Machine-Learning-Guided design and inverse engineering of 3D-printed optical filters

• 3D printing via vat photopolymerization used to create low-cost tunable optical filters. • Machine learning models predict and design filter spectra with high accuracy. • AI-guided formulations enabled 3D-printed replicas of commercial optical filters. • Material tests show printed filters are elastic, stable, and recover after strain. Optical filters are essential components in imaging, sensing, and communication systems, where they selectively control light transmission across specific wavelength ranges. Additive manufacturing offers a flexible approach to producing optical devices with customizable shapes, dimensions, and spectral responses. In this work, color-tunable filters were produced using digital light processing 3D printers and alcohol-ink-doped photoresins (developed in-house) that span the entire visible spectrum. Their optical properties were strongly dependent on ink concentration, allowing subtle control of transmission intensity and bandwidth. To predict and inversely design the optical properties of the printed filters, artificial intelligence models were trained to connect ink concentrations and compositions with measured optical transmission spectra. The Multilayer Perceptron achieved the best performance in predicting ink compositions from known optical spectra (R 2 up to 0.99), whereas the Random Forest model most accurately reconstructed the optical spectra from known ink compositions (R 2 = 0.96). Additionally, rheological testing showed that the printed materials were mainly elastic and could fully recover after large deformations, while mechanical measurements confirmed adequate strength and tear resistance for practical handling. The study demonstrates that combining machine learning with material characterization can enable data-driven design of 3D-printed photonic devices that are both mechanically stable and spectrally tunable.