InterPore2026

Micro-CT-Based Permeability Characterization of Virgin and Pyrolyzed Wood Using Deep Learning Segmentation and Image-Based CFD Simulations

21 May 2026, 09:35

15m

Oral Presentation (MS18) High-temperature heat and mass transfer within porous materials for energy and space (T > 800 °C) MS18

Speaker

Abderrahman M'jikou (I2M)

Description

Lignocellulosic biomass is already used in both energy-production and space applications, for example, in the external thermal protection system of Ariane 6. In the current context of environmental transition, a wider range of high-temperature applications is being envisioned, spanning ground-based to space environments. However, the successful use of bio-based composites under extreme conditions relies on a deep understanding and guarantee of their behavior and properties. Toward this goal, this work focuses on the evolution of their microstructure and permeability up to very high temperatures, with pine wood investigated as a proof-of-concept material.

Quantifying permeability in wood is challenging due to pronounced anatomical anisotropy and microstructural evolution during pyrolysis. We develop an image-based pipeline to compute the directional permeability tensor of maritime pine (Pinus pinaster) in virgin and pyrolyzed states from synchrotron micro-CT, deep-learning segmentation, and voxel-resolved CFD.

High-resolution X-ray microtomography was performed at the PSICHÉ beamline (SOLEIL synchrotron) with an effective voxel size of 0.32 µm, enabling visualization of cell walls, lumens, and pyrolysis-induced features. Pyrolysis was conducted in situ under nitrogen, reaching 525 °C at an average heating rate of 84 °C/min. To robustly segment the strongly orientation-dependent anatomy, we trained three independent 2D U-Net models along the longitudinal, radial, and tangential directions and fused their predictions using a majority-vote ensemble. The segmented pore space was converted into adaptive voxel–hex meshes, retaining full resolution near interfaces and coarsening in pore interiors to reduce computational cost.

Steady incompressible creeping-flow simulations were performed in OpenFOAM, ensuring Stokes-regime validity. Numerical representativity was enforced through systematic studies: a padding length of 24 µm (≈80 voxels) was sufficient to eliminate inlet/outlet boundary effects, and a permeability-based REV of 0.39 mm (1300³ voxels) was adopted for subsequent calculations. Mesh refinement tests showed rapid convergence; a fully resolved reference case yielded (for a representative subvolume) $K_L=42.9~{D}$, $K_R=0.21~{D}$, and $K_T=0.054~{D}$, while adaptive coarsening maintained errors below ~1–2% for practical settings.

Microstructural analysis based on local-thickness statistics shows a clear shift toward smaller hydraulic length scales after pyrolysis, with the median pore thickness decreasing from 32.9 to 22.8 µm (−31%) and the mean from 31.3 to 22.4 µm (−28%). The completed study will report the full permeability tensor for both virgin and pyrolyzed states and discuss how pyrolysis-driven morphological changes translate into permeability anisotropy.