Big Data Risk Assessment the 21st Century approach to safety science

Safety Science has been developed over time with notable models in the early 20th Century such as Heinrich’s iceberg model and the Swiss cheese model. Common techniques such fault tree and event tree analyses, HAZOP analysis and bow-ties construction are widely used within industry. These techniques are based on the concept that failures of a system can be caused by deviations or individual faults within a system, combinations of latent failures, or even where each part of a complex system is operating within normal bounds but a combined effect creates a hazardous situation.
In this era of Big Data, systems are becoming increasingly complex, producing such a large quantity of data related to safety that cannot be meaningfully analysed by humans to make decisions or uncover complex trends that may indicate the presence of hazards. More subtle and automated techniques for mining these data are required to provide a better understanding of our systems and the environment within which they operate, and insights to hazards that may not otherwise be identified. Big Data Risk Analysis (BDRA) is a suite of techniques being researched to identify the use of non-traditional techniques from big data sources to predict safety risk.
This paper describes early trials of BDRA that have been conducted on railway signal information and text-based reports of railway safety near misses and the ongoing research that is looking at combining various data sources to uncover obscured trends that cannot be identified by considering each source individually. The paper also discusses how visual analytics may be a key tool in analysing Big Data to support knowledge elicitation and decision-making, as well as providing information in a form that can be readily interpreted by a variety of audiences.