Learning from accidents: analysis and representation of human errors in multi-attribute events

Regardless of the evolution of engineering systems and fabrication methods, recent major accidents exposed the risk behind modern human economic activities to an inquiring and perplexed society. These events brought out the fact that interactions between complex systems, cutting-edge technologies and human factors may trigger particular accident sequences that are very difficult to predict and mitigate through traditional risk assessment tools. Thus, the purpose of this study is to overcome barriers to dealing with complex data by translating multi-attribute events into a two-dimensional visualisation framework, providing means to communicate high-technology risks and to disclose surrounding factors and tendencies that could lead to the manifestation of human errors. This paper first discusses the human error and human factors role in industrial accidents. The second part applies Kohonen's self-organising maps neural network theory to an accident dataset developed by the authors, as an attempt to improve data exploration and classify information from past events. Graphical interfaces are then generated to produce further insight into the conditions leading to the human errors genesis and to facilitate risk communication among stakeholders.