ForgeManufacturing

Forge Manufacturing operates six plants producing precision metal components, each running dozens of conveyor and press lines around the clock. Maintenance had historically followed a fixed schedule (inspect every 30 days regardless of actual wear), supplemented by reactive repairs whenever a line went down unexpectedly, which happened often enough to cause real production losses.

Plant managers knew that many failures showed visible warning signs, like a fraying belt or a misaligned roller, well before they caused a breakdown, but there was no consistent way to catch these signs between scheduled inspections. Maintenance techs simply couldn't be everywhere at once across six large facilities.

Forge wanted to use the plant floor cameras it already had installed for safety monitoring to also catch early wear signals, without asking maintenance teams to review hours of raw camera footage themselves.

Aivora AI's team spent five weeks at Forge's flagship plant, cataloguing the visible failure modes maintenance techs cared about most (belt fraying, roller misalignment, chain slack, visible corrosion) and collecting labeled footage of each. A computer vision model was trained to detect these wear patterns from the existing camera feeds, running inference at the edge on hardware already installed near each line to avoid streaming raw video back to a central server.

When the model flags a wear pattern above a confidence threshold, it creates a maintenance ticket automatically, complete with a timestamped image and the specific line and component affected, routed into Forge's existing maintenance ticketing system. Maintenance leads review flagged tickets and schedule repairs during the next planned downtime window rather than waiting for a full failure.

The flagship plant pilot ran for ten weeks before the model was retrained on footage from each additional plant's specific equipment and lighting conditions, with full six-plant rollout completed over five months.