Off-design performance analysis of a supercritical carbon dioxide Brayton cycle coupled with a scramjet based on deep learning method

Xiaofeng Ma, Yuchun Shu, Suyuan Zhao, Peixue Jiang, Yinhai Zhu,
Off-design performance analysis of a supercritical carbon dioxide Brayton cycle coupled with a scramjet based on deep learning method,
Energy,
Volume 341,
2025,
139462,
ISSN 0360-5442,
https://doi.org/10.1016/j.energy.2025.139462.
(https://www.sciencedirect.com/science/article/pii/S0360544225051047)
Abstract: The supercritical carbon dioxide (SCO2) Brayton cycle, owing to its advantages in performance and compactness, can simultaneously address the requirements of efficient thermal protection and high-power power supply for hypersonic vehicles when combined with a scramjet. This study investigated the coupled off-design performance of a dual-mode scramjet and a SCO2 Brayton cycle. To improve the off-design simulation efficiency of the coupling model, one-dimensional surrogate models of the SCO2 regenerative cooling and heat exchangers of the Brayton cycle based on a deep neural network were constructed, and the optimal neural network structure was determined through comparison. Furthermore, a deep learning-based off-design coupling simulation model for scramjet, regenerative cooling, and SCO2 Brayton cycle was established. Component-based neural network surrogate models can enhance computational efficiency while preserving the original physical constraints at the cycle system level. Effects of SCO2 cooling area and fuel equivalence ratio on cycle performance were analyzed for Ma4–Ma7 flights, comparing three cycle layouts. Results showed deep neural networks with attention mechanisms performed better, reducing simulation time by about 95 %. The thermodynamic performance of the SCO2 Brayton cycle first increased and then decreased during the flight. The limited cold source was the major reason for the significant performance degradation at the high-Mach-number stage. When comparing the off-design performances of the three cycle layouts, the net power of recuperative layout was 10 %–12 % higher than that of the simple layout and the thermal efficiency was improved by 4.5 %–6 %. The simple layout demonstrated strong regenerative cooling performance, with the regenerative cooling area increasing by more than 10 percentage points and heat absorption increasing by approximately 40 %. Split-flow recuperative layouts performed between the two.
Keywords: Supercritical CO2 Brayton cycle; Third-fluid regenerative cooling; Dual-mode scramjet; Deep learning; Off-design performance