Technature, a full-service cosmetics manufacturer, has decided to reinforce its quality control with vision solutions. The first application is a production line for cosmetic patches, which are then packaged. Mathis Le Floch managed the entire algorithm training phase on a cell provided by Psycle; he gives us an overview of the project.
What did you need Psycle for?
Mathis Le Floch: Our main objective was to recover the bags containing “compliant” patches that had been thrown away. We also wanted to relieve the operators in charge of inspection on this line. This is a high-speed line (6000 patches / hour); tracking down defects in appearance is very tiring, and sometimes non-conformities can slip through the cracks. We thought vision would be a good tool for automating control. We wanted a solution that could recognize a maximum number of different defects (creases, air bubbles, scratches, overlaps, missing parts, etc.), send a signal in the event of non-conformity and automatically trigger the ejection of the product concerned. We also wanted a system in which we could trace non-conformities, which is what Psycle offers with its PAQ software.
I learned a lot about AI-assisted machine vision through my discussions with Psycle. I now know that well thought-out AI is essential for detecting a product and its defects.
How did the Psycle team support you?
MLF: The Psycle team was at our side for 4 months. The exchanges were very fruitful, and as soon as we had an idea, we discussed it together. Psycle usually launched the necessary developments immediately. As we were new to vision, Psycle’s proximity and responsiveness were very reassuring. As did the gradual support we received as we moved towards autonomy.
How has the availability of equipment helped you?
MLF: Our project seemed complex: we have a huge number of references with different products (size, color, etc.), many defects and high production rates. Leasing appeared to be a way of validating feasibility while reducing risks. We were also able to develop our specifications. What’s more, we were able to extend the inspection to include pouches, which we hadn’t planned to do, without particularly lengthening the industrialization time.