Hub moment limit protection using neural network prediction

Nilesh A. Sahani, Joseph F. Horn, Geoffrey J.J. Jeram, J. V.R. Prasad

Research output: Contribution to journalConference articlepeer-review

8 Scopus citations


A main rotor hub moment limit protection system is proposed and evaluated in real-time piloted simulation. The system is designed to allow aggressive maneuvering while reducing the likelihood of "worst case" hub loads. Such a system might increase component life and allow less conservative structural design on future rotorcraft. The system has three major components: limit parameter response prediction, control margin calculation and pilot cueing. A method for predicting the future response of transient limit parameters using measured aircraft states has been suggested. The algorithm uses non-linear time response functions that are represented with neural networks. Stick constraints were calculated using the predicted response and the peak value of the limit parameter to a unit step control input. The stick constraints were conveyed to the pilot using softstop cues. The system was evaluated in simulation using two types of aggressive maneuvers. Results showed that the prediction algorithms were effective, as the system reduced the frequency and severity of hub moment limit violations without significantly affecting the achievable agility of the aircraft. Formal handling qualities evaluation was not conducted, but pilot comments indicated some objectionable characteristics in the soft stop cueing. Further research on new methods of tactile cueing is warranted.

Original languageEnglish (US)
Pages (from-to)775-786
Number of pages12
JournalAnnual Forum Proceedings - American Helicopter Society
StatePublished - 2004
Event60th Annual Forum Proceedings - American Helicopter Society - Baltimore, MD, United States
Duration: Jun 7 2004Jun 10 2004

All Science Journal Classification (ASJC) codes

  • Transportation
  • Aerospace Engineering


Dive into the research topics of 'Hub moment limit protection using neural network prediction'. Together they form a unique fingerprint.

Cite this