The Real-Time Prediction of Cracks and Wrinkles in Sheet Metal Forming According to Changes in Shape and Position of Drawbeads Based on a Digital Twin

In the automotive industry, extensive research has been conducted to eliminate factors negatively impacting product quality, such as wrinkles, cracks, and thickness distribution in components. The application of drawbeads often relies on the experience of field workers, leading to considerable trial and error before stabilizing the production process. Therefore, to efficiently transform these inefficiencies related to time and cost, there is a need for real-time predictive technology for forming quality based on the position of drawbeads and the bead force. This study proposes a method for predicting formability in real-time, based on a digital twin framework that considers the position of drawbeads and holder force. A digital twin was developed to predict the sheet metal forming process using Support Vector Machine, Random Forest, Gradient Boosting Machine, and Artificial Neural Networks. The machine learning models were trained using finite element analysis data corresponding to the position and bead force of drawbeads, enabling the real-time prediction of wrinkles and crack occurrences. The accuracy of the machine learning models was demonstrated, achieving 100% accuracy in determining crack occurrence, with a mean squared error (MSE) of 0.141 for wrinkle prediction and 0.038 for crack prediction, thereby ensuring the accuracy of the forming prediction model based on drawbead applications. Based on these predictive models, a user-friendly GUI has been developed, which is expected to reduce design time and costs while facilitating real-time predictions of forming quality, such as wrinkles and cracks, on-site.