Artificial Intelligence for Monitoring and Diagnosis of Robotic Spacecraft

Sammanfattning: In this thesis the application of artificial intelligence to monitoring and diagnosis of robotic spacecraft is discussed. Several software prototype systems were developed to serve as testbeds for the research and to evaluate the effectiveness of the approach against real problems and current techniques used in NASA's planetary exploration program.

Software prototypes were used to investigate the verification of robot plan execution. New artificial intelligence algorithms for monitoring and diagnosis of robot systems were designed, programmed, and tested. These included plan analysis for monitoring, sensor planning, generation of expected sensor values, and diagnosis of execution failures caused by hardware, environmental or plan anomalies. Testing was performed on a laboratory telerobotic hardware testbed for satellite servicing and on a mobile planetary rover robot operating in natural terrain.

Artificial intelligence algorithms, software prototypes, and more advanced, operationally capable systems for monitoring ground support systems and actual spacecraft in flight were designed, programmed, and tested. A ground support system that served as one test domain was the mirror cooling circuit of the 25-foot Space Simulator at the Jet Propulsion Laboratory (JPL) in Pasadena, California. A prototype monitoring system for this device based on a theory of "predictive monitoring" was developed and tested. Mission operations for the Voyager II spacecraft served as another test domain for an intelligent spacecraft health-monitoring and diagnosis system. This system was successfully tested in support of telecommunications operations during Voyager II's encounter with the planet Neptune in 1989. This was the one of the first artificial intelligence systems to be used in planetary spacecraft operations at NASA/JPL. Subsequently, this system was adapted and tested in support of operations of the Magellan spacecraft telecommunications subsystem and the Galileo spacecraft power and pyro subsystem.

Some of the specific artificial intelligence algorithms that were developed for monitoring and diagnosis included the use of heuristic and causal model-based reasoning techniques for predictive generation of sensor values, sensor selection planning, dynamic alarm limit checking, hierarchical procedure specialists for fault diagnosis, and integration of Al with conventional systems in full-scale monitoring and diagnosis applications.

In support of this overall program of research, novel software engineering tools for artificial intelligence research and application development were also developed and will be discussed in the thesis.

The application of artificial intelligence techniques to the monitoring and diagnosis of robotic space systems was shown to be very effective with specific benefits in the areas of systems autonomy, spacecraft safety, ground operations productivity and automation. As a result of this work in part, artificial intelligence is now considered by senior mission designers to be an enabling technology for on-board automation of planetary rovers and for automation in mission operations at the Jet Propulsion Laboratory.