The custom metal tin, a reusable and recyclable premium packaging, is one of Massilly's flagship products. It is now inspected with a vision solution comprising two levels of AI, developed and integrated by Psycle. Patrice Ferrero and Emeline Lautissier deployed this solution on the production line; they tell us about their experience.
Why two levels of AI?
Patrice Ferrero: We needed to automate inspection on a line that handles a multitude of designs. The human eye can tell a defect from a design element, whereas a simple camera is less capable of doing so. We absolutely needed an intelligent system. We started by working on defect detection, which is the most complex topic, with a supervised learning AI. Then we added a simpler level for design analysis. It allows us to reduce the time spent on the training runs launched before each production order.
And why Psycle?
PF: From the very first tests carried out at Psycle, it seemed to me that their AI could meet our needs. This was confirmed when we worked on our own site, with a system made available to us for validation trials.
Psycle's AI lets us choose what we want to detect, and detect it without false positives.
What feedback have you had from users?
Emeline Lautissier: Line operators are autonomous in adjusting the cameras and launching training runs at the start of each production order. The Psycle solution is accessible on their usual screen, and easy to use. The quality department, for its part, regularly feeds the defect detection algorithm with new annotated images. The process is very smooth.
It was a great project, and a discovery of AI-assisted vision that went very well thanks to the support of Natan and Baptiste.