Autonomous agents for online diagnosis of a safety-critical system based on probabilistic causal reasoning

The goal of decentralization in failure detection, identification, and recovery of high-assurance systems is to focus diagnosis on safety-critical components. The goal of probabilistic causal reasoning in diagnosis is to improve performance of fault isolation, However, this reasoning method is dependent on prior reliability knowledge. Our approach aims at focusing the overall diagnostic cycle in two independent ways: first, autonomous agents diagnose high-consequence appliances of a modular manufacturing system and second prior reliability data needed is derived from a quality assurance program. Therefore we present a hybrid technique in combining quality assurance data, failure mode and effects analysis and probabilistic causal reasoning. We develop a dynamic Bayesian network which, given evidence from sensor observations, is able to learn and reason over time. We successfully apply autonomous diagnosis agents in concert with reliability assessment to improve online diagnosis of a pneumatic-mechanical device