In composite-based VPP (Vat Photopolymerization) printing, increased resin viscosity and light scattering induced by fillers lead to simultaneous degradation of printing stability and dimensional fidelity. These issues manifest in various forms, inclu...
In composite-based VPP (Vat Photopolymerization) printing, increased resin viscosity and light scattering induced by fillers lead to simultaneous degradation of printing stability and dimensional fidelity. These issues manifest in various forms, including elevated separation force during layer detachment, boundary over-curing caused by cumulative exposure, channel blockage, and geometric distortion.
Although parameter tuning and empirical correction methods have been used to alleviate such problems, their effectiveness is limited because the optimal conditions must be repeatedly reconfigured whenever the geometry or material system changes.
To address these limitations, this study proposes a process optimization framework that integrates physics-based simulation with reinforcement learning. A separation-force model incorporating layerwise cross-sectional area and resin viscosity was established, and the stresses generated during the detachment process were quantified through finite element analysis (FEA). These physical models were incorporated into the reinforcement learning reward structure, enabling the agent to autonomously determine layer-specific bed-lifting speeds.
Experimental validation confirmed that the learned policies effectively identified stable operating conditions even for mechanically fragile composite structures, showing clear reductions in localized stresses and improved printing success rates compared with fixed-speed conditions.
To further mitigate light-scattering and cumulative exposure issues inherent to composite resins, a three-dimensional curing simulation model was developed using a Gaussian PSF–based optical diffusion model, a Beer–Lambert attenuation model, and parameter identification through Jacob’s working curve. This simulator was integrated into the reinforcement learning environment, where grayscale values at the patch level were defined as actions and optimized to minimize layerwise curing error. The learned grayscale distributions significantly reduced over-cured regions compared with the uniform grayscale-255 condition, improving geometric fidelity and preventing channel occlusion in printed gyroid structures.
Overall, this study demonstrates a process optimization system that quantitatively evaluates physical phenomena in VPP printing and autonomously improves these metrics through reinforcement learning. The simulation-driven learning approach eliminates the need for extensive pre-acquired datasets and ensures broad applicability across diverse geometries and composite material systems. The proposed framework provides a foundation for automated composite-based DLP printing and future development of shape-adaptive, material-aware optimization strategies.