Evolving regulations and consumer expectations are driving the food packaging industry to adopt new can coating chemistries, but evaluating how these coatings interact with food ingredients remains slow, empirical, and limited in mechanistic insight. We combined optical profilometry (microscopy) with unsupervised AI to automatically detect and classify coating defects across 1,000+ unlabeled images, linking defect types to the food ingredients that drive them and showing that image-derived features can predict results from conventional analytical techniques -- building predictive tools as a step toward rational design for next-generation food contact coatings.
Stiphany Tieu | Materials Science & Engineering