# CFD Reinforcement Learning Topology optimization wind tunnel

I want to create a reinforcement learning environment, designed for win tunnel simulations, where for each iteration a deep convolutional model could receive the 3D vector/scalar fields from the past simulation and output a better shape that maximizes the reward function (e.g. minimize drag, maximize lift, etc.)

The observation and action space for the neural network is the same, the inputs of the model will be 3D arrays representing velocity field, pressure field, etc. and the output will be a 3D array (created using Conv3DTranspose) with values [0, 1] which represents the mesh. I'm thinking that the architecture of the model could be something similar to an auto-encoder. My plan is to use the algorithm of Marching Cubes in order to create the mesh from those points and openFoam for the CFD simulations.

This is a small diagram showing the workflow

The goal will be to have multiple trained models specialized in optimizing a particular reward function, like minimizing drag or maximizing lift, for any object/shape given as input. What are your thoughts on this? Do you think it makes sense?