TY - GEN
T1 - Unmanned systems health analysis through evidential reasoning networks
AU - Dunham, Joel
AU - Johnson, Eric
AU - Feron, Eric
AU - German, Brian
N1 - Generated from Scopus record by KAUST IRTS on 2021-02-18
PY - 2020/10/11
Y1 - 2020/10/11
N2 - Evidential reasoning, developed by Glenn Shafer and Arthur Dempster in the 1960s and 1970s, has been extensively applied to risk analysis, sensor fusion, and system failure analysis. Use in real-time systems for health analysis has been more limited due to computational complexity requiring less either comprehensive analysis or significant computing power. Evidential reasoning networks, also known as valuation networks, reduce computational requirements by eliminating hypothesis combinations which are infeasible. Recent extensions enable these networks to learn the relationships between nodes based on evidence inputs, enabling these networks to adapt without the need of subject matter experts defining each relationship update. This system is applied to unmanned system health analysis, demonstrating the capability to run complex belief analyses in real-time on autopilot systems with low computational power using the GUST autopilot system developed at the Georgia Institute of Technology. Comparisons are made between the evidential combination approach and more traditional contingency management that uses time delays and worst-case scenario assumptions for contingency responses. Simulation training is used as a surrogate for high volumes of flight testing, and operational results are primarily tested through GUST simulation of a representative mission. Results show that evidential reasoning networks are an effective approach to real-time health analysis of unmanned systems, using the novel update rules to understand relationships based on operational outcomes. Flight demonstration is included to show the capability to run this system in real operations. This work has implications on integration of unmanned systems into the national airspace as well as on Urban Air Mobility. Results from the network are explainable, enabling human oversight of operational decisions. Real-time implementation enables integration into avionics systems. Further, the data-driven approach to learning relationships enables this system to adapt as information concerning unmanned systems steadily changes.
AB - Evidential reasoning, developed by Glenn Shafer and Arthur Dempster in the 1960s and 1970s, has been extensively applied to risk analysis, sensor fusion, and system failure analysis. Use in real-time systems for health analysis has been more limited due to computational complexity requiring less either comprehensive analysis or significant computing power. Evidential reasoning networks, also known as valuation networks, reduce computational requirements by eliminating hypothesis combinations which are infeasible. Recent extensions enable these networks to learn the relationships between nodes based on evidence inputs, enabling these networks to adapt without the need of subject matter experts defining each relationship update. This system is applied to unmanned system health analysis, demonstrating the capability to run complex belief analyses in real-time on autopilot systems with low computational power using the GUST autopilot system developed at the Georgia Institute of Technology. Comparisons are made between the evidential combination approach and more traditional contingency management that uses time delays and worst-case scenario assumptions for contingency responses. Simulation training is used as a surrogate for high volumes of flight testing, and operational results are primarily tested through GUST simulation of a representative mission. Results show that evidential reasoning networks are an effective approach to real-time health analysis of unmanned systems, using the novel update rules to understand relationships based on operational outcomes. Flight demonstration is included to show the capability to run this system in real operations. This work has implications on integration of unmanned systems into the national airspace as well as on Urban Air Mobility. Results from the network are explainable, enabling human oversight of operational decisions. Real-time implementation enables integration into avionics systems. Further, the data-driven approach to learning relationships enables this system to adapt as information concerning unmanned systems steadily changes.
UR - https://ieeexplore.ieee.org/document/9256593/
UR - http://www.scopus.com/inward/record.url?scp=85097974721&partnerID=8YFLogxK
U2 - 10.1109/DASC50938.2020.9256593
DO - 10.1109/DASC50938.2020.9256593
M3 - Conference contribution
SN - 9781728198255
BT - AIAA/IEEE Digital Avionics Systems Conference - Proceedings
PB - Institute of Electrical and Electronics Engineers Inc.
ER -