Modular Artificial Intelligence for Faults: Local Online Watch and Efficient Response

MAIFLOWER (Modular AI for Faults: Local Online Watch and Efficient Response) leverages previous NASA investments to develop a generalized architecture for fault management that is capable of being deployed across space platforms of all kinds. The MAIFLOWER Phase II effort will develop a robust, near-operational prototype to detect and diagnose faults on Astroboticrsquo;s Vertical Solar Array Technology (VSAT), a rover that will egress from its lander, transit to a desired location near the lunar South Pole, ldquo;wigglerdquo; into the lunar soil, and deploy a 60rsquo; high solar array to generate and distribute power to other lunar systems. In Phase II, MAIFLOWER will be expanded to cover mechanical faults during additional scenarios (such as Sun tracking and unwinding) as well as faults for the external electrical power system (EPS). Astrobotic will be intimately involved in Phase II development, which it will support by providing schematics and diagrams of subsystems, real data from mechanical testing labs to validate MAIFLOWERrsquo;s algorithms, and requirements for MAIFLOWERrsquo;s integration onboard a physical VSAT artifact for actual mechanical testing. MAIFLOWER will augment previous NASA-funded MAESTRO technology by introducing transformers, a machine learning method commonly utilized on series data, to the space domain for fault detection, which will enablenbsp;MAIFLOWER to not only better diagnose faults but also be alerted to novel off-nominal conditions. MAIFLOWER will make use of a suite of AI technologies: model-based reasoning, case-based reasoning, and machine learning to detect, diagnose, and triage faults as they occur; efficient algorithms to plan courses of action (COAs) and schedule responses (built on our highly successful Aurora technology); and behavior transition networks to adaptively execute selected COAs to mitigate the effects of the fault.nbsp;