EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Safe Reinforcement Learning Control Method for an Aeroengine Using a Control Barrier Function Safety Filter

Maolin Zhou, Ruichao Li, Tong Wu, Jionghao Liu, Hongtai Pang, Yingqing Guo, Chenyang Lai

Aerospace · 14.11.2026 · 13 (11) · 1067

DOI: 10.3390/aerospace13111067 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Aerospace, online erschienen am 14.11.2026. Verfasst von Maolin Zhou und 6 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Sichuan Provincial Natural Science Foundation.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

To address the conflict between fast response and strict safety constraints in aeroengine transient control, this paper proposes a hierarchical safe reinforcement learning method based on a control barrier function (CBF) safety filter. Under the same step command, the proposed method and a conventional gain-scheduled PI controller with fuel schedules and Min-Max limit selection achieve 5% settling times of 2.44 s and 2.80 s, respectively; neither violates the prescribed constraints during testing. By contrast, the soft-constraint method achieves a settling time of 2.70 s and produces maximum low-pressure and high-pressure compressor surge-margin violations of 1.34% and 7.24%, respectively. Under a wide-range amplitude-modulated pseudo-random binary sequence (APRBS) command, the proposed architecture tracks each reference change within 5 s with zero steady-state error, while removing the PI controller results in a maximum steady-state error of 1.84%. In a 500-step desktop timing test, mean and maximum controller execution times were 2.431 and 6.676 ms, respectively; no step exceeded the 20 ms sampling interval. Sensitivity analysis shows limited variation in control performance over the tested CBF-parameter and reward-weight ranges. These findings show that tracking optimization and constraint enforcement can be coordinated within the proposed hierarchical architecture under the tested conditions.

Förderung

  • National Natural Science Foundation of China
  • Sichuan Provincial Natural Science Foundation

Zitieren

APA

Zhou, M., Li, R., Wu, T., Liu, J., Pang, H., Guo, Y., & Lai, C. (2026). Safe Reinforcement Learning Control Method for an Aeroengine Using a Control Barrier Function Safety Filter. Aerospace, 13(11), 1067. https://doi.org/10.3390/aerospace13111067

Vancouver

Zhou M, Li R, Wu T, Liu J, Pang H, Guo Y, et al. Safe Reinforcement Learning Control Method for an Aeroengine Using a Control Barrier Function Safety Filter. Aerospace. 2026;13(11):1067. doi:10.3390/aerospace13111067

BibTeX
@article{Zhou2026,
  author = {Zhou, Maolin and Li, Ruichao and Wu, Tong and Liu, Jionghao and Pang, Hongtai and Guo, Yingqing and Lai, Chenyang},
  title = {Safe Reinforcement Learning Control Method for an Aeroengine Using a Control Barrier Function Safety Filter},
  journal = {Aerospace},
  year = {2026},
  volume = {13},
  number = {11},
  pages = {1067},
  doi = {10.3390/aerospace13111067}
}
Kostenlos starten →