Leveraging Cyclic Bending Under Tension Data and an Artificial Neural Network to Predict Extrapolated Strain Hardening Behavior of Dual Phase Steels

Abstract Sheet metal forming processes require accurate material behavior prediction at large strains to ensure optimal process design and final part quality. Cyclic bending under tension (CBT) testing enables mechanical characterization at these large strains but requires effective methods to translate the test data into useful material behavior predictions. In this study, an artificial neural network (ANN) is developed to predict large deformation material behavior from a CBT test, using force-displacement data as input. A comprehensive training dataset is generated through numerical simulations of the CBT process using the combined Swift-Voce hardening law to create diverse stress-strain curves. The ANN model is trained and validated using this dataset, then tested against experimental data from four dual-phase (DP) steel grades (DP590, DP780, DP980, and DP1180). Results demonstrate that the model effectively captures the stress-strain relationship across all tested steel grades and all strain regions. The model’s consistent performance, particularly in predicting behavior at large strains achieved through CBT testing, validates its potential for improving material characterization accuracy in sheet metal forming applications. This approach provides a robust method for extracting meaningful material behavior data from CBT tests, offering a practical solution for characterizing advanced high-strength steels under conditions representative of actual forming operations.