All projects

KAERI 2021 - 2023

AI-designed heat exchangers

Reinforcement learning and topology optimization, connected to metal 3D printing and testing.

My contribution
Developed thermal-fluid topology optimization and reinforcement learning methods, with a metal-printed prototype tested experimentally.
Simulation + experimental testing
14.8% higher simulated heat transfer than conventional topology optimization; metal-printed prototype tested.

Challenge

Printed Circuit Heat Exchangers (PCHEs) are compact, high-performance heat exchangers used in Small Modular Reactors (SMRs). Optimizing their complex flow channel topology to maximize heat transfer while minimizing pressure drop is a significant challenge.

PCHE Structure
Fig. PCHE structure with hot/cold fluid plates and pseudo-3D optimization domains

Approach: Topology Optimization + Deep RL

Optimization Process
Fig. DRL-assisted topology optimization: PPO generates initial design → SIMP topology optimization refines geometry

Dual-Fluid Topology Optimization

  • Pseudo-3D method: Hot plate, cold plate, and conduction plate domains
  • SIMP approach: Solid Isotropic Material with Penalization
  • Pressure constraint: Maximum pressure drop boundary condition
  • Objective: Maximize total heat transfer rate

Deep Reinforcement Learning Initial Design

  • PPO algorithm with CNN for spatial pattern recognition
  • 80 design cells determining solid/fluid regions
  • Reward: Temperature difference improvement at inlet/outlet

Results

Topology Optimization Result
Fig. Topology optimization results showing fluid (white) and solid (black) regions

Dual-fluid topology optimization (2023)

The 3D CFD study found up to 66% higher heat transfer in a specific case, and 32% on average, compared with a conventional PCHE at equal pumping power. These are simulation results from the low-pumping-power design study. Study DOI

DRL-assisted topology optimization (2024)

Simulation comparison Heat transfer improvement
DRL-assisted design vs. topology optimization alone +14.8%
DRL-assisted design vs. original PCHE +32.2%

These percentages use the baselines and conditions of the 2024 study; they are not cumulative improvements over the 2023 results. Study DOI

3D Printing Validation

3D Printed Heat Exchanger
Fig. Metal 3D printed PCHE (SUS316L) with experimental validation
  • Fabricated using powder bed fusion metal 3D printing
  • Experimental verification with thermal loop system
  • Confirmed feasibility of topology-optimized complex geometries

The DRL-assisted prototype was fabricated and tested in the 2024 study. The percentage improvements above refer to simulations, not measured prototype-to-prototype gains.

Publications

SCIE

  • G. Lee, Y. Joo, S.-U. Lee, T. Kim, Y. Yu, H.-G. Kim, “Design optimization of heat exchanger using deep reinforcement learning,” International Communications in Heat and Mass Transfer, 159, 107991, 2024. DOI
  • G. Lee, Y. Joo, Y. Yu, H.-G. Kim, “Dual-fluid topology optimization of printed-circuit heat exchanger with low-pumping-power design,” Case Studies in Thermal Engineering, 49, 103318, 2023. DOI

Conferences

  • G. Lee, Y. Yu, H.-G. Kim, T. Kim, “Application of Deep Learning Reinforcement Learning for Enhancement of Thermal Fluid Topology Optimization Performance,” 3rd Applied Artificial Intelligence Conference, 2023.
  • G. Lee, Y. Yu, H.-G. Kim, “Design of the Optimal Nuclear Heat Exchanger using Deep Reinforcement Learning,” 10th Korea-China Workshop on Nuclear Reactor Thermal-Hydraulics, 2023.
  • G. Lee, Y. Yu, H.-G. Kim, “Optimal Design of Nuclear Heat Exchanger Using 3D Printing-Based Deep Learning Reinforcement Learning,” 2023 Korean Institute of Metals and Materials Fall Conference, 2023.
  • G. Lee, Y. Yu, H.-G. Kim, “Development of 3D Printing AI-Based Component Optimization Design Technology for Small Nuclear Reactor Materials,” 2022 Korean Institute of Metals and Materials Spring Conference, 2022.
  • G. Lee, Y. Joo, H.-G. Kim, Y. Yu, “The Analysis of Topology Optimized 3D Printing Heat Exchanger for the Small Nuclear Reactor,” 2021 Korean Nuclear Society Autumn Meeting, 2021.

Research Project

(2021-2023) Development of Original 3D Printing Technology for Manufacturing Components of Small Nuclear Reactor Materials, KAERI.