An Agent-Based Architecture for Model-Based Diagnosis Using Observation Cost

Over the past decades, advanced techniques for efficient diagnosis of malfunctioning systems have been developed. Agents equipped with such techniques that provide support in the diagnosis process have resulted from these diagnosis techniques. In general, diagnosis techniques focus on the reasoning part of the diagnostic process as well as improving the efficiency thereof. In some situations however, not the reasoning part, but the observation part of the diagnostic process is the main bottleneck and consequently requires attention. These situations arise when the costs of observing during the diagnostic process are not negligible in comparison to the costs of reasoning. This paper presents such an approach by extending an existing agent-based architecture for diagnosis. Several algorithms are proposed within the agent, tested and compared with respect to the total observation cost required to identify the root cause of a problem. The tests are performed in several situations with varying circumstances showing promising results.