State estimation and system model correction of aero-engines under multi-source uncertainty: A hierarchical variational inference approach

Accurate state estimation is critical for performance optimization and reliability enhancement in modern turbine systems. Although traditional filtering methods have demonstrated strong performance in various applications, their effectiveness is limited in the presence of component performance dispersion, high-dimensional system dynamics, performance degradation and uncertain control inputs. This study proposes a variational inference-based state estimation framework for aero-engine systems to address challenges arising from multi-source uncertainty. Under the assumption of a known state-space model, a loss function based on the stochastic variational lower bound is constructed to enable joint optimization of state variables and model parameters. This allows for precise inference of component health states and reliable identification of fault-related features. In cases where the aero-engine system model is partially or completely unknown, a hierarchical variational framework is further introduced, incorporating stochastic differential equations to simultaneously infer system states and uncover underlying control dynamics. Simulation results demonstrate that the proposed method consistently outperforms traditional filtering algorithms under varying noise levels and model uncertainties. It effectively distinguishes between modeling errors and actual performance deviations of engine components, leading to improved diagnostic accuracy and robustness.