Improving the Predictability of Hazardous Scenarios by Natural Language Processing

The completeness and high predictability of hazardous scenarios by hazard identification methods are issues in risk analyses. A way to the improvement is to carry out both an exhaustive - to the extent possible - post-accident and predictive accident analysis. Currently, Natural Language Processing (NLP) allows quick processing of many accident reports. In combination with graphical tools, it is now even possible to automatically output causal diagrammatic models of accidents and visualize them on a multi-scenario accident diagram. A step forward is the application of NLP to support predictive analysis. Predictive accident analysis focuses on identifying deviations from expected or normal conditions, the subsequent events following these deviations, and their interactions leading to an accident. The expected or normal conditions are typically outlined in specifications and procedures. This paper demonstrates how NLP can assist hazard identification and predictive accident analysis during lifting operations on ships and offshore platforms.