The ticket says dust. The camera says ballast. The system says no.
Most weighbridge losses are not load-cell errors. They are a cheaper material declared for a dearer load, a tare that is a few hundred kilos heavier than that truck has ever weighed, a truck weighed with an axle off the deck, a plate typed for a different vehicle, and a ticket edited after the truck has gone. KiloSahihi closes each one at the bridge. Two of the checks are edge AI: the overview camera's frames are classified against the declared material, and every tare is scored against the vehicle's own history. Both run on the bridge PC, not in a cloud that adds seconds and a per-image bill.
The fraud and the control
Five ways a bridge loses money, and what stops each one.
- 1
Wrong commodity declared
A load of ballast is ticketed as dust at the dust price. The overview camera's edge AI classifies what is on the truck and flags a mismatch with the declared commodity before the ticket prints. The photo and the classification stay on the record.
- 2
Truck weighed half on
An axle left on the approach makes the truck lighter at tare or heavier at gross, whichever pays. Position sensors at both ends of the deck refuse the reading until the vehicle is fully on.
- 3
Wrong vehicle, wrong tare
A plate typed for a heavier truck's tare, a stored tare reused for months, or an empty truck that comes in heavy because something stayed on the deck. RFID tags or ANPR fix the identity from the vehicle itself, and every tare is scored on the bridge against that vehicle's own history: one that falls outside its normal band is held for a fresh weighing or a supervisor.
- 4
Weight typed or nudged
The weight comes from the indicator over the cable and locks only when stable. Manual weighing is a licensed, role-controlled exception, marked on the ticket, the report and the ERP posting.
- 5
Ticket changed later
Every edit, void, reprint and manual entry is logged with who, when and why, reported daily to the owner's phone, and the original values are kept.
- 6
Bridge quietly drifting
A bridge that reads light because a corner has drifted or a verification has lapsed is a loss nobody sees. Calibration-due alerts, test-weight checks and corner drift reports keep it visible.
Controls
Everything runs on the bridge.
Edge AI commodity check
The overview camera's frames are classified on the bridge PC or an edge device: commodity type, load present or empty, tarpaulin on or off. Mismatches with the ticket are flagged in the same second.
RFID vehicle identity
Windscreen or chassis tags read at the deck identify the vehicle without an operator typing a plate. ANPR does the same from the number plate; sites use one or both.
Position sensors
Loop, beam or laser sensors confirm the vehicle is fully on the deck before a weight is accepted, on road and axle bridges alike.
Stable-weight lock and audit
The reading locks only when the indicator reports stable; the operator cannot type a weight. Every change afterwards is logged and reported.
Tare anomaly detection
Every tare is scored on the bridge against that vehicle's own history and the fleet's pattern. A tare outside the vehicle's normal band is held for a fresh weighing or a supervisor, and a drift in one driver's or operator's tares over a shift is reported.
Calibration-due alerts
Verification and calibration dates per bridge with alerts before they lapse, test-weight check records, and a corner drift report from routine checks.
Owner's daily digest
Voids, manual entries, edits, overloads, commodity mismatches and rejected readings per bridge, delivered to the owner's phone and email every day.
Photo on every ticket
A picture of the truck at gross and at tare, stored with the ticket and posted with the transaction where the ERP accepts it.
No cloud dependency
Classification, identity and sensor checks run on site, so they work with the internet down and cost nothing per image.
Edge, not cloud
Why the AI runs on the bridge.
A material check or a tare check is only useful if it happens before the ticket prints, which means within a second of the weight locking. Sending each frame to a cloud model adds network latency, fails when the connection does, and charges per image at a site that weighs hundreds of trucks a day. KiloSahihi runs both models on the bridge PC or a small edge device beside the camera: the image classifier for the material, and the anomaly model that scores each tare against the vehicle's history. The result is immediate, works offline, and the images and weights stay on your premises. The cloud is used for what it is good at: the multi-site dashboard, the daily digest and long-term storage.
- Decision in the same second as the weight
- Works with the internet down
- No per-image or per-request cloud charges
- Images and weight history stay on site
Questions
Frequently asked
- How does the commodity check work?
- The overview camera captures the load while the truck is on the deck. A model running on the bridge PC or an edge device classifies what it sees, and the software compares that with the commodity on the ticket. A mismatch is flagged for the operator or supervisor before the ticket prints, and the image and classification are kept with the record.
- Is it accurate for every commodity?
- It is trained for the commodities a site handles: aggregate sizes, dust, sand, murram, cane, cherry, bagged versus bulk, empty versus loaded, tarpaulin on or off. Unusual loads are flagged for a human decision rather than guessed.
- How does tare anomaly detection work?
- Each vehicle builds up a tare history on the bridge. When it is weighed empty again, the new tare is scored against that history and the pattern across the fleet. A tare that is unusually heavy or light for that truck is held for a fresh weighing or a supervisor's approval, and the daily digest shows tares that drifted for one driver, operator or shift. The model runs on the bridge PC, so the check happens before the ticket prints and needs no connection.
- RFID or ANPR?
- RFID suits a known fleet: tags are issued to each vehicle and read without lighting or plate condition mattering. ANPR suits open sites with walk-in traffic. Many sites use RFID for their own trucks and ANPR for everyone else.
- What does a calibration-due alert do?
- Each bridge has its verification and calibration dates on record. The owner and the service team are alerted before they lapse, routine test-weight checks are recorded, and a drifting corner shows up in the report before it becomes a dispute.
- Does this need new hardware?
- The audit trail, stable-weight lock and tare anomaly detection need none; the tare model runs on the bridge PC from the bridge's own history. The commodity check needs an overview camera and a bridge PC or edge device that can run the model; RFID needs a reader and tags; position sensors are fitted at the deck ends. Each is added as the site needs it.
Keep reading
Weighbridge ANPR Cameras & Number Plate Recognition in Kenya
ANPR cameras read the number plate as the truck stops on the weighbridge, match it to the transaction and photograph every weighing, with weight and plate on one ticket.
Read moreUnmanned Weighbridge Automation in Kenya & East Africa
Automate a weighbridge in stages: ANPR cameras, boom barriers, traffic lights and driver self-service, until the bridge runs 24/7 with no operator on site.
Read moreAxle, Multi-deck & Rail Weighbridge Software in Kenya
Axle weighing with East African configurations and overload checks, multi-deck and twin bridges summed per deck, position sensors, and rail wagons weighed by wagon number.
Read moreKiloShop
Weighbridges, indicators, load cells, ANPR cameras and barriers, supplied and installed with the software.
Get started
Talk to us
Tell us what you weigh and how your site runs. We'll demo KiloSahihi with your own scenario.
+254 706 684 573 (call or WhatsApp)