Characterization of heterogeneous material architectures through x-ray computed micro-tomography

Low-density carbon-phenolic composites, such as phenolic impregnated carbon ablator (PICA), possess meso-, micro-, and nano-pores making the structure of the resin difficult to characterize. X-ray Computed Tomography (XRCT) is a widely ac- cepted method for characterizing the carbon fiber preform, yet this technique is often insufficient for visualizing the resin because of its minimal attenuation of the incident x-ray beam. Phase Contrast Retrieval (PCR) reconstruction considers not only the linear attenuation coefficient, but also the refractive indices of the constituent materials. This research leverages phase contrast tomography to characterize the meso- and micro-structure of the porous resin phase with a newly developed capability. Using a voxel size of 7 μm, the density gradient throughout the carbon-phenolic material was visualized highlighting the advancement of pyrolysis into the material. This method was used to observe charring and pyrolysis of PICA for multiple samples tested at Langley Research Center. Extending the application to FiberForm visualized regions of high and low fiber concentration which aligned with the nominal volumetric porosity of the material. Minimizing the voxel size to 1.112 μm offered novel micro-structure characterization through direct segmentation of the resin phase distinct from the fiber phase. Once the resin-void interface is located, further investigation regarding the nano-structure of the porous phenolic resin is achieved. Possessing standard voxel sizes on the micrometer scale, XRCT is incapable of resolving the nano-structure of such porous materials. Using focused ion beam (FIB) milling and scanning electron microscopy (SEM), a three-dimensional nano-structure is generated for PICA. Cleaning cross sections expose the interior structure to be captured through SEM. Voxel dimensions of 4 nm x 6 nm x 10 nm have been achieved for the fully automated FIB-SEM method. Charging effects on the non-conductive resin surface during imaging are mitigated through use of a platinum sputter coating sufficient for dissipating the build-up of electrons at the surface. Image registration within the Avizo3D software is used to correct the +52◦ offset between FIB and SEM. Semantic segmentation of the scans is performed using a developed and trained 2DCNN with a 5-layer U-Net architecture. Deconstructing the high-resolution rectangular SEM scan into patches allows for parallelization of the automated semantic segmentation. Utilization of standard volume generation methods (marching cubes algorithm, Avizo3D, etc.) extend the stack of masks to a three-dimensional surface volume revealing the nano-structure of the phenolic resin and enabling future mechanical and fluid simulations with highly realistic renderings of the material.