X-ray Micro-Computed Tomography and Deep Learning Segmentation of Progressive Damage in Hierarchical Nanoengineered Carbon Fiber Composites

Abstract

Advanced composite laminates comprised of carbon (micro) fiber reinforced polymer (CFRP) have become widespread in modern high-performance aerospace structures, providing high, tailorable mass-specific stiffness and strength. However, while underpinning such performance benefits, CFRP microstructural heterogeneity and mechanical property anisotropy concomitantly give rise to complex damage mechanisms that lead to difficult-to-predict failure, limiting CFRP understanding. Progressive damage mechanisms in CFRPs generally encompasses a spectrum of modalities, interactions, and sequences across multiple scales, exhibiting broad sensitivity to loading conditions. Dominant damage mechanisms have been identified generally as polymer matrix cracking within (intralaminar) and between (interlaminar, termed ‘delamination’) plies, fiber fracture, fiber bundle microbuckling, and fiber/matrix interfacial debonding. Two emerging solutions aiming to suppress or delay such mechanisms toward enhanced strength and stiffness are considered in this dissertation: (i) aligned carbon nanotube (A-CNT) interlaminar reinforcement (termed ‘nanostitch’) that primarily targets delaminations, and (ii) thin-ply morphology that targets intralaminar cracking and delaminations. Both solutions have demonstrated significant mechanical improvements via standard ex situ tests that lack underlying progressive damage understanding. In view of these limitations, this dissertation advances understanding of composite progressive damage by modern ex situ and state-of-the-art in situ X-ray micro-computed tomography (µCT) studies, including advancing experimental techniques via artificial intelligence (AI), in the context of aerospace-grade CFRP strengthening and toughening effects of nanostitch, thin-ply, and their combination.

Enabling here the first in situ 4D (3D spatial plus temporal) insights into these new strengthening/toughening technologies, synchrotron radiation CT (SRCT) has recently emerged with a focus on conventional composites. SRCT enables unprecedented high-resolution, non-destructive observation of the composite volume during loading, revealing full-field damage progression with resolution at the micron- or submicron-scale, though with limited fields-of-view (FoVs) in the mm to cm range. In situ SRCT of scaled-down double edge-notched tension (DENT) tests of quasi-isotropic baseline and nanostitched thin-ply laminates, coupled with extensive human tomographic segmentations, reveals their strong similarity in intralaminar matrix-dominated damage progression. Significant overall intralaminar matrix damage suppression (6.5× less surface area on average) is found in the thin-ply laminates relative to identically configured experiments with conventional (standard ply thickness) aerospace-grade CFRP composites.

At present, objective quantitative mechanistic insights are extremely challenging to extract from the big (∼10 GB/mm³), typically damage-feature-sparse SRCT datasets due to time-intensive, subjective semi-automatic (human-driven) damage segmentation techniques. Thus, a novel deep learning (DL, a sub-field of AI)-based approach to automate µCT segmentation of multiclass microscale damage in composite laminates is developed, leveraging 65,000 (trained) human-segmented tomographic images for machine development. Following downselection from 20 hyperparametrically different machines, the selected trained machine is shown to segment complex and sparse (≪1% of image volume) composite damage classes to ∼99.99% agreement, unlocking objectivity and efficiency; nearly 100% of human segmentation time is eliminated. This machine performs as well or better than the human due to inevitable, and ‘machine-discovered’, human error, with machine improvements manifesting primarily as new damage discovery, diffuse segmentation augmentation, and segmentation extension to artifact-rich exterior edges.

Next, a second in situ SRCT study coupled with DENT testing is designed to elucidate nanostitch, ply thickness (via ply blocking of thin-ply laminae), and their combinatory effects on progressive damage in cross-ply laminates, using a 20°-canted loading rig for clear 3D reconstruction of all interior features. An intermediate-thickness-ply baseline laminate (2× thicker than thin-ply, similar to conventional plies) exhibits 8% and 17% higher ultimate tensile strengths (UTSs) compared to thick-ply (4× thicker than thin-ply) and thin-ply baseline laminates, respectively, which is explained by an observed progressive damage mode transition from notch-blunting inter- and intra-laminar matrix damage-dominated (thick-ply) to brittle fiber breakage- and diffuse matrix damage-dominated (thin-ply). The overall highest UTS is achieved by the nanostitched thick-ply laminate (15% increase over baseline), which exhibits an effective combination of notch-blunting intralaminar matrix damage and greatly suppressed interlaminar matrix damage (4.3× less surface area at 90% UTS).

Finally, motivated by prior ambient environmental tests and the need to understand temperature and moisture effects for aerospace applications, hygrothermal effects are investigated for the first time for nanostitched laminates, employing quasi-isotropic conventional-thickness CFRP composites subjected to open-hole compression (OHC). The environmental conditions of -55°C/dry and 100°C/dry correspond to statistically significant ultimate strength increases of 2.6% and 5.9%, respectively, for nanostitched laminates over the baseline; however, room temperature/50% relative humidity (RT/50%RH) and 100°C/wet conditions exhibit no ultimate strength change, suggesting correlation of positive nanostitch effects with polymer brittleness. The progressive damage effects associated with nanostitch and RT/50% RH and 100°C/dry conditions are characterized via step-wise ex situ lab-based µCT of large FoVs (mm-scale) in an interrupted loading campaign, revealing similar 0° ply-dominated matrix damage at up to 98% of ultimate strength in the baseline and nanostitched specimens.

This dissertation establishes new understanding of 3D strengthening and toughening mechanisms associated with progressive matrix damage via novel in situ SRCT- and ex situ µCT-based testing of aerospace-grade CFRP laminates in multiple configurations. DL is found to be a disruptive approach to quantitative and objective structure-property characterization, enabling high-throughput, generalizable, ultra-high-resolution damage segmentation of heterogeneous materials with complex hierarchical architectures for the first time. Altogether, these techniques and findings enable and suggest future studies, including toughening and strengthening mechanisms at nearer-failure in situ CT load steps, nanostitched laminates subjected to expanded hygrothermal loading configurations, and AI machines capable of segmenting additional damage classes (e.g., fiber breaks), as vital steps toward rational optimization and prediction of emerging hierarchical composite mechanical properties, with application to layered, bio-inspired, and biological heterogeneous materials.