AI drone inspection of power lines that ranks the defects that are growing first, against last year's flight
Defects caught while they grow: every structure compared with last year's flight.
See one case, screen by screen ↓Drones are taking over from helicopter and foot patrols, and the bottleneck has moved from capturing images to reviewing them. One flight of a 41.6 km, 132 kV double-circuit line brings back 6,412 images — about 154 per km. A line inspector opens them one by one, so a cracked disc can wait six weeks before anyone sees it, and the backlog grows with every flight the contractor uploads.
Nobody lines this year's picture up against last year's. The photo of the same insulator from the last flight sits in another folder, so a Priority 3 “watch on the next flight” only works if someone looks — and measures. When a defect is confirmed, it is retyped into the work management system without the image or its position, and the line's inspection report is assembled by hand. Meanwhile the inspection cycle has to be kept, and each finding fixed within the time its priority allows.
Estimated split for one flight of one line, in calendar days, by hand and with the solution. Lumina's March–June flights took 38–47 days from flight to reviewed findings by hand; in October, 1.8 days.
Six agents match, box, compare, rank and draft every flight overnight; the inspector confirms each box.
Lumina Utilities' Overhead Lines department flies every line once a year. On Monday 12 October 2026, Corvane Aerial flies F-2609 along L-214 Harrow Ridge – Calder, a 132 kV double circuit of 148 structures. Here is what happens next, screen by screen, in the working solution.
Jo Whitaker uploads F-2609 at 16:40: 6,412 images. By 17:31 every image is matched to its structure and two re-shoots are requested (T-019 phase C, T-131 footing). By 19:52 the screener has candidates on 425 images, merged into 118 boxes on 64 structures. By 20:48, 141 of 148 structures are compared with last October's flight, and at 21:05 the queue is ranked: 2 Urgent, 30 Priority 2, 86 Priority 3. The same five steps run on demand from the Flights screen — here for F-2610, the 33 kV Dunmore tee flown this morning.
Review lag: 38–47 days by hand in March–June; 1.8 days in October.
Across the top is the line itself: all 148 structures as ticks, coloured by their worst box, with the Calder Road crossing marked. On the left, structures ranked for review, growing and Urgent first. In the centre, T-087's phase B string at 12× zoom, boxed “Cracked insulator disc · 94%”. On the right, the box under review — Urgent, was Priority 3 — and how it was made, agent by agent, with the time of each step.
“Crack grew 109% in 12 months (growth rule G-1: over 50% → Urgent) over a road crossing.”
At T-042 on the Harrow Ridge escarpment, the corroded area of the phase C suspension clamp has risen from 6% to 19% of the fitting. Rule C-1 raises it one level, to Priority 2. Ines presses C. The box is confirmed and the gallery moves on to T-118 in Westmere woods: a new nest 0.9 m from a live jumper, 97% sure.
“Confirmed B-0422 · Priority 2, within 90 days”
At T-055 the screener boxed “corrosion on tower leg” at 58% — an orange patch on a bolt group, flagged as low confidence, maybe fresh primer. It is: a repaired bolt. Ines presses R and picks the reason, “Not a defect — fresh primer or paint”. The reason goes to the audit trail and back to the screener for retraining.
“Every box at 40% or more is seen by a person.”
Back on T-087, Last year turns the image into a wipe between October 2025 (F-2511) and October 2026 (F-2609). Ines drags it across the string. The change detector found the same component on last year's flight, aligned the two on fixed points and measured the crack from the 255 mm disc: 11 mm then, 23 mm now.
“Crack length · same component, aligned by the change detector”
The comparison shows the disc full size — wipe, side by side or zoom on the box — aligned on the four disc caps. The component's history: no crack seen in October 2023; 11 mm in October 2025, logged Priority 3, “watch on the next flight”; 23 mm now, +109% in 12 months. Growth rule G-1 says a crack that grows more than 50% in 12 months is Urgent.
“crack length · scaled from the 255 mm disc, ±1 mm”
Ines promotes the box to Urgent, due 20 October 2026, with a note: the crack on disc 4 doubled since F-2511 and the string is over the Calder Road crossing. It goes to Hana Lindqvist, who agrees from her phone. Only then is the work request drafted.
“Urgent — agreed. Ines Calloway confirmed, Hana Lindqvist agreed at 09:17. Work request WR-26-4417 drafted, due 20 Oct 2026.”
WR-26-4417: replace the left-circuit phase B insulator string at T-087 — 9 × 255 mm porcelain discs, ball-and-socket, from stores; overhead lines crew 3, four people; a circuit outage of 8 hours; traffic management for the Calder Road crossing. WR-26-4420 replaces the phase C clamp and U-bolts at T-042, due 11 January 2027. Each carries the image and its position.
Drafts only. Marcus Bell, maintenance planner, releases them to the work management system.
The report writer drafts the overhead line inspection report for F-2609: summary, coverage, findings by priority, the defects that grew and the review status. Every structure has its six required views; 888 of 888 were captured. A 2% spot check of the 5,987 images with no box is under way: 64 of 120 checked, no defect missed. Rafael Moreno signs and issues it, with the meaning of his signature recorded.
The report never signs itself, and it always states the boxes still open.
The same flight, seen by the five people who fly it, review it, agree it, plan the work and sign for it.
Six agents match, box, compare, rank and draft every flight overnight; the inspector confirms each box.
Matches every image to its structure by GPS and camera heading, checks sharpness and exposure, checks each structure has its six required views, and asks the contractor for re-shoots.
Checks each image against the 31 classes of the defect catalogue, boxes each candidate with its class and confidence, and merges the same defect seen from several angles into one box.
Finds last year's image of the same component, aligns the two on fixed points such as caps, bolts and members, measures length, area or count from a known dimension, and applies the growth rules.
Sets each box's priority from the defect catalogue, its growth and its location — road, rail and river crossings, high fire-risk areas — and ranks the queue by priority, growth and confidence.
Drafts the corrective work request for each confirmed defect: structure, component, work, materials from stores, crew type, outage or live-line, due date from the priority, and the image with its box.
Writes the line's inspection report in the Overhead Lines template from the reviewed boxes: summary, coverage, findings by priority, defects that grew, review status and sign-off.
Confirms or rejects every box. The agents propose; a named person decides.
Ask in plain words — or say what to change. Every answer points to the structure, the comparison or the rule behind it.
Which defects grew since last year's flight?
23 of the 118 boxes on F-2609 grew since F-2511 in October 2025; 61 are new and 34 unchanged. The three that moved most: T-087 cracked disc, 11 → 23 mm, now Urgent; T-101 broken strands beside the clamp, 1 → 2, Urgent; T-042 clamp corrosion, 6% → 19% of the fitting, raised to Priority 2.
Why is T-087 urgent?
Because the crack doubled in a year. On F-2511 the crack on disc 4 of the left-circuit phase B string was 11 mm and was logged Priority 3, “watch on the next flight”. On F-2609 it is 23 mm, scaled from the 255 mm disc. Growth rule G-1 makes any crack that grows more than 50% in 12 months Urgent, and the string is over the Calder Road crossing. Urgent needs your confirmation and Hana Lindqvist's agreement.
Draft work requests for the confirmed defects
Drafted from what you have confirmed on F-2609, each with the image, the structure's position and the materials from stores. Urgent boxes are drafted once Hana Lindqvist agrees; T-087 would be WR-26-4417, a full string replacement over Calder Road, due 20 Oct. Marcus Bell releases them.
Make any new nest within 1 m of a jumper Priority 2
Done. New rule N-2: a nest within 1 m of a live jumper → Priority 2, applied by the severity ranker from tonight's screening. On F-2609 it matches one box, T-118, which is already Priority 2. On last year's flights it would have raised 7 boxes. Rafael Moreno is told of the change.
13 screens from the working solution, on its sample data. Pick one to see it large.
The thresholds, the priorities and their due dates, the rules that raise a priority and the people who decide are settings, chosen by the Overhead Lines department and versioned when they change.
| Setting | Default | Choose from |
|---|---|---|
| Show a box to a person from | 40% sure | Below this the box is dropped and logged |
| Spot check of images with no box | 2% | Sampled for a person after every flight |
| Compare with the last flight | On | Same component, aligned, measured |
| Re-shoot when | Any of 6 views | A required view is missing or blurred |
| Urgent | 7 days | Inspector confirms, asset engineer agrees · Priority 2 within 90 days · Priority 3 next planned cycle, at most 12 months |
| Raise the priority when | G-1 · S-2 | Crack length up more than 50% in a year → Urgent · Two or more broken strands beside a clamp → Urgent |
| Over a road, rail or river crossing · in a high fire-risk area | One level up | Applied by the severity ranker; a person confirms |
| A new nest within 1 m of a live part | Priority 2 | Applied by the severity ranker; a person confirms |
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 flight data · “target” = the design goal, measured in the live solution · “estimated” = worked from published inputs: $1.6M a year = 1,355 miles × ($1,400 − $250) per mile, the midpoints of $1,200–1,600 per mile for helicopter and $200–300 per mile for drone surveys · “typical” = published figures (Utility Dive, 2019, a US utility's projected drone and helicopter cost per mile; EPRI Journal on AI for drone image analysis). Regulatory context: NERC FAC-003; CPUC General Order 95 Rule 18. Lumina Utilities, Corvane Aerial, the lines, flights, structures and people named on this page are fictional — characters and sample data in the working solution.
Software that reviews the images from drone inspections of power lines and turns them into defect findings people can act on. Drone Inspection Analytics matches every image to its structure, boxes candidate defects against your defect catalogue, measures what changed since last year's flight and ranks the boxes so the growing and Urgent ones come first. A line inspector confirms or rejects every box.
After a flight is uploaded, four agents run overnight. The flight intake checker matches images to structures by GPS and heading and asks for re-shoots; the defect screener boxes candidates with a confidence; the change detector compares each component with last year's image; the severity ranker sets priorities. Then the work request drafter and the report writer draft the work and the report from the boxes people have reviewed.
It screens for the classes in your defect catalogue — in the working solution, 31 classes such as cracked or chipped insulator discs, tracking on polymer insulators, broken conductor strands, corroded clamps, missing cotter pins, corrosion on tower legs, bird nests and vegetation inside clearance. Each class shows how often inspectors agreed with it over the last 12 months, and a new class is tested on last year's flights before it goes live.
The change detector finds the same component on the previous flight, aligns the two images on fixed points such as caps and bolts, and measures length, area or count from a known dimension, stating the scale and its error. Growth rules then apply: in the sample case a crack that grew from 11 mm to 23 mm, more than 50% in 12 months, was promoted from Priority 3 to Urgent.
Yes. Every box at or above the threshold is seen by a person, and 2% of images with no box are spot-checked. Agents never confirm, agree, release or sign: the inspector confirms each box, the asset engineer agrees every Urgent, the maintenance planner releases work and the programme manager signs the report.
It reads the asset register for each structure's type, age, insulators, conductor and access, and drafts corrective work requests in your work management format, with the image, materials, crew, outage and due date. Released requests are exported to the work management system, where crews and outages are booked.
Yes. The screening threshold, the spot-check rate, the re-shoot rule, the due date for each priority, the growth rules, the location rules for crossings and high fire-risk areas, and the people who decide are all settings, versioned when they change.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your defect catalogue, inspection standard, report template and a sample of past flights — and it goes live once every quality gate has passed. We will walk you through it on your own flights first.
The 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 demoTransformer DGA InterpretationEvery oil result trended the day it lands, with advice set by how fast the gas is rising.
1-min demoWe'll run Drone Inspection Analytics on two years of flights of one of your own lines.