The goal is to apply TensorFlow capabilities to train a network on Fluid Dynamics data.
Current dataset shown below, contains velocity magnitude for flow in a microchannel with random obstacles:
Obtaining CFD data is not an easy task, especially having random objects makes the process more complicated. This is since the domain needs to mesh and things become complicated from the aspect of how and where data points are collected.
Here, the mesh has been fixed as a remedy, limiting the problems that can be handled. The data, therefore, consists of a velocity magnitude on an array of 50x100 domain. The first 50 marks the inlet, and therefore, their velocity magnitude is identical to inlet velocity of the problem. The rest are calculated velocity magnitudes influenced by the existence of random objects. The place of objects is marked by NaN data points (or perhaps zero velocity magnitudes).