When dealing with multiple products in a traditional computer vision project, we are generally forced to create a different program (recipe) for each of them.
This complicates the startup process and, above all, its maintenance. On one hand, commissionings take a long time because each new product involves new programming and its subsequent validation. On the other hand, maintaining these systems is a real nightmare because it's impossible to remember all the details of the dozens of generated recipes, and if improvements are needed, they must be manually incorporated into all products.
A clear example of this problem is found in a client who manufactures machinery for positioning plastic bottles. This client incorporated traditional machine vision systems into their machines to detect whether the bottles were oriented in one way or another. These vision systems were clearly the bottleneck in the commissioning of their machines because each new installation involved programming all formats. Furthermore, the maintenance of these machines was very costly.
The solution we offered to our client was to use a Deep Learning system to position the bottles without the need for recipes. Through our OneVision software, we trained a deep learning model capable of recognizing any bottle format.
The system, now standard on their machines, is infinitely more robust than the former one and is also generic for multiple formats. In other words, each new machine works without the need to prepare any specific vision program. The AI vision system recognizes plastic bottles of any format and can orient them in a generic way without specific programming.
With just one program (recipe), the position of all bottles is detected (even those it has never seen before). This has simplified the commissioning and final adjustment of their machines, reducing programming hours and associated costs. Furthermore, maintenance of these machines is much simpler since the systems are much more robust and adding new formats does not require any changes to the vision program.
Examples of some formats:
