news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

editor: lrst

[new zhiyuan introduction] research teams from the chinese university of hong kong and other institutions have developed a 3d printing path planner through deep reinforcement learning (dqn), which effectively improves printing efficiency and accuracy and opens up a new way for intelligent manufacturing.

with the rapid development of 3d printing technology, how to generate efficient and accurate printing paths on complex geometries has become one of the key challenges restricting its widespread application.

recently, research teams from the university of manchester, boston university and the chinese university of hong kong jointly proposed an innovative deep reinforcement learning (dqn)-based path planner at siggraph aisa 2024, which can generate optimized paths on graphs with a variety of different structures. 3d printing path significantly improves the efficiency and accuracy of the printing process.

paper link: https://arxiv.org/pdf/2408.09198

project link: https://rl-toolpath-planner.github.io/

in 3d printing, the path planning problem can be viewed as finding the optimal path among a sequence of nodes on a given graph. the complexity of this problem lies not only in the different graph structures of different models, but also in the large number of nodes and edges in the graph.

traditional methods, such as brute force search and heuristic algorithms, are usually difficult to provide the global optimal solution in a short time due to high computational complexity. however, the dqn optimization strategy proposed in this study dynamically constructs a local search graph (lsg). and perform path selection in it, which greatly reduces the computational complexity, enabling real-time path planning when processing graphs containing thousands of nodes.

method innovation and technological breakthrough

the core innovation of this path planner is its flexibility and adaptability. the research team designed an ingenious algorithm to encode the local search graph into a state space, so that under similar configurations, previously learned dqn policies can be reused, further speeding up the calculation of path planning.

the planner can adapt to a variety of 3d printing application scenarios by defining different reward functions, including wireframe structure printing, continuous fiber printing, and metal powder bed fusion printing.

in physics experiments, the planner demonstrated excellent performance. in wireframe structure printing, the researchers successfully printed a model containing up to 4,200 pillars, with deformation controlled within 1 mm.

in continuous fiber printing, this method can effectively avoid more than 93% of sharp-angle turns, greatly improving the printing quality; in metal printing, through optimized path planning, thermal deformation is reduced by nearly 25%, significantly improving the quality of the printed parts. accuracy and stability.

technical details

in this study, the research team proposed a flexible reinforcement learning strategy through diversified graph path planning to deal with different 3d printing challenges. in the specific implementation, they use a deep q network (dqn) as an optimizer to decide the next best node to visit.

this strategy greatly improves the efficiency of path planning by building a local search graph and using historical data to accelerate calculations. in order to further improve the performance of the model, the researchers designed an algorithm that can identify local search graphs under similar configurations, so that previously learned dqn policies can be reused. this technique not only speeds up path calculations, but also enables the method to be used flexibly in a variety of 3d printing applications.

experimental verification and wide application

in order to verify the actual effect of this method, the research team conducted physical experiments on a variety of models, covering a variety of graphic types from simple geometry to complex structures. experimental results show that whether it is wireframe structure, continuous fiber or metal printing, the path planner can generate optimal paths that meet manufacturing requirements and significantly shorten the calculation time.

for example, in one experiment, the total path planning time for using the planner to generate a complex wireframe model was only 2.05 hours, while the time to print the entire model was 6.67 hours. in contrast, traditional brute force search methods require hundreds of hours of computing time even if they are only performed within a local search scope.

the research team also noted that this learning-based approach opens up new possibilities for future 3d printing technology. by introducing reinforcement learning, path planning no longer relies solely on preset rules or heuristic algorithms, but can self-adjust and optimize based on actual printing conditions. this not only improves the success rate of printing, but also reduces material waste and the risk of printing failure.

outlook

the success of this research marks an important milestone in the field of 3d printing path planning. the path planner based on deep learning not only provides an efficient solution for printing complex geometric structures, but also lays the foundation for the development of future intelligent manufacturing systems.

the research team plans to further expand the application scope of this method in the future, including exploring its potential in fields such as multi-material printing and micro-scale printing. in addition, by combining more advanced machine learning models and optimization algorithms, researchers hope to further improve the efficiency and accuracy of path planning and provide stronger technical support for industrial production and scientific research applications.

overall, the application of learning-based diversified graphical path planners in 3d printing provides new ideas and tools for solving complex manufacturing problems, and is expected to be used in high-precision fields such as intelligent manufacturing, aerospace, and medical equipment in the future. important role.