Signed distance function-based geometric deep learning for path optimization of large-sized plates reinforcing with directed energy deposition
ADVANCES IN ENGINEERING SOFTWARE, cilt.222, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 222
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.advengsoft.2026.104293
- Dergi Adı: ADVANCES IN ENGINEERING SOFTWARE
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Applied Science & Technology Source, Compendex, INSPEC, zbMATH, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO)
- Bursa Uludağ Üniversitesi Adresli: Evet
Özet
Reinforcing large-sized plates with the Directed Energy Deposition (DED) process is more advantageous for plates with uniform thickness and higher weight. In aerospace and other high-performance industries, the ability to locally reinforce large-sized plates with DED enables weight reduction while maintaining mechanical integrity, resulting in more efficient and optimized designs. Such approaches are particularly valuable for critical components where traditional manufacturing methods may be limited in flexibility or cost-effectiveness. This study proposes a novel approach for the reinforcement and design optimization of large-sized plates, leveraging Geometric Deep Learning (GDL) techniques to efficiently capture and exploit complex geometric features efficiently. For geometric and shape design, a continuous parametric modeling approach based on Signed Distance Functions (SDF) is employed. During the design process, a beta-Variational Autoencoder ((3-VAE) is utilized for both dimensionality reduction and shape reconstruction. Structural properties of the plates, such as mass, stiffness, and frequency, are efficiently and reliably predicted using a Convolutional Neural Network (CNN). The latent variables, serving as design parameters, are combined with a Genetic Algorithm (GA) to search for the optimum design. This process allows the emergence of reinforcement paths with varying heights, widths, and curvature radius, enabling highly flexible structural solutions. The results demonstrate the effective application of the (3-VAE as a generative model for producing diverse and optimized design candidates. SDF-based GDL provides significant advantages for learning three-dimensional shapes and performing optimization through latent variables. This approach offers an effective strategy that integrates data-driven and generative modeling in engineering design.