Asset failure root cause analysis that finds the sister units before the next failure
Faster RCAs and a searchable failure library, and every sister unit found the same day.
See one case, screen by screen ↓When a transformer, a breaker, a cable joint or a pump fails, the utility owes an investigation: what happened, why, and what stops it happening again. Most of the work is gathering. Fault recorder and relay files, the SCADA sequence of events, power factor and oil results, thermography, work orders, the field team's photos and supplier bulletins each sit in their own system, and the reliability engineer pulls them by hand for two to four weeks before the real thinking starts.
Then the report is written to close the case, not to teach the next one. Root causes go in as free text with no failure code and no link to the supplier batch, so the failure library cannot be searched by failure mode. The fix is applied to the unit that failed, and the other units carrying the same model and the same seal lot are not looked at until one of them fails too. Early warnings slip past: alarm rules look at an absolute limit rather than the rise from baseline, retests are deferred when outage windows are cancelled, and bulletins land in a procurement mailbox that never reaches the asset register.
Estimated split for one major asset failure, in working days, by hand and with the solution.
Seven agents gather the evidence, build the fault tree and find the sister units; the reliability engineer confirms the root cause.
Lumina Utilities runs the Kestrel Bay 132/33 kV substation. On Sep 28 2026 at 03:47, the B-phase 132 kV bushing on transformer T2 flashes over. Here is what happens next, screen by screen, in the working solution.
B-phase HV current rises to 11.2 kA, the 87T differential operates in 38 ms and both 132 kV breakers, CB-T2-H1 and CB-T2-H2, open at 03:47:12.476. Deluge valve DV-2 opens at 03:49:05 and the turret fire is out by 03:52. The control room starts a 33 kV backfeed from Orrin Vale; 14,900 customers are back at 05:12 and the last 3,520 at 06:38. The timeline builder merges the fault recorder, SCADA and restoration log into one sequence, each entry with its source.
“Last 3,520 customers restored (2 h 51 min)” · 1.87 million customer minutes lost
The control room opens RCA-2026-0147 at 03:58 and the evidence gatherer starts on the fault recorder and relay files at 04:02. By 07:40 it has pulled 46 records: the SCADA log, seven years of tests, oil results, three work orders and the 11 photos Lina Moreau took on site. Every step of the run is listed with its time, so the engineer can see what was read and when.
“Pulled 46 records” · “SCADA, 7 years of tests, oil results, 3 work orders, 11 photos”
No cloud-to-ground strikes within 10 km between 02:00 and 05:00, and the arrester counters read 14, 9 and 11, the same as in July. T2 was at 55.8 MVA, 62 % of its 90 MVA rating, with top oil at 54 °C. Post-event SFRA matches the 2018 baseline, winding resistance is within 0.6 % and main-tank acetylene is 0.4 ppm — no internal arc.
Limit: “A branch is ruled out only with a cited record.”
Photo 9 shows the flashover track on the lower porcelain, so bushing insulation breakdown is confirmed. The bushing's power factor rose from 0.36 % at commissioning to 0.61 % in March 2025, up 69 % from baseline, with C1 capacitance +4.1 %; photo 7 shows water staining on the top layers of condenser paper, and A-phase bushing oil reads 26 ppm moisture. Photo 3 shows the top-cap gasket cracked and hardened, and in July the B-phase top cap ran 6.8 °C warmer than A and C.
Ruled out: moisture left in at the factory — “0.36% in March 2018 — normal for a new OIP bushing.”
The library matcher finds Orsella service bulletin SB-2024-07 of Nov 14 2024: the top-cap gasket compound from seal lot SL-1709 can harden early and admit moisture. It reached a procurement mailbox and was never linked to the asset register. It also finds RCA-2021-0042: Redcliff T1, the same OTB-145 bushing and the same seal lot, closed in 2021 as an “isolated defect” with no fleet check. The fault tree builder proposes “Seal compound, 2017 batch” at 92 %, and marks as contributing an alarm rule that looks only at the 1.0 % limit and a retest deferred twice.
Evidence against: “No other cause of seal damage in the work history: no top-cap work since 2018.”
Kwame Osei signed the protection review on Oct 3: the 87T, Buchholz and breaker failure schemes operated correctly. Tomás reads the tree, the evidence for and against and the contributing causes, then confirms the root cause in his own name. The meaning of his signature, root cause confirmed, goes on the audit trail.
“I confirm the root cause of this failure is: seal compound, 2017 batch.”
The fleet checker matches model OTB-145 and seal lot SL-1709 in the asset register: 33 bushings on 14 transformers at 12 substations, serving 178,200 customers. It ranks them by last test result, time since test, customers served and site exposure — 3 high, 6 medium and 5 lower — with the reason for each. Marlow Cross T1 leads at 0.58 % power factor, up 66 % from baseline, with 21,300 customers.
“Not matched: 49 OTB-145 bushings from other seal lots on 17 transformers.”
The campaign planner drafts PC-2026-009: a power factor and C1 test, top-cap seal inspection and thermography on all 14 units, the three high-risk ones on Oct 16, 20 and 23, each window with its switching plan. It totals an estimated $186,000 and lists the spares: 4 OTB-145 bushings at Calder Rise depot, 6 on order at 26 weeks, 40 seal kits with a lot-free compound. Graham approves and work orders WO-2026-14201 to 14214 are raised.
Limit: “Drafts only; work orders are raised after approval.”
The report writer drafts eight sections in Lumina's RCA template v3 — summary, event, fault tree, root cause, contributing causes, actions, fleet and sign-off — with 19 sources cited. It links the case to RCA-2021-0042 and adds failure code BSH-MOIST-SEAL, so the library can find it by failure mode. The draft is ready to send to Helen Achterberg for review.
“Drafted 8 sections from the fault tree. 19 sources cited; no sentence without one.”
Lina Moreau owns replacing the T2 B-phase bushing and resealing A and C, due Oct 31. Helen owns the new alarm on a 30 % rise in bushing power factor from baseline, due Oct 24, and the test and reseal of the 14 sister units, due Jan 11. Ruth Okonjo links supplier bulletins to serials and lots on arrival, and Graham Ives reviews deferred condition retests every month.
On Kestrel Bay T2 the new alarm “would have alarmed in March 2025, 18 months before the failure.”
The same bushing failure, seen by the five people who carry it — what their weeks looked like, and what they look like now.
Seven agents gather the evidence, build the fault tree and find the sister units; the reliability engineer confirms the root cause.
Pulls fault-recorder and relay files, the SCADA log, test and oil history, work orders, field photos and supplier bulletins for the failed asset and its time window, quoting each fact with its source.
Builds the event to the millisecond from relay, breaker and SCADA records, and the restoration timeline from the control room log.
Builds the fault tree from the failure down, attaches evidence for and against each branch, marks it confirmed, ruled out or contributing, and proposes the root cause with a confidence. It goes down until the cause is something the utility can act on — a batch, a rule, a practice.
Searches 1,284 past investigations and every supplier bulletin for the same failure mode, model and batch, and says why each one matched.
When a root cause is confirmed, finds every sister unit with the same model, batch or seal lot and ranks them by last test result, time since test, customers served and site exposure, with a one-line reason for each.
Drafts the preventive campaign: units, test windows inside the risk deadline and the outage plan, switching notes, crews, spares and cost, for the Head of Asset Management.
Drafts the RCA report and the corrective and preventive actions in the utility's template, cites every sentence and adds a failure code so the library can find the case.
Confirms the root cause. The agents propose; a named person decides.
Ask about any failure, branch or sister unit — or say what to change. Every answer points to its records.
Why did Kestrel Bay T2 fail?
Moisture got into the B-phase 132 kV bushing through a cracked top-cap seal. The gasket is from Orsella seal lot SL-1709, which the supplier warned hardens early. The bushing's power factor rose from 0.36 % to 0.61 % before it failed, and photo 3 shows the cracked gasket. Lightning, overload and a winding fault are ruled out.
Which sister units should be tested first?
Three, within 14 days: Marlow Cross T1 — power factor 0.58 %, up 66 % from baseline, 21,300 customers; Ashby Fen T3 — 0.55 %, C1 +3.2 %, top cap 4.1 °C warm; Tern Point T2 — no test since 2021, coastal, no backfeed. All three carry seal lot SL-1709.
Alarm on the rise in bushing power factor, not just the 1.0% limit
Done. New condition alarm: bushing power factor 30 % above its own baseline, or C1 change above 5 %. On Kestrel Bay T2 it would have alarmed in March 2025, 18 months before the failure. Today's rule only looks at 1.0 %.
Have we seen this failure before?
Yes. RCA-2021-0042, Redcliff T1: the same OTB-145 bushing, the same seal lot SL-1709, power factor 0.72 %. It was closed as an isolated defect with no fleet check. The library has 4 bushing moisture cases since 2019.
13 screens from the working solution, on its sample data. Pick one to see it large.
The investigation clock, what makes two assets sister units, who signs and how the agents work are settings, chosen by the utility.
| Setting | Default | Choose from |
|---|---|---|
| Investigation clock — major failures | 30 days | 30 days · 21 days · 45 days — major means more than 5,000 customers off, or an asset worth more than $1M |
| Investigation clock — other failures | 60 days | 60 days · 45 days · 90 days, counted from the failure date |
| Protection misoperations | 60 days | 60 days · 45 days — corrective action plan within that many days of the cause being found |
| Sister units must match | Model + batch or seal lot | Model + batch or seal lot · Model only · Model + manufacturer year |
| Rank sister units by | Condition first, then customers | Condition first, then customers · Customers first — from last test result, time since test, customers served and site exposure |
| High risk — test within | 14 days | 14 days · 7 days · 30 days, from the day the root cause is confirmed |
| Preventive campaigns approved by | Graham Ives | Graham Ives · Helen Achterberg — approves units, windows and cost |
| Look back in the failure library | 10 years | 10 years · 5 years · Since 2012 — past investigations and supplier bulletins |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate of where an investigation's days go · “typical” = published figures (US DOE, Large Power Transformers and the U.S. Electric Grid, 2014 — $2–7.5M per large power transformer; CIGRE TB 642 transformer reliability survey — bushings about 17 % of substation-transformer failures, not re-checked). Regulatory context: NERC PRC-004-6. Lumina Utilities, its people, substations, suppliers and records named on this page are fictional — characters and sample data in the working solution.
Software that takes a failed asset — a transformer, breaker, cable joint, pump or water main — from the trip to a confirmed root cause, a cited RCA report and corrective and preventive actions. Asset Failure Root Cause Analysis adds seven agents that gather the evidence, build the fault tree, draft the report and find the sister units with the same weakness. The reliability engineer confirms the root cause.
The fault tree builder starts from the failure and lists every credible cause — external, thermal and internal — then attaches evidence for and against each branch and marks it confirmed, ruled out or contributing. It keeps going down until the cause is something the utility can act on, such as a batch, a rule or a practice, and proposes the root cause with a confidence.
When the engineer confirms the root cause, the fleet checker matches it against the asset register — in the sample case, model OTB-145 and seal lot SL-1709 — and ranks every match by last test result, time since test, customers served and site exposure. In the sample case it found 14 transformers carrying 33 bushings at 12 substations, and left out 49 bushings from other seal lots.
Yes. The investigations in the working solution include a trunk main burst, a pumping station bearing seizure and a chlorine dosing pump trip alongside transformers, breakers and cable joints, and the failure library holds water and network cases side by side.
Every trip gets a protection review, signed by the protection engineer, on whether protection operated correctly. Protection misoperations run on their own clock — a corrective action plan within 60 days of the cause being found by default — the timeframe NERC PRC-004-6 sets for the corrective action plan.
Yes. The agents are read-only on SCADA, the relay archive and work management, and they propose and draft but never confirm. The named engineer confirms the root cause with a signature, the asset manager closes the investigation, and campaign work orders are raised only after the Head of Asset Management approves.
Yes. The investigation clocks, the condition alarms, what makes two assets sister units, how they are ranked, how soon high-risk units are tested, the approvers for root cause, protection review, close and campaigns, and how the agents start and look back are all settings.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your RCA procedure and template, condition rules and a sample of past investigations — and it goes live once every quality gate has passed. We will walk you through it on your own failures first.
Defects caught while they grow: every structure compared with last year's flight.
1-min demoAsset Health IndexThe same capital budget removes 2.5× the risk, every score traced to its evidence.
1-min demoMaintenance Knowledge SearchDecades of fixes, searchable before the expert retires.
1-min demoWe'll run Asset Failure Root Cause Analysis on a sample of your own closed investigations.