Predicting the probability of occupational fall incidents: a Bayesian network model for the oil industry

Abstract

Purpose. The probability of being injured or killed from an occupational incident is much higher than a process mishap in the oil and gas industry. The aim of this study was to establish a model for predicting the probability of occupational fall incidents using Bayesian networks. Methods. The study was performed in a selected number of oil refineries. Bayesian network variables (n = 18) were identified using literature as well as expert knowledge. These contributing factors were categorized into four layers (organizational, supervisory, preconditions and unsafe acts) according to the Swiss cheese model. Causal relationships among contributing factors were determined using expert judgment in combination with Dempster–Shafer theory. The conditional probability table of each contributing factor was measured using a questionnaire. Results. The prior probability of fall events was 5.34% (53 cases per 1000 operational workers in 12 months). The posterior probability predicted that using fall protection devices and safe working platforms will decrease more than half (58%) of fall occupational incidents. Conclusion. Bayesian network features including graphical representation, easy belief updating, performance testing and sensitivity analysis facilitate the process of predicting occupational incident probability including fall events. The proposed approach is a step toward quantitative risk analysis of occupational incidents.