Application of Deep Learning Methodologies in Steel Wheel Design

Wheel design represents a complex engineering challenge, characterized by constraints related to safety and structural efficiency. This study explores the integration of deep learning (DL) techniques into the optimization process of steel wheel design, with the goal of accurately predicting stress distributions, wheel weight, and material usage based on geometric inputs. A feed-forward neural network was trained on 172 FEM-based (finite element method–based) simulations, achieving high predictive accuracy (R2 values between 0.92 and 0.99 across all target variables). Once trained, the model is able to evaluate over 570,000 design configurations in less than 7 min, identifying optimal geometries that minimize critical stresses. Compared with conventional FEM- based workflows which typically require up to 5 days for equivalent optimization, the proposed approach drastically reduces computational time and enables rapid exploration of the design space, supporting more efficient and informed design decisions. Beyond the specific case study, the proposed workflow provides practical insights for the adoption of data-driven surrogate models in engineering applications traditionally relying on FEM-based design processes.