Automated Fault Tree Generation: Bridging Reliability with Text Mining

Proper preventive maintenance of complex systems, such as those used for power generation and medical diagnosis is dependent on the availability of their up-to-date reliability models. These models are constructed from historical maintenance and fault information of the equipment. Due to the complex nature of these machines, constructing these models involves significant manual effort which limits the widespread use of reliability-centric maintenance schemes. In this paper, we describe a process for automating the construction of fault trees, a class of non-state space reliability models, by analyzing maintenance data available as free-form text. It uses a combination of linguistic analysis and domain knowledge to identify the nature of the failure from short plain text descriptions of equipment faults. This information is used to automatically enrich and evolve existing fault trees for better reliability estimation