In industrial settings, three modeling paradigms, white-box, gray-box, and black-box, are used for prediction, monitoring, classification, and performance assessment. With the rise of AI, black-box (data-driven) models are popular, but their accuracy can suffer with noisy, scarce, or incomplete data. To address this, informed neural networks (physics-informed/hybrid models) combine data with first-principles knowledge to enhance robustness, generalization, and physical plausibility. As hybrid modeling grows in importance, it is vital to teach these methods accessibly. This paper introduces the Fiber Extrusion Device (FrED) as an educational platform demonstrating informed neural networks in manufacturing. Students use FrED to connect models with real data, train hybrid models, and analyze how physics priors and data-driven residuals complement each other amid noise and data limits.
