Mitigating challenges in Complex Space Systems
Presentation Type
PowerPoint only (no paper)
Location
Henderson Welcome Center
Start Date
17-11-2016 2:00 PM
Abstract
Space systems are important to modern industry, providing critical information for navigation, communication, weather tracking, and imagery. Systems are typically expensive and meticulously engineered. Nevertheless, spacecraft can still exhibit a range of degradation and failures. Diagnosing those failures are important, as every minute of satellite down-time results in millions of dollars in lost revenue to customers here on Earth, or potentially loss of life for travellers in space.
Fault Detection for space systems is challenging not just because of the complexity of the machines but because of dynamics in the space environment as well. Operational conditions and tempos also add to the difficulty and can cause state changes in the system. Traditional methods of fault detection use expert systems which are rules based, requiring a great deal of domain knowledge and are not general solutions-- every new scenario needs a new rule relating causes to effects. The number of relationships are superexponential to the number of variables in the system, which means rules based systems become intractable for automation.
This presentation discusses the Saber Astronautics method for modelling highly complex systems using machine learning tool called a "System Map". In particular, the goal is to find data-driven relationships between space system metrics and space environment metrics then use these relationships in models for intelligent control. Two case studies are shown. First, telemetry from the NASA Advanced Composition Explorer (ACE) satellite is observed and modeled to show time invariant general solutions can be produced with the method. Second, a human-robotic interaction study shows how System Maps portable into Mars and other Planetary Exploration analogues.
The benefits of this methodology can greatly improve efficiency, safety, and streamline the way we live and work in space.
Biographies
Mitigating challenges in Complex Space Systems
Henderson Welcome Center
Space systems are important to modern industry, providing critical information for navigation, communication, weather tracking, and imagery. Systems are typically expensive and meticulously engineered. Nevertheless, spacecraft can still exhibit a range of degradation and failures. Diagnosing those failures are important, as every minute of satellite down-time results in millions of dollars in lost revenue to customers here on Earth, or potentially loss of life for travellers in space.
Fault Detection for space systems is challenging not just because of the complexity of the machines but because of dynamics in the space environment as well. Operational conditions and tempos also add to the difficulty and can cause state changes in the system. Traditional methods of fault detection use expert systems which are rules based, requiring a great deal of domain knowledge and are not general solutions-- every new scenario needs a new rule relating causes to effects. The number of relationships are superexponential to the number of variables in the system, which means rules based systems become intractable for automation.
This presentation discusses the Saber Astronautics method for modelling highly complex systems using machine learning tool called a "System Map". In particular, the goal is to find data-driven relationships between space system metrics and space environment metrics then use these relationships in models for intelligent control. Two case studies are shown. First, telemetry from the NASA Advanced Composition Explorer (ACE) satellite is observed and modeled to show time invariant general solutions can be produced with the method. Second, a human-robotic interaction study shows how System Maps portable into Mars and other Planetary Exploration analogues.
The benefits of this methodology can greatly improve efficiency, safety, and streamline the way we live and work in space.
Comments
Visit the Research Panel: Orbital