Abstract:
Using computational fluid dynamics (CFD) to solve fluid flow problems can use a lot of
computer processing power and simulation time. Artificial neural networks (ANN) can be
regarded as universal learners that are capable of learning nonlinear patterns or
relationships among many variables. A very well-known benchmark problem for viscous
incompressible fluid flow in the lid-driven cavity problem. People have developed different
numerical procedures to solve it. It is widely regarded as the first problem people usually
try to solve when they come up with a new approach. This research aims to apply fully
connected neural networks to learn and predict fluid flow inside a lid-driven cavity. A
double lid-driven cavity with top and bottom moving walls having some internal square
objects was selected as a training data to train, test and compare several fully connected
neural networks having different parameters to predict fluid flow inside it. The results show
that by training a neural network to recognize fluid velocity patterns around simple square
objects inside the cavity, it is possible to predict fluid velocities around objects having
relatively complex geometries with significant accuracy in a fraction of the time required
by a CFD solver. The results also show the comparison between effects of using different
mesh sizes in CFD and different learning rates in the neural network model.
