Understanding the thermal behavior of subsurface porous media is critical for technologies such as nuclear waste disposal, geothermal energy, and underground thermal storage. Yet, estimating effective thermal conductivity (ETC) has long been a bottleneck — existing approaches are either overly simplified or computationally intensive.

In their latest study, Mohamed Elmorsy, Wael El-Dakhakhni, and Benzhong Zhao introduce a machine learning framework that leverages transfer learning to predict ETC in 3D digital rock samples with high accuracy and drastically reduced computational cost. By reusing pre-trained convolutional neural networks, the authors cut down on data requirements and training time while preserving fidelity.

Highlights:

• Transfer learning greatly improves ETC prediction efficiency
• Strong performance even with limited data
• Demonstrates potential for foundation models in digital rock analysis

Read the full article in InterPore Journal here: https://doi.org/10.69631/ipj.v2i3nr75