An On-Board Adaptive Modeling Approach Based on MIRLS-MA Extended Kalman Filter for Aeroengines

• Published in: IEEE transactions on aerospace and electronic systems 2024-12, Vol.60 (6), p.8855-8870
• Main Authors: Liu, Zhidan, Han, Xiaobao, Huang, Yingzhi, Gou, Linfeng, Fan, Ding
• Format: Article
• Language: English
• Subjects: Accuracy; Adaptation models; Computational modeling; Estimation; Kalman filters; Noise; Noise measurement
• ISSN: 0018-9251; 1557-9603
• DOI: 10.1109/TAES.2024.3436602

Aeroengine control applications demand on-board adaptive models with exceptional real-time performance and accuracy for precise online parameter prediction. Sensor failures, however, compromise the reliability and estimation accuracy of such models. In this article, we propose a hybrid enhanced on-board adaptive modeling approach for aeroengines, combining modified iterative reweighted least squares (MIRLS) and mixed adaptive extended Kalman filtering (MAEKF). The MAEKF algorithm outperforms conventional extended Kalman filtering by providing optimal estimates without requiring complete process and measurement noise covariance matrices, thus improving efficiency and accuracy. The MIRLS algorithm surpasses traditional iterative reweighted least squares by effectively identifying anomalous measurements. The hybrid enhanced structure integrates component-level models with adaptive linear parameter-varying models, ensuring both accuracy and real-time performance. Finally, digital simulation and hardware-in-the-loop experiments were carried out. Experimental results reveal the proposed model's high prediction accuracy and reliability. The on-board adaptive model achieves an estimation error of 0.0169% for engine parameters and component degradation levels, a significant improvement compared to existing models. The MIRLS algorithm can effectively restrain the influence of non-Gaussian noise. The MAEKF algorithm with a forgetting factor can provide more accurate estimates when the system dynamically changes. Compared with the MAEKF algorithm without a forgetting factor, the root-mean-square error of the estimation error is reduced by 42%. Furthermore, the model swiftly detects and reconstructs accurate sensor values in case of sensor failures, maintaining precise engine output parameter predictions. In conclusion, our hybrid enhanced on-board adaptive modeling approach demonstrates remarkable reliability and prediction accuracy, making it a promising solution for aeroengine control applications.