Document Type
Article
Publication Date
9-25-2025
Abstract
Reliable prediction of frictional pressure drop during condensation in mini‑ and microchannels underpins both thermal management effectiveness and overall heat transfer performance in compact two‑phase heat exchangers, cold‑plates, and on‑chip cooling loops. Excess pressure loss burdens pumps, raises electrical consumption, and can destabilize flow, whereas under‑prediction risks temperature overshoot and premature dryout. Conventional empirical correlations and flexible machine-learning models can lose accuracy once channel size, working fluid, or operating conditions stray beyond their testing range. This study uses a physics‑assisted machine‑learning framework that overlays an XGBoost residual learner on the Kim–Mudawar separated‑flow correlation to achieve high fidelity and robustness in pressure drop prediction. A curated database of 6566 condensation data points (40 studies; 0.07 ≤ Dₕ ≤ 6.22 mm; 32.7 ≤ G ≤ 1926 kg m⁻² s⁻¹ 22 fluids) was assembled. Four feature sets (physical, dimensionless, statistically selected, full) were evaluated, and Bayesian hyper‑parameter optimization combined with five‑fold cross‑validation plus fluid‑ and mass‑velocity holdouts quantified both interpolation and extrapolation. Across the full dataset, physics‑assisted machine‑learning lowered the mean absolute percentage error from 24 % with Kim–Mudawar and 9.5–10.3 % with pure machine learning to 7.4–8.3 %, achieving R² > 0.985. For the benchmark refrigerant R134a, interpolation mean absolute percentage error dropped from 22 % (Kim–Mudawar) to ≈14 %. For dielectric fluids HFE7000/HFE7100 (unseen during training) extrapolation error fell from >150 % with pure machine learning to ≈40 %. Mass‑velocity holdouts confirmed ≤15 % error at high mass velocity and ≤42 % at the most challenging low mass velocity conditions. These advances enable more reliable pump sizing, manifold design, and thermal‑resistance budgeting, directly supporting the development of energy‑efficient, high‑heat‑flux thermal management hardware for electronics, electrified vehicles, and aerospace platforms.
Keywords
condensation, machine learning, physics-assisted machine learning, pressure drop, thermal management, two-phase flow
Language
English
Publication Title
International Journal of Heat and Mass Transfer
Grant
N00014-24-1-2039
Rights
© 2025 The Authors. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/), which permits non-commercial copying and redistribution of the material in any medium or format, provided the original work is not changed in any way and is properly cited.
Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 4.0 International License.
Recommended Citation
Golezani, F. B., Li, J., Naderi, F., Meylikhov, A., Pirnstill, L. M., Kim, S. M., ... & Kharangate, C. (2026). Enhanced prediction of condensation pressure drop in mini and microchannels using physics-assisted machine learning. International Journal of Heat and Mass Transfer, 255, 127838. https://doi.org/10.1016/j.ijheatmasstransfer.2025.127838
Manuscript Version
Final Publisher Version