RL4MicrostructureEvolution

Reinforcement Learning Environments for Structure Guided Process Optimization Tasks

Prerequisites

• Download compiled microstructure-path simulation (uniax_simulator_for_microstructure_evolution_40tasks) and material model from https://fordatis.fraunhofer.de/handle/fordatis/201 and put to /msevolution_env/assets/sim
• Intel Fortran environment to run the simulations and proper environment variables (eg. export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/<username>/intel/compilers_and_libraries_2019.3.199/linux/compiler/lib/intel64_lin:/home/<username>/intel/compilers_and_libraries_2019.3.199/linux/mkl/lib/intel64_lin)

Install and run microstructure-evolution environments

• cd RL4MicrostructureEvolution
• pip install .
• cd msevolution_env/examples
• python sg_random_agent.py for single-goal version or python meg_random_agent.py for multi-equivalent goal version

Cite

@article{dornheim2022deep,
  title={Deep reinforcement learning methods for structure-guided processing path optimization},
  author={Dornheim, Johannes and Morand, Lukas and Zeitvogel, Samuel and Iraki, Tarek and Link, Norbert and Helm, Dirk},
  journal={Journal of Intelligent Manufacturing},
  volume={33},
  number={1},
  pages={333--352},
  year={2022},
  publisher={Springer}
}