Search

Human Errors, e.g. a pilot mismanaged the fuel system causing engine failure and fuel starvation, are known to contribute to over 66% of aviation accidents. However, in some cases, the real sources of the errors are the design of aircraft, e.g.the pilot was confused with the different fuel systems across different models in the same manufacture. The failed collaboration between human operators and the systems therefore has been a major concern for aviation industries. Aviation accident reports are critical information sources to understand how to prevent or reduce such problematic collaboration. In particular, the portions of the reports describing how the behaviour of human operators deviated from an established norm and how the design of aircraft systems contributed to this deviation are particularly important. However, it is a time-consuming and error-prone task to manually extract such information from the reports. One reason is that current accident reports do not aim specifically at capturing the information in format easily accessible for aircraft designers. Therefore, an automatic approach that identifies the sentences describing Human Errors and Design Errors is needed. A preliminary test using hand-crafted cue phrases, i.e. a special word or phrases that are used to indicate the types of sentences, showed a limited identification performance. Therefore, a machine learning technique that uses a greater variety of the linguistic features of the cue phrases than the pre-defined ones and makes the identification decisions based on the combinations these features, looks promising. The examples of the features are active or passive sentence styles and the position of keywords in the sentence. This paper presents the results of developing such automastic identification approach. [ABSTRACT FROM AUTHOR]

While knowledge is viewed by many as an asset, it is often difficult to locate particular items within a large electronic corpus. This paper presents an agent based framework for the location of resources to resolve a specific query, and considers the associated design issue. Aspects of the work presented complements current research into both expertise finders and recommender systems. The essential issues for the proposed design are scalability, together with the ability to learn and adapt to changing resources. As knowledge is often implicit within electronic resources, and therefore difficult to locate, we have proposed the use of ontologies, to extract the semantics and infer meaning to obtain the results required. We explore the use of communities of practice, applying ontology-based networks, and e-mail message exchanges to aid the resource discovery process. [ABSTRACT FROM AUTHOR]