Efficient data-driven prediction of effective thermal conductivity in porous phenolic resins from scanning electron microscopy images

Phenolic-resin-based porous composites are widely used in thermal protection systems for near-space applications, where fast and reliable prediction of effective thermal conductivity is crucial. This study proposes a scanning electron microscopy (SEM)-image-based, data-driven approach to predict the effective thermal conductivity of phenolic resin materials with different porous microstructures. Representative SEM images are processed using a porosity-calibrated threshold segmentation to ensure consistency between image-derived and experimental porosity. Key microstructural descriptors, including porosity, equivalent particle diameter, and bonding ratio, are extracted and used as inputs to the SVR model. Application-level validation on three newly acquired specimens showed good agreement with experiments, yielding a mean relative error of 5.41%, and the SVR predictor consistently outperformed classical analytical correlations. Sensitivity analysis indicated that porosity dominated the global importance (60.9%), followed by particle diameter (27.8%) and bonding ratio (11.3%). However, the locally dominant factor can shift across thermal-conductivity regimes. Overall, the proposed method provides an experimentally anchored, ultra-low-cost workflow for predicting thermal conductivity directly from SEM images and offers quantitative guidance for microstructural optimization of phenolic-based thermal protection materials.