Application of Bayesian networks for the prediction of gas accidents in semi-mechanized underground operations on the southern coast of Peru

Abstract

Occupational health and safety are key factors in the development of underground operations and works. Reducing the risks associated with accidents caused by gases is essential to prevent risk situations and protect the integrity of workers. The present investigation evaluated the use of Bayesian networks as a different tool in accident investigation. The general objective is to propose an accident investigation model as a predictive tool for the control and subsequent reduction of gassing accidents. To establish this model, Bayesian networks and structural models were used that complemented the operation of the first iterations. Bayesian networks were used to identify related risk factors, assess their impact, and understand the interaction between them. The study was based on a comprehensive analysis of gas accidents over a 15-year period. The main finding of the investigation focuses on the identification of 3 critical zones within the Cinco Cruces operation with associated probabilities of 0.712, 0.446 and 0.652. The value of the Bayesian inference obtained is 0.36, which through the analysis of the ROC curve establishes it as a non-false positive of regular prediction. This makes it possible to identify which are the future conditions in which the events can be repeated and to which key safety factors they are linked. Based on them, an action plan was proposed to create a PETS (Written Safe Work Procedure), which includes recommendations, methodologies, equipment, and tools to prevent future gassing accidents. The incorporation of Bayesian networks makes it possible to adhere to predictive approaches to mining accident investigation processes.