Development of a Repeatable and Objective Decision Model for the Human Error Probability Estimation According to the HEART Framework

Abstract

Abstract The proposed methodology aims at developing a model to provide a structured and repeatable approach to estimate the error probabilities of the human actions carried out during well construction and control operations according to the Human Error Assessment and Reduction Technique (HEART) approach. This relies on a combination of AI-based and traditional techniques that exploit all the available Knowledge, Information and Data (KID) in the field. The modeling framework consists of different steps. According to the decision science paradigm, identification of the criteria to assign a nominal Human Error Probability (HEP) to each human action considered in the analyzed risk scenario, according to the HEART framework. The development of a Bayesian Network to characterize the qualitative categorical variables for each HEP assignment criterion, based on the available KIDs, and describe the uncertainty on the factors that HEART considers to modify the nominal HEP into the actual HEP. Encoding of the estimated task actual HEP into the risk scenario probabilistic model. The developed methodology has been applied to a case-study related to a "blow-out prevention scenario" in the context of a drilling platform. All the human actions involved in the prevention ofblow-out and carried out on the rig site have been analyzed, collecting the available KID. This provides evidence to build a Bayesian Network model, which allows obtaining a probability distribution for each task over the possible generic tasks, thus preventing the limiting approach of relying on a single assignment, which inevitably neglects some aspects and, thus, results in an inaccurate HEP estimation. This allows overcoming the limitations of the HEART methodology, which is characterized by vague descriptions, as we focus on understandable criteria harmonizing HEART categories with the task features. The results show the complexity deriving from this attribution process, highlighting how the generic task which shows more compatibility with the analyzed task is generally not the most conservative and robust choice. The novelty of the approach is in the rationalizationof the application of the HEART methodology, by developing a HEP attribution methodology that exploits all the available KID without reducing to traditional qualitative categorization.