Keeping an AI on it

Manual inspection of seals, even by expert operators, can be prone to inconsistencies. To overcome this, the Trelleborg facility in Czechowice, Poland, has embraced digital transformation by developing an in-house, AI-powered automated visual inspection (AVI) system to support its quality process.
​

“What sets this project apart is that development of the AI model and software was entirely done in-house by our AI, Data Science, and Digital Manufacturing teams,” says Arkadiusz Chrobaczynski, Engineering Manager, Manufacturing. “The cross-functional collaboration ensured a solution tailored to the unique challenges of inspecting flexible rubber.”
​

The new system inspects up to 2,200 seals per shift, using a rotating table, four high-resolution Keyence cameras, and a ResNet-based AI model to analyze each part.

Overcoming Complex Challenges

The project was not without its difficulties. The part chosen for the AVI system’s initial project was a "wobbly gasket."  This was mechanically difficult to handle and keep stable for the AI cameras. The complexity led to challenges such as seal twisting, inconsistent defect presentation, and hardware sensitivity.
​

“The team overcame these challenges through iterative model training, hardware redesign and expert review,” Chrobaczynski explains. “With the AI-controlled AVI system, each seal is analyzed and sorted in just 12 to 15 seconds, which is about three times faster than manual inspection. It gives zero false positives, something that is almost impossible to achieve manually. Despite the sensitivity of the equipment that could mean something as minor as a water droplet could be seen as a fault, false rejects run at around 25% — meeting all KPIs for the safe launch phase. "
​

The successful implementation has not only improved quality but also reallocated employee resources. The three operators previously performing a repetitive manual task can now be reassigned to other valuable areas of the business. "This is important for staff retention in a demanding labor market. We are more likely to keep employees who have fulfilling jobs," says Chrobaczynski.

Paving the Way for the Future

The lessons learned from this challenging initial project have been invaluable. An internal playbook has been created to guide future vision AI projects, ensuring a more streamlined process. This playbook is being shared across the organization to serve as a starting point for other teams.
​

The next project will benefit from this experience. It involves a stiffer, rubber-to-metal part and will be integrated directly into the production line, providing a much shorter feedback loop.
​

"The deployment of the new system will be much faster as we will execute the playbook that we drafted," explains Chrobaczynski. The goal is to develop a scalable solution that can be extended to other Trelleborg production facilities, adapting the specific AI model for each unique part while leveraging the standardized project approach.