Machines and belts working at the right pace

The AI locates heavy equipment on the construction site, mine or port — excavators, backhoes, wheel loaders, crawler dozers, rollers, cranes and trucks — and classifies, second by second, whether each one is operating (arm, boom or bucket working, or maneuvering back and forth with a load), traveling (moving without working) or idle. The result is each machine's real utilization, idle time, number of work cycles and an equivalent hour meter — without installing anything on the machines.

How it works

1. 1Open-vocabulary detection (YOLO-World): machines are found by text description, without specific training; each one's type is confirmed by a CLIP classifier on the crop and decided by voting during the first few seconds.
2. 2Camera movement (drone, swaying pole, zoom) is estimated from the scene background and discounted; only the machines' real movement is analyzed. Residual shake sets an adaptive noise floor.
3. 3Within each machine, movement is split into translation (the whole machine moving) and articulation (arm, boom, bucket or upper-structure swing moving differently from the body).
4. 4Operating = sustained articulation or back-and-forth maneuvers (wheel loaders, rollers, dozers); a stationary truck being loaded or dumping also counts as operating. Traveling = moving without articulating. Idle = no relevant movement, even with the engine running.
5. 53 s windows, hysteresis and a minimum dwell of 1.5 s in each state keep camera shake or dust from becoming 'operating'; people are detected and their movement is discounted. Pauses of up to 5 s between cycles (filling or emptying the bucket) still count as operating.
6. 6Cycles: for excavators and backhoes, each round trip of the arm between digging and dumping counts as one cycle; for wheel loaders, each back-and-forth round trip.
7. 7Alerts: machine idle beyond the configured time and fleet utilization below target.

Benefits

• Real per-machine utilization, with no onboard telematics, working with machines of any brand or age.
• Identifies trucks waiting for loads, idle machines with engines running and operational bottlenecks.
• Equivalent hour meter and cycles for preventive maintenance, rental measurement and productivity per shift.

A camera above the conveyor belt measures, frame by frame, how much of the belt width is covered by grain (soybeans, corn, wheat, coffee), meal, ore, crushed stone or sand. The system shows the load in real time and over time, identifies the belt running empty (energy and wear with no production), overload with a risk of spillover, and off-center load that causes mistracking and spillage — and combines load and speed into a flow index to balance the feed.

How it works

1. 1The user marks the belt section with 4 corners; the image is rectified (perspective removed) and split into transverse scan lines.
2. 2Text-prompted segmentation (CLIPSeg) separates material from empty belt, without training; a color and texture model fitted on the frame itself refines the load edge.
3. 3On each scan line, the fraction of the width covered is measured; the average is the load, and the load centroid indicates centering.
4. 4Apparent speed is measured by phase correlation along the flow, discounting the movement of the adjacent structure (compensates for camera movement).
5. 5Configurable bands (empty, low, optimal, overload) trigger alerts with photos: empty belt, overload/spillover and off-center load.

Benefits

• Eliminates hours of belts running empty — direct savings on energy and on belt and idler wear.
• Prevents overload and spillover, which cause product loss, cleanup and downtime.
• Detects off-center load before belt mistracking.
• Provides continuous visibility of the feed to balance hoppers, dryers and crushers.