Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


41783.pdf (3.49 MB)

ETD Abstract Container

Abstract Header

Run Time Assurance for Intelligent Aerospace Control Systems

Dunlap, Kyle
http://orcid.org/0000-0002-5869-1380
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1649768916078616

Abstract Details

2022, MS, University of Cincinnati, Engineering and Applied Science: Aerospace Engineering.
Safety is a critical component for aerospace systems, where a mistake or fault on board an aircraft or spacecraft could result in hardware damage, mission failure, or even the loss of human lives. In the absence of scalable advanced control and system verification methods, one approach is to monitor the behavior of control techniques at run time and intervene to assure safety of the system. Run Time Assurance (RTA) systems are online safety assurance techniques that filter the output of a primary controller to actively assure safety of the system. RTA can be used in safety-critical control applications where a performance driven primary controller may cause the system to violate safety constraints. RTA is designed to be completely independent of the primary controller, and therefore can be applied to any aerospace control system. This research evaluates four categories of RTA approaches for application to different aerospace control systems. Each RTA approach is classified based on its membership to explicit or implicit monitoring and switching or optimization based intervention. First, to show the feasibility and compare performance, an unconstrained linear quadratic regulator is used as the performance driven primary controller for spacecraft docking, while all four RTA approaches are demonstrated to adhere to velocity limit safety constraints. Second, all four RTA approaches are evaluated in a fixed-wing aircraft formation flight scenario with safety constraints on position and velocity. Third, this scenario is expanded to use quadrotors instead of fixed-wing aircraft to evaluate the performance of RTA on a smaller scale, where it can easily be implemented and tested in the real world. Fourth, all four RTA approaches are applied to assure safety during reinforcement learning training in a simplified spacecraft docking scenario. The impact of each RTA on the training time and ultimate performance of the trained controller are compared to reward-shaping approaches that do not feature RTA but penalize unsafe states. The simplest RTA is then used to train controllers with reinforcement learning and genetic fuzzy systems for spacecraft docking in order to compare and evaluate the performance and explainability of each approach. Finally, different time optimal control methods are compared for their use as primary controllers, where RTA can still be used to assure safety.
Kelly Cohen, Ph.D. (Committee Member)
Timothy J. Arnett, Ph.D. (Committee Member)
Anoop Sathyan, PhD (Committee Member)
Ou Ma, Ph.D. (Committee Member)
Kerianne Hobbs, Ph.D. (Committee Member)
112 p.
Aerospace Materials
Run Time Assurance; Safety Critical Control; Reinforcement Learning; Genetic Fuzzy Systems

Recommended Citations

Citations

  • Dunlap, K. (2022). Run Time Assurance for Intelligent Aerospace Control Systems [Master's thesis, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1649768916078616

    APA Style (7th edition)

  • Dunlap, Kyle. Run Time Assurance for Intelligent Aerospace Control Systems. 2022. University of Cincinnati, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1649768916078616.

    MLA Style (8th edition)

  • Dunlap, Kyle. "Run Time Assurance for Intelligent Aerospace Control Systems." Master's thesis, University of Cincinnati, 2022. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1649768916078616

    Chicago Manual of Style (17th edition)

ucin1649768916078616
296
© 2022, all rights reserved.
This open access ETD is published by University of Cincinnati and OhioLINK.
Release 3.2.12