Search

Directed Energy Deposition (DED) is crucial in the ongoing industrial revolution, providing a unique ability to fabricate high-quality components with complex shapes. However, the determination of key process parameters, such as scan sequence, laser power, and scanning speed, often relies on offline trial-and-error or heuristic methods. These methods are not only suboptimal but also lack generalizability. A major challenge is the non-uniform temperature distribution during manufacturing, which affects the uniformity of the mechanical properties. To overcome these challenges, we have developed a framework based on Deep Reinforcement Learning (DRL). This framework dynamically adjusts process parameters to achieve an optimal control policy. Additionally, we introduce a cost-effective temperature simulation model of the deposition process. This model is particularly useful for researchers testing the proximal policy optimization algorithm. The experimental results demonstrate that DRL policies substantially improve temperature uniformity in Inconel 718, enhancing hardness variability with improvements of 31.8 % and 27.1 % in horizontal and vertical building directions, respectively. This research marks an important step toward achieving a highly intelligent and automated optimization of process parameters. It also proves to be robust and computationally efficient for future online implementation.