서강대학교

초록 (요약문)

The profiled ring rolling procss is hard to control, so major forming defects such as unfilling commonly occur. This research provides an insight into the performance of a deep neural network (DNN)-based algorithm for groove-section ring rolling process design. The main design variables were defined through finite element analysis. Each defined design variable was non-dimensionalized and utilized as training data through the DOE (Design of Experiment) technique, preventing the overfitting due to inappropriate data configuration. A total of 156 numerical analysis results were utilized as data for DNN model, 80% divided into training sets and 20% into test sets. In addition, process design was conducted through the commonly used regression method. The performance of each model was evaluated by calculating the groove formability index to identify unfilling defects in the cross-section. As a result, the DNN model showed 15.1% lower groove formability index than the regression model in average. The developed DNN model based on the applied data configuration method showed the potential to be used as a design technique for ring rolling processes that have high complexity and shape diversity.