An Expert Systems Approach to Automated Fault Diagnostics

Future long duration manned space missions will require regenerative Life support subsystems which operate efficiently and reliably with minimum attendance from the crew. An essential prerequisite to achieving this goal is the development of highly reliable controllers to control the various life support processes, to monitor subsystem performance, and to provide automatic shutdown of a process when out of tolerance conditions persist. While life support controller designs have improved dramatically over the years, it is now no Longer good enough to continue a philosophy where the only result of the fault monitoring process is to provide an automatic shutdown of the subsystem. Indeed, if the continuous operating time (as measured by mean-time-between-shutdowns) of environmental control equipment is to be actively increased, we must incorporate the developer's expert knowledge of the life support process into the controller to create automatic shutdown avoidance techniques where possible. The recent emergence of powerful artificial intelligence (AI) software development tools promises to reduce the time and effort associated with transforming an expert's knowledge into the operational software system required to implement automatic shutdown avoidance. To conduct an evaluation of the applicability of using AI techniques for life support processes, the National Aeronautics and Space Administration (NASA) developed a prototype expert system for automated fault detection, isolation, and correction of a regenerative carbon dioxide (CO₂) removal subsystem. This paper describes the development of the prototype and presents the results of that evaluation.