A dairy unit is moving to a verified count for every crate it ships

Challenge

A dairy line moves a large number of milk packets into crates every shift. When packets are counted by hand, a short crate can leave the plant without anyone noticing, and the gap only shows up when a distributor complains.

The dairy unit wanted every crate to leave with a count it could trust, without slowing the line.

Hand counting also takes people away from other work on the floor, and it gets harder to do accurately as volumes grow or shifts run long.

Solution

We are building an AI vision system for the line. It runs on site at the plant, without sending images to the cloud, so it keeps working whatever the connection.

Running the system on site matters on a plant floor. Images stay inside the plant, there is no dependence on an internet link, and checks happen at the speed of the line rather than after a round trip to a data centre.

• Count: cameras number every packet in every crate as it passes.
• Flag: a crate that is short is flagged before it is dispatched.

Results

The system is in development, so there are no production results to report yet. We would rather say that plainly than publish numbers before the system has earned them.

The same approach, cameras and checks running on site, can apply to other lines where items are counted or inspected before they ship.