Process safety consequence modeling using artificial neural networks for approximating heat exchanger overpressure severity

Abstract

One challenge in accounting for process safety incidents is that accurate modeling is complex, time-intensive, and requires many inputs. Process safety consequence modeling using first principles can be complicated. At the same time, setting up experiments is not always practical. This work proposes an artificial neural network (ANN) framework to predict process safety metrics to prevent overpressure during tube rupture scenarios with reasonable accuracy. Specifically, we apply a feed forward neural network to predict heat exchanger safety rating that is proportional to the heat exchanger pressure normalized with respect to the maximum allowable pressure. By training ANN to a set of tube rupture simulation data, we are able to bypasses the need for solving tedious dynamic and non-smooth system of equations. The ANN-based models yield safety rating predictions that comply with API 521 overpressure standards. We further demonstrate how these predictions can be used to perform real-time monitoring for a network of heat exchangers in a plant setting.