A physics-data hybrid modeling method for aeroengine component via CMA-ES-based steady-state calibration and Mamba-based dynamic compensation

Accurate component-level modeling of aeroengines remains challenging because component degradation, engine-to-engine variability, and model simplifications introduce persistent discrepancies between physics-based models and real gas-path responses. Existing hybrid approaches improve modeling accuracy but still face difficulties in calibrating strongly coupled physical parameters, capturing long-term dynamic residuals, and effectively coordinating physics- and data-driven models. To address these challenges, a hybrid aeroengine component modeling framework is proposed by combining covariance matrix adaptation evolution strategy (CMA-ES) based steady-state calibration with Mamba-based dynamic residual compensation. First, key correction factors governing component flow capacity, efficiency, pressure ratio, and temperature characteristics are introduced into the physics-based model and globally optimized using CMA-ES, thereby reducing engine-specific steady-state mismatch while preserving physical interpretability. Subsequently, residual sequences between the calibrated model outputs and measured engine responses are modeled using a Mamba selective state-space model to capture long-range temporal dependencies and compensate for unmodeled dynamic behavior. The calibrated physics-based model and residual compensation module are integrated in a cascaded architecture, enabling simultaneous preservation of physical consistency and adaptation to engine-specific dynamics. Validation using real aeroengine operating data demonstrates substantial reductions in both steady-state bias and transient prediction errors. The proposed framework also maintains consistent accuracy across different engine units and operating transitions, indicating strong adaptability and generalization capability. These results provide an effective physics-data hybrid modeling approach for high-fidelity aeroengine performance prediction and model-based health management.