Application of machine learning methods for process safety assessments

Abstract

The accurate prediction of gas dispersion and the potential consequences of gas explosions hold a pivotal role in the assessment of explosion design loads for oil and gas processing facilities. This often involves the use of computational fluid dynamics (CFD) simulations, a widely adopted practice in the field. The extent of CFD simulations required depends on the specific characteristics and size of the facility. In many cases, a substantial number of simulations, often in the thousands, are needed to comprehensively assess the potential outcomes in the event of a hydrocarbon loss of containment. These simulations account for the complex three-dimensional nature of the facility, the surrounding environmental conditions, and the properties of the leaking hydrocarbon fluids. Although unquestionably invaluable, CFD simulations impose significant temporal constraints upon their execution and necessitate the allocation of substantial efforts and Central Processing Unit (CPU) time. In this paper we develop a neural model tailored specifically for the analysis of CFD gas dispersion and gas explosion scenarios. This model leverages the capabilities of machine learning algorithms to expedite the execution of these complex studies. The proposed neural network model has the advantage of being able to handle a wide range of scenarios in a fraction of time it takes to perform the CFD simulations, making it particularly useful for large-scale processes facilities. The accuracy of the predictions is remarkably high, providing a high level of confidence in the predictions of the flammable gas clouds sizes across various scenarios, as well as the resulting explosion overpressures.