An improved nonlinear onboard adaptive model for aero-engine performance control

The onboard adaptive model can achieve the online real-time estimation of performance parameters that are difficult to measure in a real aero-engine, which is the key to realizing model-based performance control. It must possess satisfactory numerical stability and estimation accuracy. However, the positive definiteness of the state covariance matrix may be destroyed in filter estimation because of the existence of some uncertain factors, such as the accumulated measurement error, noise, and disturbance in the strongly nonlinear engine system, inevitably causing divergence of estimates of Cholesky decomposition-based Spherical Unscented Kalman Filter (SUKF). Therefore, this paper proposes an improved SUKF algorithm (iSUKF) and applies it to the performance degradation estimation of the engine. Compared to SUKF, the iSUKF mainly replaces the Cholesky decomposition with the Singular Value Decomposition (SVD), which is numerically stable without any strict requirement for the state covariance matrix. Meanwhile, a correction factor is designed to assess the measurement deviation between the real engine and the nonlinear onboard model to correct the state covariance matrix, thus maintaining better numerical stability of parameters estimated by the filter. Then, an offline correction strategy is also proposed to eliminate the influence of the degradation of unestimated health parameters or the filter’s inadequate estimation of the coupled health parameters. This action effectively promotes the onboard adaptive model’s estimation accuracy concerning the degradation of the engine’ real health parameters and its performance parameters. Finally, the simulation results show that the iSUKF can maintain the numerical stability of the filter’s estimation of health parameters. Compared with the existing methods, the offline correction strategy improves the estimation accuracy of the iSUKF-based nonlinear onboard adaptive model for the performance parameters of the real engine by more than 50%. The proposed method will provide feasible technical support for model-based aero-engine performance control.