Heat Transfer Model For Micro Scale Flow Boiling Based On Interpretable Machine Learning

Abstract High heat transfer is achieved in micro scale flow boiling by coupling latent heat of vaporization and vibrant vapor bubbles in microchannels. The difference between micro scale and conventional flow boiling is the relative significance of surface-tension in the four forces, which are included in: Froude number (FrLO), Boiling number (Bo), Bond number (Bd), and dimensionless gas velocity (JG) as the core influential factors, revealing their synergistic interactions, identified by SHAP analysis of developing an interpretable machine learning model by covering complex micro scale flow boiling regimes, wide operating conditions range, and significant scale effects inherent. The analysis reveals that FrLO, Bo, Bd, and JG are the core features influencing the micro scale flow boiling heat transfer, with their effects exhibiting distinct segmented or threshold behaviors. This makes our model be universal for various working fluids, saturation temperature, tube diameters, vapor qualities, mass fluxes and heat fluxes based on dimensionless numbers to achieve high-precision prediction of the heat transfer coefficient in evaporating microchannels. Based on the physical thresholds and interaction mechanisms determined by the SHAP interaction analysis (FrLO ≈ 50, JG ≈ 12, Bd ≈ 1, Bo ≈ 00005), a new correlation was developed by embedding the co-ordination coupling terms (FrLO×Bd and JG × Bo) and the vapor quality penalty factor into the correlation structure. This work is not only helpful to understand the basic mechanism of microscale flow boiling, but also provides a new theoretical method for improving the prediction of heat transfer in microcircuits.