and a structured fault knowledge base. The process works as follows:

• Symptom capture — you describe the fault in plain English. For example: "the RCD protecting the downstairs ring main trips after about 10 minutes. It resets fine and does not trip immediately. Started happening three weeks ago. No new appliances connected."
• Symptom analysis — the AI extracts the key diagnostic information from your description: the affected circuit (downstairs ring main), the protective device behaviour (delayed trip, not immediate), the time to trip (approximately 10 minutes), the recurrence pattern (repeatable after reset), the onset (three weeks ago), and any changes (none reported).
• Pattern matching — the extracted symptoms are compared against the fault database. The delayed trip pattern (10 minutes) strongly suggests a thermal or moisture-related cause rather than a fixed fault, because fixed faults typically trip the RCD immediately or within seconds.
• Probability ranking — the AI ranks the probable causes by likelihood based on the symptom pattern, the circuit type, and historical fault data. For this example, the most likely causes would be ranked as: (1) an appliance with a developing fault that worsens as it heats up, (2) moisture ingress at a junction or connection point, (3) cable insulation deterioration that becomes conductive when warm.
• Test sequence recommendation — for each probable cause, the AI recommends the specific test to confirm or eliminate it. For the example above: disconnect all appliances and test if the RCD still trips (eliminates cause 1); carry out insulation resistance testing on each section of the ring with the circuit warm (tests causes 2 and 3); check all accessible junction boxes for signs of moisture.

AI Fault Diagnosis — Built Into Elec-Mate

Describe fault symptoms in plain English and the AI suggests probable causes ranked by likelihood, with diagnostic test sequences for each.

Try it free for 7 days

Pattern matching is the core of AI fault diagnosis. The AI maintains a knowledge base of fault patterns — combinations of symptoms that are characteristic of specific root causes. These patterns are derived from real-world fault reports, manufacturer fault data, and the published technical literature on electrical fault diagnosis.

Some patterns are simple and well-known. An insulation resistance reading of less than 1 megohm on a circuit almost certainly indicates cable insulation breakdown, moisture ingress, or a faulty connected device. An earth fault loop impedance reading that exceeds the maximum permitted value for the installed protective device indicates either a high-resistance earth path or incorrect/degraded connections.

Other patterns are more subtle. The AI can identify correlations that a human diagnostician might miss because they require cross-referencing multiple pieces of information simultaneously. For example:

• Time-dependent tripping — if an RCD or MCB trips only after a period of operation, the AI identifies this as a thermal pattern and prioritises causes that worsen with temperature: failing appliance heating elements, cable insulation that becomes conductive when warm, or loose connections that expand with heat.
• Weather-dependent faults — if the fault occurs more frequently during rain or high humidity, the AI identifies moisture ingress as a primary suspect and suggests checking outdoor junction boxes, cable entries, and IP ratings of outdoor equipment.
• Load-dependent symptoms — if the fault occurs only when specific loads are connected, the AI identifies the load as either the cause itself (appliance fault) or a trigger (marginal insulation that fails only under load current heating).
• Multiple circuit involvement — if symptoms appear on multiple circuits simultaneously, the AI identifies common-point failures: supply-side issues, main earthing problems, or shared neutral/earth faults.

The pattern matching also considers the installation context. A fault in a 1960s installation with rubber-insulated wiring has a different probability distribution than the same symptom in a 2020 installation with thermoplastic cables. The AI adjusts its probability rankings accordingly.

The quality of the AI's fault diagnosis depends heavily on the quality of the symptom description. The more detail you provide, the more accurately the AI can narrow down the probable causes. The AI extracts and analyses the following categories of information from your description:

• Affected circuit(s) — which circuit or circuits are experiencing the fault. Single-circuit faults have different causes to multi-circuit or whole-installation faults.
• Protective device behaviour — which device trips (RCD, MCB, or both), whether it trips immediately or after a delay, whether it can be reset, and whether it trips on a specific operation (switching on a load, turning on the supply).
• Temporal pattern — when the fault occurs (time of day, weather conditions, seasonal pattern), how often it occurs, and when it first started.
• Recent changes — any changes to the installation, connected appliances, building structure, or environmental conditions that preceded the fault onset.
• Test results — any test readings you have already taken, such as insulation resistance, continuity, or earth fault loop impedance values.
• Installation age and type — the approximate age of the installation, wiring system type, and earthing arrangement, which influence the probability of different fault types.

If your initial description is missing key information, the AI asks clarifying questions before generating its diagnosis. For example, if you report an RCD tripping but do not mention whether it trips immediately or after a delay, the AI will ask — because the timing is a critical diagnostic indicator that significantly changes the probable cause ranking.

For each probable cause identified, the AI recommends a specific diagnostic test to confirm or eliminate it. The tests are ordered to follow the most efficient diagnostic path — starting with the simplest, quickest tests that can eliminate the most probable causes, then progressing to more detailed investigation if needed.

For an RCD tripping fault on a ring final circuit, the AI might recommend the following diagnostic sequence:

• Step 1: Appliance isolation — disconnect all appliances from the affected circuit and retest. If the RCD no longer trips, reconnect appliances one at a time to identify the faulty unit. This is the quickest test and eliminates the most common cause (faulty appliance).
• Step 2: Insulation resistance test — safe isolation and proving dead are mandatory before connecting the tester (BS 7671 Reg 643.3.1 / GN3 Reg 2.8). With all appliances disconnected and the circuit isolated, carry out an insulation resistance test at 500 V DC between live conductors and earth. The minimum acceptable result for a standard LV circuit is 1.0 MΩ per Table 64 (Reg 643.3.2). Note: SELV and PELV circuits must only be tested at 250 V DC (minimum 0.5 MΩ per Table 64) — applying 500 V to SELV/PELV circuits risks equipment damage. Test each leg of the ring separately to localise the fault.
• Step 3: Split the ring — if insulation resistance is marginal, disconnect the ring at the consumer unit and test each leg independently. A significantly lower reading on one leg localises the fault to that half of the ring.
• Step 4: Junction box inspection — visually inspect all accessible junction boxes, socket outlets, and connection points on the affected section for signs of moisture, heat damage, or loose connections.
• Step 5: Sectional isolation — if the fault is still not found, progressively disconnect sections of the circuit to narrow down the fault location using a process of elimination.

Each suggested test includes the expected result if the probable cause is confirmed, what to do next if the result is inconclusive, and the relevant testing procedure from the testing sequence guide. This systematic approach reduces the time spent on fault finding by eliminating guesswork and ensuring you test in the most efficient order.

The AI fault diagnosis system improves over time through two mechanisms: the collective fault database and individual feedback.

The collective fault database aggregates anonymised fault data from all users. When an electrician reports a fault, diagnoses it, and confirms the root cause, that data point is added to the database. Over time, this builds an increasingly accurate picture of which symptoms typically lead to which root causes in different types of installation. The more faults that are reported and resolved, the better the AI becomes at ranking probable causes for new faults.

Individual feedback refines the AI's recommendations for your specific work. If you consistently find that a particular fault type is more common in your area (for example, if you work in a coastal area where moisture ingress is more prevalent due to the salt air), the AI adjusts its probability rankings for your future diagnoses.

This learning mechanism is particularly valuable for emerging fault patterns. For example, as more homes install EV chargers with Type A or Type B RCDs, new fault patterns emerge that are not yet documented in textbooks. The AI captures these patterns from early adopters and makes them available to all users, accelerating the spread of diagnostic knowledge across the trade.

All fault data is anonymised before being added to the collective database. No client details, property addresses, or personally identifiable information is shared. Elec-Mate complies fully with UK GDPR for all data processing. See our guide on AI tools for electricians for more on privacy and data handling.

Here are some of the most common fault scenarios that electricians encounter, and how the AI approaches each one:

• RCD tripping — immediate — suggests a direct earth fault on one of the protected circuits. The AI recommends isolating circuits one at a time to identify the affected circuit, then isolating appliances on that circuit. If no appliance is at fault, insulation resistance testing is recommended. See our guide on RCD keeps tripping.
• RCD tripping — delayed — suggests a thermal or load-dependent cause. The AI prioritises appliance faults (heating elements with developing insulation breakdown), moisture ingress that worsens with ambient temperature, and cumulative leakage from multiple circuits.
• MCB tripping — the AI distinguishes between overcurrent tripping (circuit overloaded or short circuit) and magnetic tripping (high inrush current). For overcurrent, it suggests load analysis. For magnetic tripping, it suggests checking for appliances with high starting currents.
• Partial loss of supply — the AI considers lost neutral (extremely dangerous on TN-C-S systems), single-phase loss on a three-phase supply, supply fuse failure, and internal distribution faults.
• Intermittent faults — the hardest faults to diagnose. The AI asks for detailed timing patterns, weather correlations, load correlations, and temperature correlations to narrow down the cause. It may suggest extended monitoring or thermal imaging to capture the fault during occurrence.
• High earth fault loop impedance — the AI considers loose or degraded earth connections, undersized CPCs, long cable runs, and supply-side earth path issues. It recommends systematic testing from the origin outward to localise the high-resistance point. When reviewing measured Zs values, always compare against a temperature-corrected limit rather than the raw Table 41.4 maximum — conductor impedance at test temperature is lower than at maximum operating temperature, so a raw comparison can give a false pass or false fail. GN3 Reg 3.18 and Appendix A3 set out the correction method; GN3 Appendix A provides adjusted tabulated Zs values for standard thermoplastic (PVC) circuits.
• A4:2026 — lighting circuit RCD trips — BS 7671:2018+A4:2026 Reg 411.3.4 now requires additional protection by a ≤30 mA RCD on AC final circuits supplying luminaires in domestic premises. An RCD trip on a domestic lighting circuit may therefore be correctly-operating A4 protection responding to a real leakage event, rather than a nuisance trip or wiring fault. The AI differentiates this from pre-A4 installations where lighting circuits were not required to be RCD-protected.
• A4:2026 — AFDD trips — Reg 421.1.7 (A4:2026) recommends arc fault detection devices (AFDDs) on AC final circuits to mitigate fire risk from arc fault currents. An AFDD trip is a distinct symptom from an MCB overcurrent trip or an RCD residual-current trip — the AFDD detects the characteristic signature of a series or parallel arc fault. If a consumer unit has AFDD devices fitted, the AI treats an AFDD operation as a separate fault category requiring arc-fault investigation (loose connections, damaged cable insulation, deteriorated wiring) rather than a standard overload or earth fault diagnostic path.

For each scenario, the AI provides both the technical diagnosis and a plain-English explanation suitable for communicating with the client. This dual output saves time and helps you explain the fault and required repair to homeowners who do not understand electrical terminology. The Client Explainer feature produces even more detailed client-facing explanations when needed.

AI Fault Diagnosis + Client Explainer

Diagnose the fault with AI assistance, then generate a clear, plain-English explanation for the client.

Try it free for 7 days

AI fault finding is a powerful diagnostic aid, but it has clear limitations that every electrician should understand:

• No physical inspection — the AI cannot see the installation, smell burning, feel a warm connection, or hear an arcing fault. These sensory observations are often critical diagnostic clues that only a person on site can detect.
• No testing capability — the AI suggests tests but cannot carry them out. You still need your Megger or Metrel, your proved voltage indicator, and your competence in safe isolation and testing procedures.
• Depends on symptom quality — the AI diagnosis is only as good as the symptom description. If you describe the fault inaccurately or miss a key detail, the AI's ranking of probable causes may be misleading.
• Unusual or unique faults — the AI works best for fault patterns that have been seen before. Truly novel faults — caused by unusual equipment combinations, manufacturing defects, or unprecedented environmental conditions — may not match any pattern in the database.
• Does not replace experience — the best fault diagnosticians combine AI assistance with their own experience, intuition, and on-site observations. AI is a tool in the diagnostic toolkit, not the entire toolkit.

Think of AI fault finding as a colleague with a very good memory. They can recall every fault pattern from a database of thousands of cases and suggest what to check first. But they are not on site, they cannot hold a test lead, and they rely on you to describe what you see. The combination of their knowledge and your hands-on skills is more effective than either alone.