EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 2.5

Robustness Analysis of Conjugate Gradient Algorithms in Neural Network Based Aircraft Parameter Estimation under Measurement Uncertainty

Challa Parvathi Rudesh, Sanjay Singh, P. Srinivas Rao, Yash A. Gujar

Defence Science Journal · 27.10.2026 · 76 (6) · 797-803

DOI: 10.14429/dsj.21727 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Defence Science Journal, online erschienen am 27.10.2026. Verfasst von Challa Parvathi Rudesh und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 2.5).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Estimation of stability and control derivatives is critical for flight dynamics analysis and control law design. Artificial Neural Networks (ANNs) offer a strong model-free framework for system identification; handling sensor noise remains a major hurdle. In this focused, preliminary theoretical investigation, we analyse the robustness of a Feed-Forward Neural Network (FFNN) using the Delta Method to estimate the longitudinal derivatives of the Hansa-3 aircraft. This study compares the performance of three Conjugate Gradient training algorithms Fletcher-Reeves (CGF), Polak-Ribiére (CGP), and Powell-Beale (CGB) under realistic measurement uncertainty. Synthetic flight data is generated using a 4th order Runge-Kutta simulation and corrupted with 5 % Gaussian white noise using the Equation Error formulation to ensure physical consistency. Results indicate that while all algorithms achieve high accuracy ( > 0.99) under ideal conditions, their performance splits significantly when noise is added. The Powell-Beale (CGB) algorithm demonstrated superior performance, especially for the pitch control derivative achieving an estimation error less than 9 %, effectively halving the error rate compared to the 19.0 % yielded by CGF. Additionally, CGB reduced estimation variance by nearly three times for control parameters, proving that its restart strategy effectively stabilizes the gradient search on noisy error surfaces. These findings establish CGB as a highly effective training strategy among the tested algorithms for neural network-based parameter estimation under 5 % noise conditions.

Zitieren

APA

Rudesh, C. P., Singh, S., Rao, P. S., & Gujar, Y. A. (2026). Robustness Analysis of Conjugate Gradient Algorithms in Neural Network Based Aircraft Parameter Estimation under Measurement Uncertainty. Defence Science Journal, 76(6), 797-803. https://doi.org/10.14429/dsj.21727

Vancouver

Rudesh CP, Singh S, Rao PS, Gujar YA. Robustness Analysis of Conjugate Gradient Algorithms in Neural Network Based Aircraft Parameter Estimation under Measurement Uncertainty. Defence Science Journal. 2026;76(6):797-803. doi:10.14429/dsj.21727

BibTeX
@article{Rudesh2026,
  author = {Rudesh, Challa Parvathi and Singh, Sanjay and Rao, P. Srinivas and Gujar, Yash A.},
  title = {Robustness Analysis of Conjugate Gradient Algorithms in Neural Network Based Aircraft Parameter Estimation under Measurement Uncertainty},
  journal = {Defence Science Journal},
  year = {2026},
  volume = {76},
  number = {6},
  pages = {797--803},
  doi = {10.14429/dsj.21727}
}
Kostenlos starten →