gym-pfyr

Gym environment that runs PyFR on a 2D flow over a circular cylinder with rotational control.

Installation

1. Install gym from here
2. Install PyFR version v1.7.6
3. Copy gym.patch to the PyFR directory and use git apply gym.patch to apply it
4. From the top level directory, gym-pyfr, run pip install -e .

Contents

• algorithms/ - Contains sample algorithms for solving the vortex suppression problem. Includes Deep RL and proportional controller approaches
• baseline_solutions/ - Contains steady-state solutions for different Reynolds number flows around a circular cylinder. Note that these are approximate
• gym_pyfr/ - This folder contains the gym environment code
• init_states/ - Contains solution states where vortex shedding has already started for different Reynolds numbers
• meshes/ - Contains the available meshes for the 2D circular cylinder

PyFR environment Constructor Arguments

The following are the options available when constructing a gym-pyfr environment. Default values are given with argument = default_val.

• mesh_file - The location of the mesh used by PyFR
• init_file = None - The initial solution file *.pyfrs that PyFR uses to initialize
• config_file = os.path.join(__location__, 'config_base.ini') - The PyFR configuration file
• baseline_file = None - The baseline solution file to compare the state to (to compute reward)
• backend = "cuda" - The PyFR backend
• discrete = False - Whether or not to discretize the action space
• n = 20 - The number of actions to discretize the action space to
• action_multiplier = 0.01 - Multiplier on the actions (the space is set from -2 to 2 so that initially there is not cutoff in the network)
• verbose = False - Whether or not to display more information
• save_dir = "." - The directory to save plots and models
• sol_dir = 'sol_data' - directory to store solution data in
• print_period = 100 - Frequency of printing stats when verbose is off
• plot_best_episode = False - Whether or not to plot the reward and action vs iteration and any new best rewards
• save_epsiode_animation = False - Whether or not to create an animation of each episode
• animation_period = 1 - Number of timesteps between animation frames
• Re = None - Reynolds number override
• tend = None - end time override
• write_state_files = False - Whether or not to save the state files
• write_state_period = 1 - Period of saving state files