Thermodynamic characteristics evaluation of a biomimetic printed circuit heat exchanger inspired by spine morphology integrating machine learning prediction and interpretability insights

This study introduces a novel Spine-Inspired Printed Circuit Heat Exchanger (SI-PCHE), leveraging the physiological curvature and spinous processes of the human spine to optimize thermodynamic characteristics. While the inherent mechanical resilience and flexural rigidity of the vertebral structure provide a foundational guarantee for equipment integrity, its potential for thermodynamic characteristics intensification through bending features and spinous process inducing geometric perturbations remains to be systematically elucidated. Utilizing three-dimensional numerical simulations, the research systematically evaluates the impact of critical design parameters, like the velocity ratio of cold- and hot- side fluid ( K ), bending angle of channel ( α P ), inclining angle of spinous processes ( η P ), and layout configurations on overall thermodynamic characteristics. To transcend the limitations of conventional empirical correlations, a high-fidelity predictive framework is constructed using four machine learning architectures (Random Forest, CatBoost, MLP, and CNN). The comprehensive thermodynamic characteristics induced by structural variables and boundary conditions are decoded through the integration of SHAP and Sobol global sensitivity analyses. The findings reveal that unidirectional spinous process arrangements tend to induce flow stagnation, alternating configurations facilitate periodic geometric perturbations that effectively destabilize the thermal boundary layer. The mechanism promotes sustained secondary flow intensification, leading to an optimized trade-off between heat transfer enhancement and pressure-drop penalty. Under the configuration of α P = 45° and η P = 30°, the heat capacity flux-weighted PEC + and comprehensive PEC A achieve values within the ranges of 1.07–1.15 and 1.21–1.46, respectively. Sensitivity analysis underscores a significant synergy between α P and η P , demonstrating that η P ≤ 30° is essential for maximizing biomimetic benefits, whereas η P ≥ 45° results in inefficient performance regimes ( PEC < 1.0).The interpretable machine learning approach quantifies the “geometry–thermodynamic synergistic effect”, establishing a robust quantitative foundation for the advanced development of biomimetic PCHE in high-power-density energy conversion systems.