Development of a Neural Network to Predict Path of an Object in a Two Dimensional Potential Flow

Abstract

Neural networks have been widely used in various elds, including uid dynamics, to predict complex phenomena that are di cult to model analytically. In this research, a neural network is developed to predict the path taken by a circular body in a two-dimensional uidic domain. The study involves simulating the potential ow over a rectangular domain inside which a circular body is placed. Fluctuations in di erent parameters such as pressure, forces, and velocity eld during the motion of the body are studied. The Laplace equation is solved at each time step by applying the techniques of nite element method ( FEM) to obtain accurate data, which is fed into the neural network. The neural network comprises of three layers input , middle and output layer. The study is carrried out using computational methods that relies on open-source software Python and its modules like NumPy. The results of the neural network's predictions are compared with accurate data to analyze the error. Fluctutaion of error with respect to di erent hyperparameters of the network is calculated and accordingly suitable hyperparameters of the network are determined.