Performance enhancement of a C-shaped printed circuit heat exchanger in supercritical CO₂Brayton cycle: A machine learning-based optimization study

The present work is focused on enhancing the overall thermo-hydraulic performance of a previously proposed C-shaped printed circuit heat exchanger (PCHEs) using Machine Learning (ML) Algorithms. In this context, CFD analysis is carried out on 81 different channel configurations of the C-shaped channel geometry, and computed data is used to train three ML algorithms. Later, C-shaped channel geometry is optimized by coupling the trained ML model with the multi-objective genetic algorithm (MOGA). Finally, the optimized channel geometry (called optimizedML) is investigated numerically for a wide range of Reynolds numbers. Its performance is compared with the zigzag geometry, C-shaped base geometry, and previously optimized C-shape channel geometry using response surface methodology (RSM). The findings showed that the multilayered approach combining MOGA, CFD, and machine learning techniques is beneficial to accomplish a robust and realistic optimized solution. Comparing the thermo-hydraulic characteristics of the optimizedML channel geometry with zigzag channel geometry shows that the former is up to 1.24 times better than the latter based on the performance evaluation criteria (PEC). Furthermore, the overall performance of the optimizeML channel geometry was found up to 21% and 16% higher than the optimized RSM geometry on the cold and hot sides, respectively.