Physical Field Reconstruction and Structural Optimization of Gas Turbine Hirth Couplings Based on Graph Learning Method

Hirth couplings connect gas-turbine rotor discs and transmit torque, making them critical to efficient energy conversion and power output. To address the current lack of strength research, limited optimization studies, and inefficient optimization methods for Hirth couplings, a Hirth Coupling Strength Prediction Graph Convolutional Network (HSP-GCN) is proposed for multiple mechanical field reconstruction and structural optimization. HSP-GCN can rapidly reconstruct the displacement, stress, and contact pressure fields on the Hirth tooth surfaces according to structural parameters. Its rapid and accurate performance superiority is demonstrated through comparisons with deep neural network and convolutional neural network models. To the best of our knowledge, this is the first instance that a neural network-based surrogate model has been used for field reconstruction and performance prediction of Hirth couplings. The results show that the field mean absolute errors are below 0.02 for all reconstructed fields except ux. The relative errors of the maximum von Mises stress and maximum contact pressure are generally within ±5%, with an error of −5.76% observed in one representative case near the boundary of the design space. Meanwhile, a multi-constraint optimization method for Hirth couplings based on HSP-GCN is proposed. The optimized designs reduce the maximum von Mises stress by 15.4%. This study can provide an effective and novel tool for accelerating analysis and optimization.