Learning from accidents: Analysis of multi-attribute events and implications to improve design and reduce human errors

摘要

© 2015 Taylor & Francis Group, London.High-technology accidents are likely to occur under a complex interaction of multiple active failures and latent conditions, and recent major accidents investigations are increasingly highlighting the role of human error or human-related factors as significant contributors. Latent conditions might have long incubation periods, which implies that a number of design failures may be embedded in systems until human errors trigger an accident sequence. Consequently, there is a need to scrutinise the relationship between enduring design deficiencies and human erroneous actions as a conceivable way to minimise accidents. This study will tackle this complex problem by applying an artificial neural network approach to a proprietary multi-attribute accident dataset, in order to disclose multidimensional relationships between human errors and design failures. Clustering and data mining results are interpreted to offer further insight into the latent conditions embedded in design. Implications to support the development of design failure prevention schemes are then discussed.