Inference Techniques for Diagnosis Based on Set Operations

Abstract

*† State identification is a crucial function in an autonomous system. The result of state identification is the basis for fault diagnosis and autonomous planning in an autonomous agent. NASA has developed Livingstone to perform state tracking, which is the kernel in their remote autonomous agent. The Remote Agent has been successfully tested in Deep Space One spacecraft. The theory behind Livingstone is based on General Diagnostic Engine (GDE) which aims to detect all possible states. On the other hand, Livingstone only tracks several most possible states. Based on this observation, we believe that GDE is computationally too intensive for state identification in spacecraft subsystem diagnosis. We have also observed that the number of sensors is relatively smaller than the number of components in a system. This motivated us to develop a symptom driven state-tracking algorithm which reduces the memory requirement and increases the execution speed. In this paper, we analyze simple examples and summarize characteristics of an autonomous system which can help simplify the diagnosis. Given a structural and behavioral model of a system, we can use techniques from set theory to diagnose faults. The diagnosis process is triggered by a discrepancy between the observation and the model prediction. Our method avoids the exponential computational problem faced by the traditional approaches such as GDE. Probability theory is used to support our approach and a command-based probing (configuration) technique is also discussed. A diagnosis system is implemented based on our technique and applied to a simple spacecraft propulsion subsystem diagnosis.