검색 상세

A Process-Response-Guided Data-Driven Framework for Preform Design in Hot Forging Processes

  • 박준희 (서강대학교 일반대학원)

초록(요약문)

Hot forging is widely used to manufacture high-strength mechanical components because it provides excellent structural integrity, dense microstructures, and reliable mechanical performance. However, the quality and efficiency of hot forging strongly depend on the preform geometry. An inappropriate preform can cause underfill, excessive flash, folding defects, high forging load, non-uniform deformation, and undesirable microstructural evolution, which may lead to poor dimensional accuracy, reduced material yield, die failure, and non-uniform mechanical properties. Conventional preform design is still largely dependent on expert intuition, repeated finite element method simulations, and trial-and-error modification. Therefore, a systematic and automated preform design methodology is required for complex forged components. This dissertation develops a process-response-guided data-driven preform design framework for hot forging processes. The proposed framework integrates FEM-based dataset generation, image- and voxel-based geometry representation, defect and field-variable representation, deep learning-based preform prediction, quantitative quality evaluation, FEM-based physical validation, geometry reconstruction, and automated workflow implementation. The overall objective is to reduce the dependence on empirical design experience and to provide a practical design-support methodology for complex forging preform design. First, the data-driven preform design concept was applied to deformation homogeneity and microstructural uniformity in flex spline forging. A CNN-designed preform was compared with an existing industrial preform through FEM simulation, hot forging, heat treatment, microstructural analysis, and hardness measurement. The CNN-designed preform redistributed effective strain across the gear, diaphragm, and flange regions and reduced the standard deviation of effective strain from 0.81 to 0.42. It also reduced the standard deviations of grain size, grain orientation spread, kernel average misorientation, and hardness by 59%, 22%, 55%, and 74%, respectively. These results demonstrate that preform design affects not only macroscopic forming quality but also microstructural evolution and mechanical-property uniformity. Second, a defect-aware three-dimensional CNN-based preform design framework was developed for complex hot forging geometries. The target shape, preform, forged result, flash, underfill, and folding regions were represented using voxel grids, allowing the preform design problem to be formulated as a three-dimensional learning problem. The Forging Volume Efficiency Index was used to quantitatively evaluate preform quality by considering volume balance, flash, underfill, and folding. The proposed framework was applied to a brake caliper, EV manifold ＃1, and EV manifold ＃2. The final preforms achieved FVEI values of 2.99, 2.20, and 2.54, respectively, while eliminating or significantly reducing major forming defects and satisfying the forging load constraint. Generalization tests using unseen geometries further showed that the trained model could provide physically meaningful preform candidates for new forging components. Third, a manufacturable extruded preform design framework was developed for an industrial A6082 EV manifold. A self-evaluating 3D TransU-Net was used to generate extrusion-compatible preform geometries and predict forging quality. Active latent optimization and geometry reconstruction were then applied to obtain a practical extruded preform profile. The final hybrid design achieved a penalty-free FVI of 2.63, reduced input volume by 2.65% relative to the reference profile and reduced forging load by 23.5% in the blocker stage and 30.2% in the finisher stage. The design was validated through FEM simulation and a 1600-ton press trial, confirming its industrial applicability. Finally, NSM Designer was implemented as an integrated design-support platform that organizes the developed preform design methodology into a configuration-driven FEM–AI workflow. The system integrates initial data generation, FEM simulation, field extraction, voxel conversion, neural-network training, preform prediction, geometry reconstruction, FEM validation, quantitative shape-defect evaluation, process-response evaluation, integrated objective-score evaluation, result analysis, monitoring, and iterative dataset accumulation. In addition, the platform was extended to a process-response control problem related to die-life improvement, where intermediate-stage geometry was iteratively improved to reduce the maximum principal stress range in the final forging stage. Overall, this dissertation demonstrates that process-response-guided data-driven preform design can provide a practical alternative to experience-dependent trial-and-error design in hot forging. By linking geometry prediction, defect suppression, deformation homogenization, microstructural stability, manufacturability, FEM validation, automated iteration, and software-based design support, the proposed framework provides a foundation for intelligent forging process design.

more

목차

Contents Ⅰ
List of Tables Ⅴ
List of Figures Ⅵ
Abstract ⅩⅩ

1. Introduction 1
2. Background and Unified Methodology 4
2. 1. Hot Forging and Preform Design 4
2. 1. 1. Role of Preform Design in Axisymmetric Component 4
2. 1. 2. Role of Preform Design in Non-Axisymmetric Component 5
2. 2. Data-Driven Preform Design Framework 5
2. 2. 1. Pixel-Based Framework for 2D Preform Design 5
2. 2. 2. Voxel-Based Framework for 3D Blocker Preform Design 6
2. 2. 3. Voxel-Based Framework for 3D Extruded Preform Design 8
3. Strain and Property Homogenization in Flex Spline Forging 9
3. 1. Flex Spline Forging Process and Material Conditions 10
3. 1. 1. Material Properties and Chemical Composition 10
3. 1. 2. Temperature Measurement and Boundary Conditions. 11
3. 1. 3. Grain Growth Model 12
3. 2. CNN-Based Preform Design and Forging Performance 13
3. 2. 1. Target Shape and Data Structure. 13
3. 2. 2. CNN Model and Iterative Preform Design 14
3. 2. 3. Existing and CNN-Designed Preforms 15
3. 2. 4. Influence of Preform in the Forging Process 17
3. 3. Microstructural and Mechanical Property Evaluation 21
3. 3. 1. Microstructure of Flex Spline after Hot Forging 21
3. 3. 2. Microstructure of forged flex spline after heat treatment 26
3. 3. 3. Hardness Analysis 33
3. 4. Summary 36
4. Volume-Driven Preform Design for Complex Three-Dimensional Forging 37
4. 1. Voxel-Based 3D Preform Design Framework 38
4. 1. 1. CNN Model Structure and Training Objective 38
4. 1. 2. Initial Position Encoding in Voxel Space 41
4. 2. Forging Volume Efficiency Index and Dataset Configuration 43
4. 2. 1. Forging Volume Efficiency Index 43
4. 2. 2. Target forging shape 45
4. 2. 3. Initial Dataset Generation 46
4. 2. 4. Deformation-Based Data Augmentation 48
4. 3. Numerical Validation and Design Results 49
4. 3. 1. Material Properties 49
4. 3. 2. Effect of Defect and Load Constraints 51
4. 3. 3. Iterative Preform Design Results 56
4. 3. 4. Experimental Validation Using Polymer Clay 60
4. 3. 5. Generalization to Unseen Geometries 62
4. 4. Summary 66
5. Manufacturable Extruded Preform Design and Industrial Validation 67
5. 1. Extruded Preform Design Framework 67
5. 1. 1. Target Component and Industrial Motivation 67
5. 1. 2. FEM-Based Dataset Generation 68
5. 1. 3. Defect Representation and Forging Volume Index 69
5. 1. 4. Initial Datasets 70
5. 2. Self-Evaluating TransU-Net and Target-Specific Design 71
5. 2. 1. Network Architecture and Training Objective 71
5. 2. 2. Preform Inference and Geometric Reconstruction 73
5. 2. 3. Dataset Accumulation and Model Convergence 75
5. 3. FEM and Press Trial Validation 76
5. 3. 1. Experience-Based Design 77
5. 3. 2. Data-Driven Design 78
5. 3. 3. Hybrid Design 79
5. 4. Summary 81
6. NSM Designer: Integrated Platform for Data-Driven Preform Design 82
6. 1. Design Concept and System Overview 83
6. 2. Master Configuration and Initial Data Generation 85
6. 3. Objective Definition and Integrated Evaluation Score 86
6. 4. Result Visualization and Monitoring 89
6. 5. Automated Iterative Design Loop 90
6. 6. Demonstration of Iterative Preform Improvement 91
6. 7. Summary 92
7. Conclusions 93
7. 1. Overall Conclusions 94
7. 2. Main Contributions 95
7. 3. Future Works 96
Reference 98

more
반출 Meta View 목록