Deep Learning-Based Super Resolution Applied to Finite Element Analysis of Fused Deposition Modeling 3D Printing

Area of Science:

• Additive Manufacturing
• Computational Mechanics
• Artificial Intelligence

Background:

• Finite element analysis (FEA) is crucial for predicting temperature and displacement in fused deposition modeling (FDM).
• Traditional FEA using fine meshes is computationally expensive, limiting its practical application.
• Coarse mesh models offer faster computation but sacrifice accuracy.

Purpose of the Study:

• To develop a deep learning (DL)-based super-resolution (SR) approach to enhance the accuracy of coarse mesh FEA models.
• To reduce the computational time required for accurate FEA predictions in FDM.
• To improve the quality of temperature and displacement predictions in FDM.

Main Methods:

• An analogy between FEA grids and image resolution was established, treating FEA elements as pixels.
• A modified super-resolution residual network was employed to reconstruct high-resolution (HR) results from low-resolution (LR) FEA data.
• The DL-SR model was trained to map coarse mesh results to fine mesh accuracy levels.

Main Results:

• The DL-SR approach significantly reduced the error between predicted and actual high-resolution results compared to traditional interpolation.
• Improved accuracy was demonstrated for both temperature and displacement fields in FDM simulations.
• Quantitative metrics, including Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM), confirmed enhanced image quality.

Conclusions:

• The proposed DL-based SR method effectively enhances the accuracy of coarse mesh FEA models in FDM.
• This technique offers a viable solution to reduce computational costs while maintaining high prediction accuracy.
• The study highlights the distinct mapping variations between temperature and displacement fields.