Photovoltaic production relies on specialised imaging to reveal defects that are hard to spot under normal lighting. The images are large, the relevant patterns can be small or ambiguous, and their significance depends on production experience. Reviewing them therefore demands close visual attention and specialist knowledge. A manufacturer wanted to explore whether artificial intelligence could help direct attention to suspicious regions, without replacing human judgement.

The project combined process analysis, literature research, image annotation, AI-based object detection and software development. A modular solution was built to analyse high-resolution inspection images and feed marked regions back into an existing quality-control workflow. The technical implementation, image-analysis performance and initial user feedback were each assessed separately.

The project delivered a reusable software foundation linking AI-based image analysis with an existing inspection workflow. Suspicious regions can be flagged for review, while production experts retain the final decision. Technical testing showed that images could be processed and the markings fed back through the intended workflow. A controlled image comparison and a brief visible pilot gave encouraging preliminary signs, but did not yet establish representative production accuracy or measurable productivity gains. For the client, the key benefit is a modular basis for extending visual assistance to further inspection tasks, without building a separate application for every use case. The work also clarified the most important next steps: cleaner, expert-reviewed data, independent evaluation, ongoing monitoring and controlled model management. This gives the manufacturer a concrete path forward while preserving the existing production process.