Predicting Occupational Struck-by Incident Probability in Oil and Gas Industries: a Bayesian Network Model

• Main Authors: Yaser Shokouhi, Parvin Nassiri, Iraj Mohammadfam, Kamal Azam
• Format: Article
• Language: English
• Published: Tehran University of Medical Sciences 2018-12-01
• Series: International Journal of Occupational Hygiene
• Subjects: Bayesian network; Incident prediction; Oil industry; Struck-by incident
• Online Access: https://ijoh.tums.ac.ir/index.php/ijoh/article/view/400
• ISSN: 2008-5109; 2008-5435

Risk of injury or death due to occupational incidents in the oil and gas industries is higher than that of major incidents such as fire or explosion. In 2017, the largest proportion (36%) of fatalities and greatest number of incidents (24%) in the oil and gas industries were categorized as Struck-by. This study was aimed to develop a Bayesian network (BN) model for predicting occupational struck-by incident probability. Nineteen struck-by causal factors were extracted from the literature. Expert knowledge in addition to Dempster-Shafer theory was used to construct a BN. A questionnaire was developed to measure conditional probabilities of causal factors among participants. Struck-by probabilities of different states of causal factors were also estimated. The prior probability of struck-by incident was 3.09% (approximately 31 per 1000 operational workers per year). Belief updating predicted that preventing workers from being in improper position (in line of fire) would decrease the struck-by incidents by 37%. In contrary, failure of hazard warning (true state) and violation of procedures increased the struck-by probability by 4.08% (an increase of 32%) and 3.96% (an increase of 28%), respectively. The proposed BN model predicted that preventing workers from being in improper position (in line of fire) would decrease the struck-by occupational incidents by 37%. This approach was a step toward quantification of risks associated with occupational incidents. It had advantages including graphical representation of causal factors relationships, easily customizing model, and simply introducing of new evidence (belief updating).