Optimizing Thermal Uniformity in Laser Powder Bed Fusion through Deep Reinforcement Learning-Based Toolpath Generation

Temel Kavramlar

A deep reinforcement learning-based framework is developed to generate optimized toolpaths that achieve uniformly distributed temperature fields and avoid extreme thermal accumulation during the laser powder bed fusion process.

Özet

The paper presents a deep reinforcement learning (DRL)-based toolpath generation framework for the laser powder bed fusion (LPBF) process. The goal is to achieve uniformly distributed temperature fields and avoid extreme thermal accumulation regions during printing.

Key highlights:

• A simplified numerical model is developed by considering the relationship between turning angles and thermal distributions to improve computational efficiency.
• The reward function is designed to minimize the input energy density, aiming to ensure a stable temperature field.
• Special environments, such as unqualified-points, isolated-points, and sensitive regions, are defined and handled during the training process.
• Numerical simulations for a polygon case study demonstrate the effectiveness of the DRL-based approach in obtaining uniformly distributed temperature fields and avoiding excessive thermal accumulation.
• Experimental results show that the DRL-optimized toolpath can reduce the maximum distortion by approximately 47% compared to zigzag patterns, 29% compared to chessboard patterns, and 17% compared to adaptive toolpath generation (ATG) patterns.
• The study presents a promising approach for using machine learning to optimize toolpath patterns in the LPBF process and provides a foundation for further research in this area.

İstatistikler

The maximum depth of the molten pool increases significantly as the turning angle decreases below 90 degrees, but remains constant around 45 μm when the turning angle is greater than 90 degrees.

Alıntılar

"The elimination of extremely accumulated thermal regions will be accomplished by maximizing the reward function, which can be designed to minimize the energy density as much as possible."
"Our DRL-based approach provides a promising solution to address the problem of residual stress accumulation during the LPBF process."

Önemli Bilgiler Şuradan Elde Edildi

Deep Reinforcement Learning Based Toolpath Generation for Thermal Uniformity in Laser Powder Bed Fusion Process

by Mian Qin,Jun... : arxiv.org 04-12-2024

https://arxiv.org/pdf/2404.07209.pdf

Daha Derin Sorular

How can the proposed DRL-based framework be extended to consider inter-layer thermal accumulation in 3D geometries?

To extend the DRL-based framework to consider inter-layer thermal accumulation in 3D geometries, the reward function can be modified to take into account the thermal history of each layer. By incorporating the temperature distribution from previous layers, the algorithm can optimize the toolpath to minimize thermal accumulation across layers. Additionally, the neural network architecture can be adjusted to include memory units that store information about previous layers, allowing the model to make decisions based on the cumulative thermal effects throughout the printing process. This approach will enable the algorithm to optimize toolpaths not only within each layer but also across multiple layers to achieve more uniform thermal distribution and reduce inter-layer thermal accumulation.

How can the proposed DRL-based framework be extended to consider inter-layer thermal accumulation in 3D geometries?

To address the noise points generated around the boundary due to precision issues in uniform sampling, several strategies can be explored. One approach is to implement a filtering mechanism that identifies and removes noise points based on predefined criteria such as proximity to the boundary or deviation from the expected distribution. Another strategy is to refine the sampling algorithm to generate points more accurately, reducing the occurrence of noise points. Additionally, post-processing techniques such as smoothing algorithms can be applied to the generated toolpaths to eliminate irregularities caused by noise points. By implementing these strategies, the DRL-based framework can improve the accuracy and reliability of the generated toolpaths.

How can the computational efficiency of the toolpath generation process be further enhanced to handle more complex structures?

To enhance the computational efficiency of the toolpath generation process for handling more complex structures, several strategies can be implemented. One approach is to optimize the neural network architecture by reducing the number of parameters or implementing parallel processing techniques to speed up the training process. Additionally, the use of advanced optimization algorithms such as genetic algorithms or simulated annealing can help in finding optimal toolpaths more efficiently. Furthermore, leveraging cloud computing resources or GPU acceleration can significantly reduce the computational time required for training the model. By implementing these strategies, the DRL-based framework can efficiently handle the generation of toolpaths for complex structures while maintaining high computational efficiency.