Stage – AI inspection for inkjet printing with rapidly changing designs

VIVENTRIS – Smart production solutions for the industry of tomorrow

Viventris develops smart and flexible production solutions that help companies improve their processes and make them future-proof. We specialize in visual inspection and assembly, with applications in consumer products, high-tech systems, and healthcare, among others.

Our strength lies in combining generic building blocks with customer-specific customization. This allows us to realize efficient and reliable solutions for the production of compact and often complex products. We work pragmatically and closely with our clients to achieve results quickly and make a real impact on the shop floor.

Within our focus area of assembly, we automate manual processes and complex operations, leading to increased product quality and improved efficiency. Within our inspection expertise, we develop advanced vision systems utilizing both rule-based algorithms and AI. This enables accurate product inspection and reliable defect detection.

With our solutions, we contribute to smarter production processes and consistent product quality.

AI-powered inkjet print inspection for rapidly changing designs

In many printing processes, quality inspection is still based on template matching or manually programmed feature detections per product. This makes it time-consuming to add new designs, even though modern printers can technically produce these variations without issues. For new products or prints, the inspection often needs to be reprogrammed, while the printing process, and sometimes even colors and materials, remain the same.

In this internship, you will investigate how AI can make this inspection process more generic and reduce the engineering effort for new products.

Goal

Develop a flexible AI-driven print inspection approach that can start with limited or no labeled data and then quickly adapt to new prints and defect types.

What are you going to do?

• Investigating how visual print inspection can be built with limited labeled data in combination with synthetic data
• Applying techniques such as data augmentation, transfer learning, and domain adaptation to photos of previous products and prints to quickly develop an inspection model for a new print
• Compare the different supervised AI methods with an anomaly detection approach.
• Exploring continuous and incremental learning strategies (such as few-shot and online learning) to enable pre-existing models to grow with new products and defect types.

Required skills

• Experience with Python
• Interest in computer vision and machine learning
• Affinity with industrial applications or quality inspection
• Pre: experience with synthetic data or data augmentation techniques