Human-reliability analysis of cooperative redundancy to support diagnosis

This paper presents a human reliability analysis method of cooperative redundancy in support of diagnosis. We use a particular redundancy on which a diagnosis function is realized jointly by interacting human and automated controllers. The proposed cooperative redundancy supports a retrospective and experiential failure location function. It integrates an automated controller, based on several points of view: on a list of possible system failures, on a model of coherence between points of view, and on different operations such as focusing on or excluding components or cancellation of a previous operation. A particular dialogue interface is specified for an industrial case to join the human and automated controllers' reasoning, in order to optimize their mutual understanding when making inferences on failures. Such a cooperative redundancy is analyzed with a discrete human reliability approach in order to evaluate its efficiency. This new approach is a conditional, multi-objective probabilistic method which takes into account two types of constraints: constraints based on human behavior, e.g., the time consuming human reasoning, and constraints regarding the system being diagnosed, e.g., the quality of the diagnosis. It is based on different modes of reasoning: the normal mode, the degraded mode, the failed mode, and the success mode. Both normal and degraded modes concern the human behavior dependent constraints. The outputs of both modes are either the success mode if system dependent constraints are satisfied, or the failed mode if not. The cooperative redundancy for diagnosis support is applied to the phone network troubleshooting, and experimental results are analyzed by using the defined analysis method. Conclusions have shown the feasibility of applying cooperative redundancy to diagnosis support. However, even if this cooperative redundancy improves the quality of the diagnosis, future researchers have to improve this redundancy in order to reduce the average delay to make a diagnosis.