For solar teams
Thermal pictures in.
A checked inspection report out.
One faulty solar panel can hide among a thousand good ones. PanelProof looks at every thermal picture, puts the suspicious panels in front of your technician first, and gives you a signed report where every finding opens its picture.
Checked panel by panel1,485 panels on one real roof
Made for the teams that keep solar plants working
The workspace
This is the app itself, open on a real inspection.
Eight drone photos of a rooftop plant, 1,485 panels. Choose a panel, open the review list, read the report. Everything you see came out of the app.
Opening the inspection
Why PanelProof
A faulty panel wastes power every sunny hour. From the ground it looks like all the rest.
A thermal camera sees it at once: a hot cell, a warm strip, a whole panel running warm. But one flight over one plant brings back hundreds of photos and thousands of panels. Checking them by eye takes days, and the report at the end is often a spreadsheet nobody can trace back to a picture.
Tested on3,002 pictures it had never seen
See the resultsWhat you get
- On the photoRow and column of every panel
- In the reportA small picture beside each finding
- For your clientOne private link, read only
A hot part, or a suspected faultEvenly warm
- Most suspicious firstThe review list is sorted for you
- Keyboard shortcutsN, F and C decide a panel
- Kept on recordA new decision never erases the old one
What the app said
Photo 3, row 4, column 6
Suspected fault, type not clear
It stands 14.3 spreads above the ordinary panels of this inspection.
Your decision
No fault NSuspected fault FCannot tell C
- Panel picturesPattern faults only
- Full photosEach panel compared with its neighbours
- TemperaturesIt prints none, and says so
The two bright rows are evenly warm panels. The app found them by comparing neighbours, not by looking at each panel alone.
- PDF and CSVOne row for every panel
- VersionsThe signed one is always kept
- Check itAnyone with the link can test the fingerprint
Thermal inspection report
Rooftop plant, Catanzaro
Findings are suspected faults seen in thermal pictures. This report is not a safety certificate and is not a statement that the inspection meets any standard.
| Pictures | Full thermal photos, cut into panels with the grid |
| Panels checked | 1,485 |
| Flagged for a person | 638 |
| Not flagged by the app | 847 |
Model pp-crops-0.1.0-level-camera. Review rule review-chooser-0.1.0.
Core features
Six parts, each with one job.
Small models do the looking. Plain rules check them. A person decides.
92 of 100 faulty panels found
On 3,002 test pictures. 39 of every 1,000 normal panels were flagged by mistake.
Before anything is judged, the app turns back what it cannot use: a file that is not a picture, or a picture in a colour scheme with no camera data behind it. It also straightens a panel that was saved with black corners. It did that for all 3,675 pictures of one public set.
A small model scores every panel. On 3,002 pictures it had never seen, it found 92 of every 100 faulty panels and flagged 39 of every 1,000 normal ones by mistake.
Every number, with its rangeThe app names a fault type only when that type passed its test. One of five did. "Bypass diode active" was named rightly 99.6 times in 100 and caught 93 of every 100 of its kind. Every other suspected fault reads "type not clear", and your technician chooses the type.
On full photos each panel is compared with its place among its neighbours. On one real roof it marked 162 evenly warm panels, 150 of them in 10 groups. The fault finder alone had called 15 of those 162 a suspected fault. No technician has marked those photos, so this is what the app said, not a measured score.
The roof, photo by photoThe app sets its levels from each batch, so a camera that reads bright does not flood the list. It sorts the batch into three piles and puts the most suspicious panels first. A batch needs at least 200 panels for levels. A smaller one is listed whole.
Thermal inspection report
Rooftop plant, Catanzaro
Findings are suspected faults seen in thermal pictures. This report is not a safety certificate and is not a statement that the inspection meets any standard.
| Pictures | Full thermal photos, cut into panels with the grid |
| Panels checked | 1,485 |
| Flagged for a person | 638 |
Model pp-crops-0.1.0-level-camera. Review rule review-chooser-0.1.0.
The lead signs with their own password. The PDF and the CSV are never changed after that, and the app keeps a fingerprint of each so anyone with the client link can check them.
The numbers
0of 100
Faulty panels found on 3,002 test pictures. The likely range is 90.8 to 93.5.
0in 1,000
Normal panels flagged by mistake. The likely range is 31 to 51.
0of 100
Faults that show a pattern inside the panel, found on pictures from another camera and site.
0of 100
Evenly warm panels found there. A single panel picture cannot show them.
How PanelProof AI works
Small models do the looking. Rules check them. A person decides.
Here is one real inspection, step by step, with the numbers it produced.
The first rule
The app suggests. Your technician confirms, changes or rejects every finding.
What the app said and what the person decided are stored apart, and both stay on record.
Product rule oneIn every PanelProof report
One real run
Step 1
Read the file
8 photos, each 640 by 512 pixels. The camera's own reading for every pixel was read out and drawn as a white hot picture.
Step 2
Cut the panels
16 grids placed on the photos. 1,485 panels cut out, turned upright and resized.
Step 3
Score every panel
One score for each. 289 panels stood far above the rest of the batch.
Step 4
Compare the neighbours
162 evenly warm panels, 150 of them in 10 groups. 51 panels with a hot part.
Step 5
Sort for review
289 suspected faults, 323 more to review, 873 not flagged. A person decides from here.
The tests
Four tests, and what each one found.
Each test was fixed in writing before its pictures were opened.
3,002 pictures were held back before any model was trained. The model found 92.2 of every 100 faulty ones and flagged 39.4 of every 1,000 normal ones. It found 95 of every 100 faults that show a pattern, and 62 of every 100 evenly warm panels.
3,675 pictures of one array, made by other people with other equipment, with no training on them. Pattern faults: 98.3 of every 100 found. Evenly warm panels: 7.7 of every 100. Taken together that is 51.8 of every 100, which is below the 70 we had set as our own stop rule.
1,485 panels on a rooftop plant. The neighbour check marked 162 evenly warm panels and 51 panels with a hot part. Nobody has marked these photos by hand, so we cannot say how many of those marks a technician would agree with.
The rule that sorts a batch was aimed at 50 false alarms in every 1,000 normal panels. It gave 86 on the test pictures and 75 on the other camera, where a fixed level gave 781. It is steadier than a fixed level, and it is not exact.
Case studies
Real pictures. Real misses shown.
Every case study is a public set of pictures run through the app as it is. None of these plants is a PanelProof customer, and none took part.
Featured case study
Catanzaro
Eight drone photos of a 1.8 MW rooftop plant
More case studies
How it works
Four steps from the flight to the report.
Nothing to install. The whole job happens in the browser.
Upload the pictures.
Panel pictures with one panel in each, or the full photos a drone or a handheld thermal camera takes. Many files at once, or one zip.
Place the grid.
On a full photo, click the four corners of a block of panels and type its rows and columns. The app cuts out every panel and remembers its row and column.
What the app said
Photo 3, row 4, column 6
Suspected fault, type not clear
It stands 14.3 spreads above the ordinary panels of this inspection.
Your decision
No fault NSuspected fault FCannot tell C
Review the flagged panels.
The most suspicious panels come first. Confirm, change or reject each one with a single key, and add a note if you want one.
Thermal inspection report
Rooftop plant, Catanzaro
Findings are suspected faults seen in thermal pictures. This report is not a safety certificate and is not a statement that the inspection meets any standard.
| Pictures | Full thermal photos, cut into panels with the grid |
| Panels checked | 1,485 |
| Flagged for a person | 638 |
Model pp-crops-0.1.0-level-camera. Review rule review-chooser-0.1.0.
Sign and share.
The lead signs. The PDF and the CSV are locked, and your client gets a private link to the report and its pictures.
Prices
Pay for the panels you check. Nothing for the ones you do not.
Priced in naira, with no sales call
Try it
See your own panels go through the app.
What is in it
- One inspection of up to 300 panels
- The review list
- A signed report
- A client link
Free
You need no card.
Get startedPay as you go
For a plant now and then.
What is in it
- Any number of panels
- Review, signed report and client link
- Panel pictures and full photos
₦30 for each panel
From ₦15,000 an inspection.
Write to usDrone service firms are priced on request. Write to us.
Who uses PanelProof
Made for the people who answer for a plant.
If your name goes on the report, this is for you.
A hand over report for your client, and proof for a warranty claim.
When: the day you hand a new system over, while the installer and the client are both on site.
Solar companies
Find faults early across many sites, and keep a record for each.
When: at every routine inspection.
Mini grid operators
Keep small plants working with few staff.
When: when output drops and nobody knows why.
Drone service firms
Turn the photos you already take into a report your client can read.
When: after every flight.
Owners of large systems
An independent check on your contractor, for a factory or a campus.
When: before you sign off the work.
What it does not do
Questions
Plain answers, before you upload anything.
Questions answered
- Which pictures can I upload?
- JPEG, PNG or TIFF. Panel pictures with one panel in each, or full thermal photos with many panels. Many files at once, or a zip.
- Must the camera be set to white hot?
- If the file holds the camera's own readings, no. The app draws its own white hot picture. If it is only a picture in a colour scheme, the app turns it back and asks for the original camera file.
- Which cameras has it been tested on?
- Full photos from a DJI Zenmuse H20T, and panel pictures from two public sets. A camera that saves grey thermal pictures works from brightness. DJI has no connection with PanelProof.
- How often is it wrong?
- On 3,002 test pictures it missed 8 of every 100 faulty panels and flagged 39 of every 1,000 normal ones. On another camera and site it found 98 of every 100 pattern faults and 8 of every 100 evenly warm panels.
- Why does it say “type not clear”?
- It names a fault type only when that type passed its test. One type did: bypass diode active. Your technician can choose any type.
- Why do small batches get no levels?
- The app sets its levels from the batch, and needs at least 200 panels to do that. Below that it lists every panel, most suspicious first.
- Can a signed report be changed?
- No. A correction makes a new version, and the signed one is kept.
- What does the free plan cover?
- One inspection of up to 300 panels, with the review list, the signed report and the client link.
- Who can see my pictures?
- Only the people in your workspace, and anyone you give a client link to. We never train on a customer's pictures without written agreement.
- How do I start a paid plan?
- Write to us from the contact page and say which plan you want.