A 3D Preform Design Method for Reduction of Forging Load and Flash Volume and Improved Grain Flow
- 주제어 (키워드) Preform Design , Forging , Generative Artificial Intelligence , Grain Flow , Finite Element Analysis
- 발행기관 서강대학교 일반대학원
- 지도교수 김낙수
- 발행년도 2025
- 학위수여년월 2025. 2
- 학위명 박사
- 학과 및 전공 일반대학원 기계공학과
- 실제 URI http://www.dcollection.net/handler/sogang/000000079637
- UCI I804:11029-000000079637
- 본문언어 영어
- 저작권 서강대학교 논문은 저작권 보호를 받습니다.
초록 (요약문)
This study presents a novel 3D preform design method that utilises generative artificial intelligence to optimise the hot forging process for complex geometries. The proposed approach integrates a β-variational autoencoder (β-VAE) as a generative model for generating preform shapes and a deep neural network (DNN) as a surrogate model to predict forging outcomes efficiently. The methodology introduces two objective functions: the first minimises forging load and flash volume to enhance material efficiency, while the second optimises grain flow to improve the yield strength of forged components. A grain flow evaluation function was developed to quantitatively assess the density and alignment of grain structures relative to principal stress directions. The proposed design framework was validated through physical and numerical experiments on target geometries, including a brake calliper and an EV manifold. The results demonstrated significant reductions in forging load and flash volume, alongside improvements in grain flow alignment, resulting in enhanced mechanical properties. This study contributes to the advancement of 3D preform design by offering a systematic and efficient approach to address both process performance and microstructural optimisation. The findings highlight the potential of generative artificial intelligence to transform forging design, providing a robust solution for creating high-performance forged components across various industrial applications.
more목차
1 Introduction 11
2 3D Preform Design Method 15
2.1 Definition and Discretisation of 3D Forged Shape 15
2.2 3D Preform Design Procedure 16
2.3 Initial Training Set Generation From Isosurfaces of Laplace's Equation 18
2.4 Generative Artificial Intelligence Model: β-VAE 18
2.5 Surrogate Model: Deep Neural Network (DNN) 21
2.6 Objective Function for Forging Load and Flash Volume 21
3 Grain Flow and Evaluation Function 23
3.1 Grain Flow and Yield Strength 23
3.2 Grain Flow Evaluation Function 25
4 Experiments 28
4.1 Experiments for Forging Load and Flash Volume 28
4.1.1 Target Forged Shapes for Forging Load and Flash Volume 28
4.1.2 Forging Experiments With Model Material 28
4.1.3 Polymer Clay Compression Tests 30
4.1.4 Similarity Law for Model Material 31
4.2 Numerical Experiments for Grain Flow 32
4.2.1 Microstructural Analysis and Grain Flow Evaluation 32
4.2.2 Finite Element Model of the Microstructure 34
4.2.3 Cohesive Zone Model and the Material Properties 34
4.2.4 Target Forged Shapes for Grain Flow 36
5 Results and Discussion 41
5.1 Preform Design for Forging Load and Flash Volume 41
5.1.1 Model Training and Preform Shape Decision 41
5.1.2 Forging Experiments 51
5.2 Preform Design for Grain Flow 54
5.2.1 Grain Flow Evaluation Function and Yield Strength 54
5.2.2 Preform Design and Grain Flow 56
5.3 Comparison With Previous Studies 61
5.3.1 3D Preform Design Using Voxel and CNN 61
5.3.2 2D Preform Design 62
6 Conclusion 65
A Generated preform shapes from β-VAE 67
B Voxel and CNN Based Preform Design Method for 3D Shapes 69
C Generative Artificial Intelligence Model for 2D Shapes 71
References 72
