Knowledge-based management of failures in autonomous systems

This paper presents the design of a fault management system (FMS) for an unmanned and untethered platform. The system must automatically detect, diagnose, localize and reconfigure the system to cope with failures. Traditional fault tolerant approaches used in telephone switching, manned and unmanned satellites, commercial banking, airline reservations, air traffic control, and others are reviewed. Expert system´s technology is used to extend these traditional approaches to achieve a highly reliable design capable of sustaining operation over many months with little or no communication. An existing simulator has been modified to allow fault injection and to model fault propagation. This provides a testbed for evaluating system candidates. A specific fault management hardware and software architecture has been selected. Expert system diagnostic rules, which run on the fault tolerant base, are discussed. Diagnostic rule performance in detecting, localizing, and recovering from Autonomous Systems (AS) sensor, actuator, and computer subsystem failures during AS operation is analyzed.