0404
Inspection, prediction and compliance on one recordOperations & maintenance
Plan campaigns, rank anomalies by real severity, watch telemetry for drift and dispatch the work, all against the model built during the build.
The asset produces more evidence every year than the team can act on
A plant in operation generates inspection reports, telemetry history, work orders and photographs continuously, and each lands somewhere different. The result is not a shortage of data. It is that deciding what to do this week requires one person to hold four systems in their head, so what gets done is what somebody remembered, and the record of why is a conversation.
A thermal anomaly, corroborated and dispatched inside a day
The pass is planned from what changed. The finding is checked against what the sensors saw. The work is raised only if the asset's own history says it should be.
Bypass diode failure, string 14
B13.S14.M07
- confidence
- 0.94
- delta T
- 21.4 K
- corroborated
- yes
- standard
- IEC TS 62446-3
Work raised inside the standing limit
Corroborated, in warranty, part on the shelf and under the two-hour labour threshold for this site, so it proceeded without waiting for Monday.
Identifiers, timings and values are illustrative.
What the layer does once the plant is running
Four capabilities, and the fourth is the one that makes the first three compound rather than accumulate.
Fly what changed, not the calendar
The next pass is planned from drift, from what was skipped for weather and from what has not been looked at, rather than from a fixed annual schedule.
Check the picture against the sensors
A visual finding is compared with what the equipment was doing at the time. Corroborated findings rank above ones only a single source saw.
Watch the trend, not just the event
Drift and degradation across the fleet, so the useful question becomes which asset is changing rather than which one has already failed.
Close the loop
Outcome, time taken and part consumed written back, so the next plan is built from what the last one actually found.
The part that repeats
Operations is not a sequence with an end. Four moves, and the fourth changes the first, which is the only reason the model gets better rather than merely bigger.
Plan from what changed
Drift since the last pass, blocks skipped for weather, assets nobody has looked at in a while. The route is built over those rather than over the whole site.
Capture and classify
RGB and thermal come back attached to the assets they cover, classified with a confidence rather than pinned on a map.
Corroborate and rank
Telemetry for the same window, compared against peer equipment. What two sources agree on outranks what one of them saw.
Act, or stop and ask
Inside your threshold the work is raised and dispatched. Outside it, the finding goes to a person with the evidence attached and the reason it stopped.
How often, and what for
A starting point rather than a rule. The planner moves from what the site is actually doing, and the last column is what the interval is anchored to.
- Utility-scale PVTypical cadenceThermal twice a year, RGB quarterlyWhat the pass looks forHotspots, diode and string failures, soiling, tracker alignment, vegetationReferenceIEC TS 62446-3
- Wind turbinesTypical cadenceBlades annually, tower and foundation on conditionWhat the pass looks forLeading-edge erosion, lightning damage, cracks, coating loss, oil tracesReferenceIEC 61400
- Substations and linesTypical cadenceThermal twice a year, corridor on conditionWhat the pass looks forConnector hotspots, corrosion, insulator damage, encroachmentReferenceIEC 62443
- Structures and civilsTypical cadenceAnnually, or after an eventWhat the pass looks forCracking, spalling, corrosion, settlement, drainage and access conditionReferenceISO 55001
These intervals are the industry starting points the standards describe, not a measured result of ours. Insurers, warranty terms and your own experience will move them, and the point of planning from condition is that the interval stops being the thing that decides.
All four modules, in this phase
This is the only phase where the whole layer is in play, because it is the only one where there is a history to reason over and something worth acting on every week.
Detect
Flies the planned pass and classifies what it finds against the asset it belongs to.
Read moreAnalyse
Streams the telemetry a visual finding is corroborated against, and watches the trend between passes.
Read moreOperations
Holds the history, parts and warranty state that decide whether a finding is urgent or routine.
Read moreDecide & act
Plans the pass, ranks what was found, and raises the work inside the limits your team set.
Read moreWhat leaves this phase
A ranked list of what to do this week
Findings ordered by corroborated severity against the asset's actual situation, not by detection confidence alone.
Work in your system, with evidence
Raised in O-OPS or written back into the maintenance system you already run, with the frame and the trace attached.
A versioned visual history
Every pass kept, so an asset can be compared with itself over years rather than with a written description of itself.
A decision record per action
What was read, which rule applied, what confidence it carried, and who approved it.
- 01O-EYEDetectInspection & digital twin
- 02O-ARCAnalyseSensors, IoT & telemetry
- 03O-OPSOperationsThe system of record
- 04actsO-AGENTDecide & actThe agentic layer
Every action writes back to the record, so the model sharpens with every job.
Questions from this phase
Can this run alongside our existing inspection contract?
Yes, and that is usually how it starts. The contractor keeps flying. The difference is that what they produce lands in a model that also holds the telemetry and the work history, instead of landing in a report.
How much autonomy do teams actually grant at the start?
Almost none, and we would be suspicious of a team that granted a lot on day one. The normal path is that the agent proposes everything for a few months, the proposals get reviewed, and limits are opened on the specific action types where the record shows it was consistently right.
What if a finding is wrong?
It is caught by the same threshold that lets the agent act. Below the confidence bar for its class a finding is proposed rather than actioned, and every action carries the evidence it was based on, so a wrong call is reviewable rather than mysterious.
Does this reduce headcount?
That is not the claim we make. What changes is what the week is spent on: less assembling evidence and chasing four systems, more of the work that needs judgement. If a headcount case is what you need, we would rather build it with your numbers than assert one here.
