Efficient reliability-based design optimization for compressor root and groove structure

To prolong the low-cycle fatigue (LCF) life and improve the reliability of gas turbines, reliability-based design optimization (RBDO) of compressor root–groove (RG) structures is of great significance. In this study, a hierarchical surrogate-assisted framework is developed for the LCF reliability assessment and RBDO of compressor RG structures. First, a graph convolutional network (GCN) based physical-field reconstruction model is established to predict the stress and strain fields of the RG structure, providing an efficient alternative to repeated finite element analyses. Second, a reliability assessment method is constructed by combining the global limit sampling (GLS) active learning strategy with a radial basis function (RBF) surrogate model. The GLS adaptively identifies informative samples near the limit-state boundary and enables efficient reliability assessment of a numerical example and the RG structure. Finally, the GLS-RBF model is embedded into a decoupled RBDO procedure with an adaptive most probable target point (MPTP) search strategy to improve convergence stability. The proposed framework is validated through benchmark examples and the compressor RG problem by comparison with existing methods. For the compressor RG structure, the failure probability is reduced while the LCF life increases. The results demonstrate that the framework can effectively improve LCF reliability and optimization efficiency for complex engineering problems.