AI solutions for manufacturing

Project challenge:

Quality control in a manufacturing setting can be a difficult challange. There is a set of problems, where out-of-the-box solutions can help, however there is also an abundance of challenges, where bespoke models are required.

We solved such a complex quality control problem for a global tyre manufacturer. We created a solution for real-time detection of multiple defects in images of tyres (molding, cord spacing, air bubbles, external particles, stains) within the production process.

We applied and developed multiple neural networks (CNN, defect segmentation, anomaly detection, image enhancing) machine learning architecture optimized for quality of the prediction and minimal processing time.

As a result, we reduced human QA cost by over 80% while improving the detection rate of several complex defects. Additionally, our solution enables precise assessment of defect type, severity and location – even in cases barely detectable by a human expert.