Application of Artificial Neural Network for Prediction of Sprinkler Actuation Time in Fire

Fire phenomena are complex and involve non-linear interactions between environmental and fire parameters. Computer based methods (Field model and Zone models), by use of numerical methods applyied on mathematical models in the form of differential equations have been widely used for predicting the results of fire phenomena. In the present work an Artificial neural network (ANN) model has been used to predict the time for sprinkler actuation as a function of heat release rate, vertical distance from the ceiling and horizontal distance from the fire source. A set of input data has been generated from numerical studies where the code BRANZFIRE for a typical geometry of Emergency Switch Gear Room (ESGR) of a Pressurised Water Reactor (PWR) for different heat release rate, vertical distance from the ceiling and horizontal distance from the fire source for training the ANN model. The trained ANN model was used for the predictions of the time to sprinkler actuation. The ANN predictions were in good agreement with the predictions from the numerical analysis results. This paper also discusses briefly the ANN back propagation model used in this study.