🔍 ORAFOL Halved Optical Inspection Times with Zebra Vision!

[Hasan S. Cemkan]

ORAFOL Halves Optical Inspection Times with Zebra Vision!​

ORAFOL Fresnel Optics GmbH has succeeded in halving the inspection time of its optical lenses thanks to Zebra Technologies' AI-based machine vision solution. Faced with the complexity of manually inspecting plastic components sometimes only 4 mm in diameter, the German company automated the process to guarantee continuous 24/7 operations and ensure end-to-end traceability for the more than 1.5 million lenses already inspected.

Automation of Complex Quality Control​

Before this integration, inspecting the microstructures of Fresnel lenses required 15 operators to examine each part under specific lighting. To overcome the visual fatigue and lack of consistency inherent in this repetitive task, ORAFOL opted for Zebra Aurora Vision Studio software combined with the Aurora Deep Learning module. This code-free platform, compatible with standard cameras, was paired with an optical turret specifically developed to capture the smallest details of the lenses. According to Rico Traxdorf from the Production and Technology team, the system now applies absolutely objective rejection criteria, eliminating any human interpretation variability when dealing with scratches or defects.

Scaling with Partner Integration​

Following the success of this initial implementation and an increase in production volumes, ORAFOL partnered with ID Engineering, a Zebra partner integrator, to design a second, higher-throughput inspection system for other types of optical windows. As Jason Harvey, VP of AI and Automation EMEA at Zebra Technologies, emphasized, integrating deep learning closer to production lines bridges the gap between digital and physical environments, solving complex industrial optical inspection challenges while freeing up operators for higher-value tasks.

Optical Challenges of Fresnel Lenses and Traditional Vision​

Automated inspection of Fresnel lenses represents one of the most demanding challenges in machine vision. Unlike a standard convex lens with a smooth surface, a Fresnel lens consists of a series of stepped concentric rings (micro-prisms). During camera inspection, these concentric ridges refract and reflect light chaotically, producing intense optical noise, moiré effects, and parasitic reflections. Traditional "rule-based" machine vision algorithms (those looking for sharp contrasts or defined geometric contours) cannot differentiate between a genuine scratch, a plastic short shot defect, and a normal reflection of a micro-prism. This is where Deep Learning (via convolutional neural networks) becomes indispensable. Instead of being programmed with rigid mathematical rules, the AI model is trained on thousands of images of "good" and "defective" lenses. Thus, it learns on its own to ignore the optical noise inherent in the Fresnel structure and focus only on structural anomalies, making it possible to automate the inspection of components as small as 4 mm in diameter.