| Abstract [eng] |
Stereolithography is an additive manufacturing technique that fabricates three-dimensional objects with high speed and precision by selectively exposing a liquid photopolymer using a laser beam. It is most commonly applied in prototyping and the production of medical devices and implants. To achieve highly dimensional accuracy, stereolithography processes are typically described using physical models, chemical kinetic models, or hybrid approaches, while certain aspects of the process can also be optimized using machine learning. However, under high exposure conditions, the experimentally observed relationship between exposure and photopolymer cure depth – known as the working curve – occasionally deviates from classical models. One proposed mechanism for this deviation is the optical bleaching phenomenon: as the photopolymer cures, its optical properties change, allowing greater light penetration and increasing the exposure received by the uncured material. In this thesis, a data-driven approach is developed that extends a classical physics-based model to account for optical bleaching, and the model is validated using real experimental data. The extended model uses four parameters that are optimized in a two-stage function fitting process. Two parameters, identical to those in the original model, are extracted from the linear portion of the data, while two additional parameters are optimized from the full data. The performance of the extended model was evaluated using two experimental datasets comprising a total of 75 experiments. For fully linear experiments, both models achieved similar accuracy, indicating that the extended model does not deviate from the original formulation under normal conditions. For super-logarithmic cases, where the working curve increases at higher exposures, the extended model consistently outperformed the baseline. However, for the sub-logarithmic cases, where the working curve decreases with higher exposures, the extended model underperformed relative to the original model. These results indicate that optical bleaching effectively captures only the super-logarithmic behavior of the working curve. Finally, the consistency of the extended model was evaluated using three similar experiments, demonstrating that the model remains relatively stable under comparable conditions. |
