Previous Case Study: Patates Dolbec
Next Case Study: Restos Plaisirs

How to detect anomalies on a production line

Pulp and paper industry

Our client is a Canadian leader in the pulp and paper industry. They produce paper for domestic and industrial use, cardboard, packaging products and specialty papers, amongst others.


One of their factories uses cardboard boxes to package their products. They are transported on several conveyors, which are sometimes difficult to reach. While being carried, some of the boxes often fall or end up in an invalid position (at an angle, upside down, skewed, etc.). If not detected in time, these mishaps sometimes force the company to shut down the production line – a procedure that can be costly. 

To remedy the situation, the company installed a set of sensors along the conveyors to measure the height of the boxes and detect those that had fallen. This rudimentary equipment, however, failed to recognize boxes that were simply at an angle or skewed. This method thus proved to be very imprecise and regularly required human intervention. An employee was sometimes tasked with the constant inspection of all the factory’s surveillance cameras in order to confirm anomalies.


The typical approach for this kind of problem would be to install a more traditional, but very expensive (up to $30 000), vision system equipped with specialized cameras, lenses and other components to detect invalid boxes.

Our method is much simpler, but mostly a lot more affordable. First, we connected to the surveillance cameras that were already in the factory to deploy a vision system. Next, we trained an AI algorithm that detects boxes in invalid positions, while ignoring those that are upright. As soon as an anomaly is detected, the employees receive real-time notifications and can go straighten the box to avoid downtime further down the conveyor.


The vision system’s performance is significantly better than their old system. What's more, the company didn't need to buy expensive new equipment – we just had to connect to the infrastructure already in place and make their cameras "smart", allowing them to save time and money. The simplicity of this vision system also allows the solution to be deployed quickly, around a month in their case.