Abstract: To ensure the reliability of gas turbine compressors under multi-source uncertainties, a composite framework for the reliability-based design optimization (RBDO) of the compressor blade disc is established. For the problem with multi-source uncertainties, a method that integrates a two-stage active learning Kriging model and the decoupled approach is proposed. The proposed method adaptively selects points to construct the surrogate model during the optimization process and performs RBDO in two stages. The proposed method improves the efficiency of RBDO while maintaining accuracy. Meanwhile, an active learning strategy based on improved constrained boundary double-point sampling is proposed. The computational cost is further reduced through the design-driven local approximation strategy and the double-point enrichment method. The superiority of the method is verified by a numerical example of a short column. The RBDO is carried out for a compressor blade disc while considering the low cycle fatigue. The results indicate that the proposed method significantly improves the average low cycle fatigue life and the reliability of the model.