| Title |
Prediction of Forming Limit Diagram by Convolution Neural Networks and Crystal Plasticity Finite Element Method |
| Authors |
이한준(Han-jun Lee); 정재면(Jaimyun Jung); 오세혁(Sehyeok Oh); 김동규(Dong-kyu Kim); 김지훈(Ji Hoon Kim); 이호원(Ho Won Lee) |
| DOI |
https://doi.org/10.3365/KJMM.2025.63.7.542 |
| ISSN |
1738-8228(ISSN), 2288-8241(eISSN) |
| Keywords |
Forming limit diagram; Texture; Deep learning; Convolutional neural networks |
| Abstract |
This paper presents a novel deep learning (DL) approach to predict the forming limit diagrams (FLDs) of sheet metals, addressing the limitations of traditional experimental and numerical simulation methods. By utilizing orientation distribution function (ODF) data and crystal plasticity finite element (CPFE) results, an efficient DL prediction model was developed. First, numerous orientation datasets of the AA6061 alloy were generated and converted into ODFs to assess the impact of various texture components. Then, FLD data were computed using virtual forming tests via CPFE analysis and transformed into a format suitable for training the DL model. The model is based on the ResNet18 architecture and its hyperparameters were optimized using the Optuna framework. The model demonstrated rapid loss reduction, achieving an R2 score of approximately 0.7 on the test dataset. However, a decline in prediction performance was observed, particularly when minor strain values were greater than zero. This suggests the need to incorporate additional input data such as the positional information of representative volume elements in future work. This study demonstrates that leveraging microstructural data for FLD prediction via DL can reduce time and costs compared to traditional methods, and it opens new possibilities for predicting material behavior under diverse conditions.(Received 24 June, 2024; Accepted 21 April, 2025) |