Synthetic data generation framework for quality control automation in gravure printing
Creating fake defects to train machines that catch real printing flaws
A printing factory can now automate quality control without manually photographing thousands of defects. Researchers built software that generates synthetic images of common printing problems—creases, streaks, misalignment—and trained an AI detector on these fake images. When tested on real factory output, the detector caught defects with 80.9% accuracy, matching the performance of systems trained on actual photos.
Rotogravure printing currently relies on human inspectors to spot flaws, a slow and inconsistent process that slows production. This framework eliminates months of manual photo collection, letting factories deploy automated quality control within days instead. Since the synthetic images are generated instantly at zero cost, even small printers can afford to automate their inspection lines.