TRACLabs is working in conjunction with Vanderbilt University on a NASA-funded project to diagnose faults in complex systems. Space systems such as habitats, vehicles and rovers have multiple, interconnected subsystems each of which has its own set of telemetry and commands. Faults in any subsystem may manifest in a number of different ways, and for lack of sufficient sensors be difficult to isolate. Even if faults are isolated, the remote location of the spacecraft or rover may make it hard to repair. Recovery from faults often involves a lengthy process of sending commands, verifying their success, and then re-evaluating system performance. Given the insufficiency of sensors, isolating faults may require issuing commands to reconfigure the system or subsystem to obtain more information or to rule out certain hypotheses. is focused on developing an active diagnosis system that can iteratively evaluate diagnosis hypotheses to arrive at a definitive diagnosis and then recommend recovery actions. The research uses a model-based diagnosis system combined with an adjustably autonomous procedure execution system in a novel architecture that fits into NASA’s concept of operations for space missions. The architecture includes a model-based diagnosis component that compares system sensors with values generated from a model. It also includes a procedure executive that operates on a clearly specified linear set of commands and telemetry checks to move the system from one configuration to another. Between these two components is a functional reasoner that maps diagnosis hypotheses to procedures that can generate more information. Another component is a cost/benefit analyzer that ranks procedures based on their suitability for the current situation. Together, these components can reason about different diagnoses and take action on the system to overcome uncertainty and recover from faults. The figure below shows the complete system. Additional components such as a predictor (for determining future system state), a decision-maker (for altering system activities) and a replanner (for determining new system goals) are also being integrated.

An architecture for active diagnosis