Abstract: Aeroengine blade flutter poses a critical safety hazard induced by aeroelastic instability, with traditional prediction methodologies requiring extensive unsteady flow field simulations that entail substantial computational time. This paper proposes a method for calculating aerodynamic loads on aero-engine blades based on modal shape decomposition combined with a Gated Recurrent Unit (GRU) neural network model, and applies it to rapid estimation of aerodynamic damping and flutter analysis of aero-engine blades. We first perform bending-torsion decomposition on the blade's natural vibration modes and calculate corresponding aerodynamic modal forces for both bending and torsion modes. Subsequently, temporal GRU neural networks are employed to establish mapping relationships between aerodynamic modal forces and generalized motion variables for both bending and torsion modes. These established mappings are then applied to compute aerodynamic damping and predict flutter for blades with different bending-torsion ratios within specific frequency ranges. Using the NASA Rotor67 rotor model as a case study, the proposed aerodynamic modal force model for bending-torsion modes is validated, and flutter analysis is implemented for blades with varying bending-torsion ratio modes. Results demonstrate that the model can accurately estimate unsteady aerodynamic loads on aero-engine blades within certain frequency ranges and enable efficient aerodynamic damping estimation for blades with different bending-torsion ratio vibration modes. The method developed in this study significantly accelerates the flutter design process of aero-engine blades.