Health-aware decision making under uncertainty for complex systems
Abstract/Contents
- Abstract
- Aerospace vehicles, particularly spacecraft, often operate in harsh and uncertain environments, where decisions critical to mission success may need to be made quickly and with incomplete information. This is especially true when the vehicle experiences component faults or failures. The field of system health management has evolved over the last several decades from simple automated alarms to sophisticated artificial intelligence algorithms designed to analyze such off-nominal conditions and generate appropriate responses. This evolution took place largely apart from the development of automated system control, planning, and scheduling (referred to collectively as decision making). While there have been efforts to establish information exchange between system health management and decision making, successful practical implementations of integrated architectures have remained rare. This thesis consists of three major parts. In the first part, the limitations of the currently prevalent system health management methodology are described and illustrated through numerical examples. In particular, prognostics (a relatively recent addition to the field of system health management) is shown to be meaningful only in a narrow subset of applications and, even then, challenging to implement in an effective manner. Instead, an approach is proposed that unifies system health management and operational decision making in their formulations in order to overcome these shortcomings. The thesis discusses implementation details of the new approach -- referred to as Health Aware Decision Making, HADM -- and provides an analysis of its computational complexity. One of the key ingredients for successfully implementing HADM for realistic systems operating in harsh and uncertain environments is availability of decision making algorithms that can reason over large, continuously valued action and observation domains. The second part of the thesis describes an algorithm developed for this category of problems, Large Problem Decision Making (LPDM). The algorithm is based on Determinized Sparse Partially Observable Trees (DESPOT), a state-of-the-art solver for problems formulated as partially observable Markov decision processes (POMDPs). LPDM incorporates novel methods for handling complex model spaces and is shown to outperform both the original DESPOT and a version of DESPOT augmented with the Blind Value algorithm (a recent method of handling large, continuously valued action spaces) on benchmarking problems. The third major part of the thesis applies the methodology and the algorithms developed in the first two parts to create an advanced decision support system for space missions: System Health Enabled Realtime Planning Advisor (SHERPA). SHERPA is designed to be model-based, modular, and adaptable to different use cases throughout the lifetime of a mission. The system is targeted for first use on a NASA robotic rover mission to the Moon, scheduled for launch in 2023. The mission, Volatiles Investigating Polar Exploration Rover (VIPER), intends to land the solar-powered rover in a lunar polar region and use it to characterize the distribution of water ice and other volatiles in preparation for establishing a permanent human base. The thesis describes in detail one SHERPA use case developed for the VIPER mission. In the use case, focused on rover traverse evaluation and refinement, a traverse template is provided to SHERPA specifying the science activities to be performed at an ordered set of waypoints. SHERPA uses mission simulations with optimized action selection to evaluate the robustness of the proposed template to uncertainties that are likely to be a factor during the mission, then recommends a schedule of battery recharge periods that maximizes the chances of a successful traverse. Another use case, currently under development, generates a full traverse for the VIPER rover taking only the high-level mission objectives and constraints as inputs. The latter use case will also form the foundation for SHERPA's landing site selection and vehicle parameter optimization capabilities
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2020; ©2020 |
| Publication date | 2020; 2020 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Balaban, Edward |
|---|---|
| Degree supervisor | Kochenderfer, Mykel J, 1980- |
| Thesis advisor | Kochenderfer, Mykel J, 1980- |
| Thesis advisor | Alonso, Juan José, 1968- |
| Thesis advisor | Schwager, Mac |
| Degree committee member | Alonso, Juan José, 1968- |
| Degree committee member | Schwager, Mac |
| Associated with | Stanford University, Department of Aeronautics and Astronautics. |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Edward Balaban |
|---|---|
| Note | Submitted to the Department of Aeronautics and Astronautics |
| Thesis | Thesis Ph.D. Stanford University 2020 |
| Location | electronic resource |
Access conditions
- Copyright
- © 2020 by Edward Balaban
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...