Title:	Flow control using reinforcement learning and computationl fluid dynamics
DNr:	LiU-compute-2025-42
Project Type:	LiU Compute
Principal Investigator:	Saeed Salehi <saeed.salehi@liu.se>
Affiliation:	Linköpings universitet
Duration:	2025-10-28 – 2026-01-01
Classification:	20306
Keywords:

Abstract

This small compute project aims to provide computational resources for a master’s student to perform preliminary training and testing related to the upcoming thesis project “Accelerating reinforcement-learning-based flow control using transfer learning – 30 ECTS”. The goal is to enable the student to become proficient in running OpenFOAM-based CFD cases and coupled DRL frameworks before the official start of the thesis (January 2026). During the project, the student will learn how to set up, execute, and post-process two-dimensional laminar flow simulations, focusing primarily on vortex shedding behind a cylinder. The work will include familiarization with the OpenFOAM solver and Python-based reinforcement learning libraries such as Stable Baselines3 (PyTorch) and Tensorforce (TensorFlow). These tools will be used to train simple DRL controllers for proof-of-concept flow-control cases. The project will also serve as a testbed for verifying the stability and performance of the coupled CFD–DRL framework on the local cluster environment, identifying any computational or parallelization issues before the full-scale thesis computations. The expected workloads include short to medium OpenFOAM simulations (up to 2D laminar and weakly turbulent regimes) and small-scale DRL training runs involving repeated CFD episodes. The outcome will be a fully functional OpenFOAM–DRL environment, tested on the cluster, with example cases and documentation prepared for future use in the main thesis project and in other related research on DRL-based flow control. In this application, I request a small compute project allocation of 5,000 core-hours for two months to support these preparatory activities. At the beginning of next year, a medium-scale compute project (approximately 40,000 core-hours) will be applied for to perform the full thesis computations.