High-cycle fatigue reliability prediction method for single-crystal turbine blades integrating a hybrid DMD-POD-LSTM neural network and a multi-factor synergistic model

Purpose Accurately predicting the high-cycle fatigue (HCF) reliability of single-crystal turbine blades is essential for the structural integrity and service safety of aero-engine components. However, this task remains challenging due to two main limitations in existing methods: the high computational cost required to quantify aerodynamic excitation dispersion and the absence of models capable of characterizing HCF strength dispersion under multi-factor coupling. To overcome these limitations, this study aims to develop an integrated framework with two major innovations. Design/methodology/approach First, a novel rapid aerodynamic response prediction method is proposed by combining dynamic mode decomposition, proper orthogonal decomposition and long short-term memory networks. Second, a multi-factor synergistic HCF strength model is established based on systematic tests of DD6 single-crystal specimens under varying crystal orientations, temperatures and stress ratios. This model improves the K-T diagram by integrating the EI-Haddad intrinsic crack length, critical distance theory and a Gerber-based mean stress correction. Findings This hybrid DMD-POD-LSTM model achieves high-fidelity predictions of unsteady aerodynamic loads with errors below 6.6%, while providing a computational speedup of over 11.4 times. This method makes the previously exhaustive quantification of aerodynamic excitation dispersion feasible. The multi-factor synergistic HCF strength model improves the K-T diagram by integrating the EI-Haddad intrinsic crack length, critical distance theory and a Gerber-based mean stress correction. It accurately captures the coupled effects of multiple factors, with prediction errors below 7.7%. Originality/value By integrating these two high-precision models within the reliability prediction framework, a comprehensive HCF reliability analysis is achieved that simultaneously accounts for uncertainties in aerodynamic excitation and material HCF strength. The analysis yields a probabilistic reliability of 96.82% for the blade under design conditions, offering a robust and quantitative basis for fatigue-resistant design.