Optimized Design of Mechanical Cantilever Structure Parameters and Stress Field Inversion Study Based on Deep Neural Network

With the advancement of industrial technology, the increasing complexity and diversity of mechanical structures impose heightened demands on engineering design and manufacturing processes. Traditional stress–strain analysis methods, however, often encounter limitations in both accuracy and efficiency, particularly when addressing complex geometries and nonlinear material properties. To overcome these challenges, this study introduces the DGUNet framework, which integrates a Graph Neural Network (GCN), Gated Recurrent Unit (GRU) and Deep Deterministic Policy Gradient (DDPG) algorithms for precise prediction of the stress–strain state in intricate mechanical structures such as cantilever beams. The framework conducts comprehensive modeling of cross-sectional shapes, material characteristics and loading conditions through GCN, captures temporal dependencies via GRU, and optimizes stress–strain predictions within continuous action spaces using DDPG, thereby enhancing the accuracy and robustness of predictions. In the experimental evaluation, the DGUNet framework successfully captures the intricate interrelationships and stress evolution patterns within the structures, demonstrating superior performance in terms of Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and other metrics compared to traditional stress–strain analysis methods. The results obtained from both public datasets and finite element analysis (FEA) datasets substantiate that DGUNet significantly elevates the accuracy and efficiency of mechanical structure analysis, offering novel technical approaches and strategies for the design optimization of complex engineering structures, and providing robust methodological support for achieving more efficient structural optimization and safety assessments in engineering design.