Deep learning models for optically characterizing 3D printers.

Area of Science:

• * Additive Manufacturing
• * Optical Engineering
• * Computational Materials Science

Background:

• * Multi-material 3D printing enables complex material arrangements with diverse optical characteristics.
• * Accurate optical printer models are essential for predicting and reproducing these properties from printer control values (tonals).

Purpose of the Study:

• * To develop and evaluate deep learning-based models for accurate optical characterization of 3D printers.
• * To reduce the number of training samples required for high-accuracy optical property prediction.

Main Methods:

• * Development of a Pure Deep Learning (PDL) model, a black-box approach.
• * Implementation of a Deep-Learning-Linearized Cellular Neugebauer (DLLCN) model, leveraging deep learning to linearize the tonal-value-space of a cellular Neugebauer model.
• * Testing on two six-material polyjetting 3D printers to predict reflectance and translucency.

Main Results:

• * Both PDL and DLLCN models achieved high accuracy in predicting optical properties (reflectance and translucency).
• * The models required significantly fewer training prints compared to a standard cellular Neugebauer model.
• * Accuracies achieved were sufficient for most practical applications.

Conclusions:

• * Deep learning offers an effective strategy for optically characterizing multi-material 3D printers.
• * PDL and DLLCN models provide accurate predictions with reduced data requirements, advancing material design in 3D printing.
• * These models enhance the ability to reproduce desired optical properties in 3D printed objects.