Introduction of the human factor in the generic failures frequencies of accidents through fuzzy logic

The frequency of an accident scenario is one of the key aspects in the risk assessment field and it is most commonly assessed by a generic failure frequency approach; the accuracy of the calculations is based on the quality of the data used. There exist different sources of generic failure frequencies such as the Reference Manual Bevi Risk Assessments (2009), the Failure Rate and Event Data for use within Risk Assessments of the HSE (2012), and the Handbook of Failure Frequencies of the Flemish Government (2009). Each one of the aforementioned sources takes into account different variables, but aspects such as the mechanical failures or the human factor are not explicitly detailed. Although the mechanical failures may have been considered indirectly, the human factor is difficult to quantify. The latter is a major cause of undesired events in process industries. Due to the complexity of quantifying human error and the causes that lead to it, this factor is not often considered in most of the generic failure frequencies databases. In the present work, the generic failure frequency is modified through the use of fuzzy logic. This theory allows including qualitative variables not considered by traditional methods and deal with the uncertainty involved. A fuzzy modifier has been developed in order to introduce the human factor in the failure frequency estimation. The model takes into account the inclusion of human factors variables such as Organizational factors (Contracting, Training, Communication and Reporting), Job Characteristics Factor (Workload Management, Environmental Conditions, Safety Equipment), Personal Characteristics Factor (Skills and Knowledge, Personal Behaviour). In order to design the proposed model experts’ opinion is necessary. Therefore a questionnaire to gather information on the aforementioned variables has been designed. 40 international experts in the field of risk and human factors have replied, providing interesting results on the different weight of the variables and on the inputs required for the model. As a first attempt to test the model, this has been applied to a real case study of a chemical plant, obtaining new frequency values for a selected event tree. Since the human factor is now reflected in the failure frequency estimation, these new values are more realistic and accurate. As a result, an improvement of the final risk assessment is achieved.