Accident analysis reinforced by natural language processing: the case of interactions between maritime and diving operations

Abstract

Maritime and offshore activities carry significant safety risks. Accident analysis is crucial for identifying hazards and justifying safety measures. To inform the planning of future maritime operations and avoid overlooking potential hazardous events, accident analysis should strive to identify as many hazards as possible. However, few of the existing accident analysis methods actually pursue completeness in the analyses. One way to improve completeness has been the analysis of multiple accident reports through post-accident analyses to assess as many recorded events as available. Yet, it is possible to develop a more complete analysis by carrying out complementary predictive analysis and using error mode checklists. Conducting such analyses and exploiting the required information could be challenging, particularly when the operation under analysis consists of many tasks. More challenges appear when each task is associated with numerous hazards or safety measures that are insufficiently documented across scattered sources. In this paper, we address these challenges using natural language processing tools. The paper, therefore, describes the implementation of the Accident Anatomy method; a method that uses both post-accident and predictive analyses, and error mode checklists supported by natural language processing tools. To illustrate the implementation of the method and tools we selected a challenging case. We investigate the hazards associated with commercial diving operation accidents and the respective safety measures, especially those related to subsea cable installation, for which poor data and limited number of assessments are publicly available. Emphasis is given to the interactions between maritime and diving operations because these are also significantly underreported.