Power-based self-tuning model for aero-engine using multilayer perceptron-assisted inverse control with filtered error feedback

Analytical redundancy is of critical importance in improving the fault tolerance capability of engine control systems. The component-level model (CLM) is widely employed to provide analytical redundancy for gas path sensors. However, the prediction accuracy of the CLM is inevitably influenced by measurement uncertainties in fuel flow and nozzle area. To address this limitation, this paper proposes a power-based self-tuning model (PSTM) to improve the gas path prediction accuracy of a dual-shaft engine. By employing multilayer perceptron (MLP)-assisted inverse control with filtered error feedback, the PSTM dynamically adjusts CLM inputs based on deviations in power representative parameters between the model and the physical engine. The power representative parameters include the low-pressure shaft speed and the engine pressure ratio. The proposed method is validated with simulation data covering a wide range of operating conditions, as well as with actual ground test data. In the envelope points test, the average mean absolute error (MAE) and root mean square error (RMSE) of gas path predictions decrease from 0.916 % to 0.028 % and from 1.113 % to 0.039 %, respectively. In the flight profile test, the average MAE and RMSE of gas path predictions are reduced from 0.703 % to 0.069 % and from 0.914 % to 0.275 %, respectively. In the actual ground data test, compared with the CLM, the average MAE and RMSE of gas path predictions are reduced from 1.642 % to 0.920 % and from 2.072 % to 1.233 %, respectively. The verification results indicate that adjusting CLM inputs according to deviations in power representative parameters effectively reduces the impact of measurement uncertainties on model predictions. The proposed method achieves higher modeling accuracy compared with the conventional CLM, thereby enhancing the reliability of analytical redundancy for gas path sensors.