Prediction of Bifurcation Phenomenon Inside a Sudden Expansion Pipe Using Neural Network

The current study investigates internal flow through a sudden symmetrical expansion pipe of different expansion ratios. The flow pattern is studied using Finite Volume based computational tool which trained a neural network to predict bifurcation phenomena at different expansion ratios and Reynolds Number. This study focuses on the training of neural network using the dataset generated by the Finite Volume based computational tool and the predicted dataset is validated by comparing it with the results of experimental investigation obtained from relevant literature. Latin Hypercube Sampling is used to generate random values of Reynolds Number, which is given as an input to the solver, from where we get outputs for vortex lengths. This process is repeated in a loop for the sample space of Reynolds Number and varying expansion ratios. A neural network is trained by the dataset generated by the solver using Reynolds Number as input and vortex lengths as output.