Deep reinforcement learning based dynamic printing control in Large Format Additive Manufacturing
研究利用变压器架构预测打印表面温度，结合深度强化学习算法动态调整挤出速度，以提升大幅面增材制造的效率与产品质量。
Large Format Additive Manufacturing (LFAM) utilizes the fused filament fabrication technique, where an extruder continuously deposits melted thermoplastic material layer by layer to create structures typically exceeding 1m3 in volume. Since the temperature of the base layer significantly impacts the quality of the final product, dynamically controlling the extruder’s speed becomes essential, as it enables precise temperature management, ultimately enhancing both efficiency and product quality. To effectively control the extruder speed, it is essential to understand the thermodynamic behavior of the part’s surface during printing. This paper employs a transformer architecture to analyze the spatiotemporal relationships among cooling profiles across different locations on the part’s surface, providing accurate temperature predictions. Building on the accurate temperature predictions, this paper integrates these insights into a custom-designed environment that simulates the LFAM printing process. This environment serves as a foundation for training a deep reinforcement learning agent using the enhanced Environment Aware Proximal Policy Optimization (EAPPO) algorithm. By leveraging the spatiotemporal temperature predictions, the EAPPO agent dynamically adjusts the extruder speed to maintain optimal surface temperatures, ensuring consistent material deposition and improved part quality. The proposed method is validated through a real-world hexagon printing case study, demonstrating its capability to enhance printing efficiency and produce high-quality structures.