FACT: Fibril Analysis for Cellulose Technology

• FACT is a comprehensive framework that integrates high-resolution imaging, segmentation, and reconstruction to transform cellulose fibril characteristics into quantitative descriptors.
• It employs diverse modalities—synchrotron microtomography, SEM, AFM, and neutron tomography—to analyze morphology, orientation, porosity, and mechanical responses at multiple scales.
• FACT standardizes metrics for localized mechanics and chemical disassembly, offering actionable insights for designing composites, membranes, and paper networks.

Fibril Analysis for Cellulose Technology (FACT) denotes a set of integrated analytical workflows for cellulose-based materials in which modality-specific measurements are converted into quantitative descriptors of fibril morphology, orientation, porosity, mechanics, interfacial state, or process response. In the reported literature, FACT is used for high-resolution synchrotron phase-contrast microtomography and 3D pore morphometrics in flax fibres, for machine-learning segmentation and morphological thinning in negative contrast SEM images of cellulose nanofibers, for AFM–CLSM–SEM mapping of humidity-dependent local mechanics in single cellulose fibres, and for multiscale orientation analysis in neutron tomography and flow-stop birefringence experiments (Quereilhac et al., 2023, Baez et al., 8 Sep 2025, Auernhammer et al., 2020, Busi et al., 2024, Rosén et al., 2018). The term therefore refers not to a single instrument or a single algorithm, but to a methodological framework in which fibril-resolved measurements are standardized, spatially registered, and interpreted against processing history and macroscopic performance.

1. Conceptual scope and analytical logic

FACT is unified by a recurring sequence: acquisition of structurally informative data, segmentation or reconstruction of cellulose-relevant features, extraction of physically interpretable metrics, and correlation of those metrics with mechanics, transport, or processing behavior. In the flax-fibre study, the framework is explicitly described as leveraging high-resolution synchrotron phase-contrast microtomography, advanced 3D segmentation, and quantitative morphometrics to link cell-wall ultrastructure, porosity, microfibrillar angle perturbation, and mechanical performance (Quereilhac et al., 2023). In the CNF image-analysis study, FACT is defined through binary segmentation, topology-preserving thinning, and whole-network width measurement from NegC-SEM images (Baez et al., 8 Sep 2025). Auernhammer et al. formulate a FACT workflow in which maps of $E(x)$, $F_a(x)$, $E_{diss}(x)$, $S(x)$, and $\theta(x)$ are correlated to segment a single fibre into regions of distinct behavior under controlled relative humidity (Auernhammer et al., 2020).

This common logic makes FACT inherently multiscale. The reported workflows encompass sub-elementary features at $\approx 0.44$ nm and $\approx 0.88$ nm in TEMPO-oxidized nanofibers, pore layers with median radial width $0.957\,\mu$m in flax-fibre kink-bands, and centimetre-scale foam specimens imaged by neutron tomography (Silvestre et al., 2020, Quereilhac et al., 2023, Busi et al., 2024). Taken together, these reports suggest that FACT is best understood as a quantitative bridge between cellulose ultrastructure and engineering observables, rather than as a narrowly defined characterization protocol.

2. Structural imaging, reconstruction, and segmentation

A central branch of FACT is high-resolution imaging of fibrillar architecture and its defects. In flax fibres, measurements were performed on individual fibres at the ANATOMIX beamline using a polychromatic X-ray beam centered at $\sim 12$ keV, with a $2048\times 2048$-pixel detector, effective pixel size $F_a(x)$0m, voxel size $F_a(x)$1, a $F_a(x)$2 mm$F_a(x)$3 field of view, and total scan time per $F_a(x)$4 mm fibre segment of $F_a(x)$5 min. The raw projections were reconstructed with PyHST2 using filtered back-projection and a Paganin phase retrieval with $F_a(x)$6 kernel length $F_a(x)$7m, followed by a non-local means filter, global thresholding of voids, manual separation of the lumen from defect-induced pores in Avizo 2021.1, and “Separate Objects” analysis for individual-pore morphometrics (Quereilhac et al., 2023). This pipeline isolates ultrastructural void organization inside kink-bands rather than treating the fibre as a homogeneous solid.

In image-based CNF metrology, FACT requires a binary image separating foreground from background and implements two pixel-classification approaches: Weka for small datasets and rapid per-image training, and a modified U-Net CNN for larger, heterogeneous datasets. The U-Net reduces parameters from $F_a(x)$8 M to $F_a(x)$9 M by halving feature-map counts in each layer; the input workflow mirror-pads each $E_{diss}(x)$0 image to $E_{diss}(x)$1, extracts overlapping $E_{diss}(x)$2 patches with $E_{diss}(x)$3 px overlap, applies rotations by $E_{diss}(x)$4, $E_{diss}(x)$5, $E_{diss}(x)$6, $E_{diss}(x)$7 and mirror at $E_{diss}(x)$8, and splits data into $E_{diss}(x)$9 train, $S(x)$0 validation, and $S(x)$1 test, totaling $S(x)$2 patches from $S(x)$3 images. Inference takes $S(x)$4 s per $S(x)$5 image on NVIDIA RTX A5000, and the full pipeline with a pre-trained U-Net is reported as less than $S(x)$6 min per image (Baez et al., 8 Sep 2025). The key structural operation after segmentation is iterative thinning,

$S(x)$7

continued until convergence to a one-pixel-wide skeleton.

A further extension of FACT is multiscale 3D orientation imaging by multi-directional dark-field neutron tomography. Here a single-absorption grating with period $S(x)$8m creates a 2D spatial intensity modulation, the sample is placed $S(x)$9 cm upstream of the detector, and the real-space correlation length is set by

$\theta(x)$0

with $\theta(x)$1 nm under the stated geometry. Three in-plane scattering-sensitivity directions, $\theta(x)$2, $\theta(x)$3, and $\theta(x)$4, are acquired over $\theta(x)$5 projections across $\theta(x)$6, each with $\theta(x)$7 min exposure, and reconstructed with the SIRT algorithm in the ASTRA toolbox (Busi et al., 2024). Unlike X-ray or electron workflows, this route is explicitly positioned as non-destructive and suitable for fragile hierarchical biomaterials.

3. Quantitative descriptors: porosity, width, and orientation

FACT is defined as much by its metrics as by its instrumentation. In the flax-fibre implementation, local porosity in a sub-volume is

$\theta(x)$8

with $\theta(x)$9, and the local orientation proxy for cellulose microfibrils is expressed through the inclination of pore-layer long axes relative to the lumen axis, with first two statistical moments

$\approx 0.44$0

Using this framework, $\approx 0.44$1 kink-bands were studied over a cumulative fibre length of $\approx 0.44$2 mm; pore-layer thickness had median $\approx 0.44$3m with range $\approx 0.44$4–$\approx 0.44$5m; inter-layer spacing had median $\approx 0.44$6m with range $\approx 0.44$7–$\approx 0.44$8m; and $\approx 0.44$9 individual pores were extracted, with volumes from $\approx 0.88$0 to $\approx 0.88$1 and median $\approx 0.88$2. Local $\approx 0.88$3 rises from $\approx 0.88$4 in intact regions to peaks approaching $\approx 0.88$5 in kink-band zones, while pore inclination angles lie between $\approx 0.88$6 and $\approx 0.88$7, with $\approx 0.88$8 and $\approx 0.88$9 over $0.957\,\mu$0 measurements (Quereilhac et al., 2023). The linear increase of pore-layer radius versus distance from the lumen with slope $0.957\,\mu$1 is taken to demonstrate a concentric “onion-peel” arrangement at successive interfaces of the S2(G) growth layers.

In whole-network CNF morphology analysis, the defining FACT observable is width derived from the Euclidean distance transform. If $0.957\,\mu$2 is the final skeleton and $0.957\,\mu$3 the unfiltered binary foreground, then

$0.957\,\mu$4

The width set $0.957\,\mu$5 is summarized by

$0.957\,\mu$6

Validation on idealized rectangular branches with true widths $0.957\,\mu$7, $0.957\,\mu$8, $0.957\,\mu$9, and $\sim 12$0 px yielded FACT peak means $\sim 12$1 px, corresponding to error $\sim 12$2. On low-branching CNFs, five images at $\sim 12$3 nm/px gave FACT $\sim 12$4 nm versus manual $\sim 12$5 nm, and one image at $\sim 12$6 nm/px gave FACT $\sim 12$7 nm versus manual $\sim 12$8 nm. For high-branching CNFs, FACT produced $\sim 12$9 nm versus manual $2048\times 2048$0 nm, with the reported skew toward larger widths attributed to the fact that longer, thicker fibrils contribute more skeleton pixels (Baez et al., 8 Sep 2025).

Orientation analysis is also formalized in tensorial or order-parameter terms. In neutron tomography, the second-moment orientation tensor is

$2048\times 2048$1

and Herman’s orientation factor in a reference direction $2048\times 2048$2 is

$2048\times 2048$3

The dark-field implementation does not reconstruct a full continuous ODF, but instead infers local anisotropy through an eccentricity $2048\times 2048$4 computed from orthogonal dark-field channels. Cross-validation by SAXS gave Herman factors at $2048\times 2048$5 nm of $2048\times 2048$6 for the CNC shell, $2048\times 2048$7 for the CNF shell, and $2048\times 2048$8 for the CNF-uit core (Busi et al., 2024). A plausible implication is that FACT descriptors can remain comparable across modalities only when the underlying orientation proxy is stated explicitly.

4. Mechanical and hygro-mechanical mapping

A second major branch of FACT concerns local mechanics under controlled environmental state. Auernhammer et al. describe a single cotton-linter fibril suspended over a $2048\times 2048$9 mm trench and clamped at both ends inside a humidity-controlled AFM chamber at RH $F_a(x)$00, $F_a(x)$01, $F_a(x)$02, and $F_a(x)$03, with equilibration for $F_a(x)$04 min after each RH step. A spherical SiO$F_a(x)$05 colloidal probe of diameter $F_a(x)$06m is attached to a V-shaped cantilever with $F_a(x)$07–$F_a(x)$08 N/m, and static force–distance curves are acquired at points spaced by $F_a(x)$09m. Local stress and strain are computed as

$F_a(x)$10

and the Young’s modulus map is extracted from the linear regime of $F_a(x)$11–$F_a(x)$12 curves. Combined with CLSM-based swelling, $F_a(x)$13, and SEM-derived fibril orientation $F_a(x)$14, the workflow identifies “wet spots” characterized by local maxima in $F_a(x)$15 and minima in $F_a(x)$16 and $F_a(x)$17, accompanied by peaks in $F_a(x)$18 and $F_a(x)$19 (Auernhammer et al., 2020). The reported quantitative ranges are sharply humidity dependent: $F_a(x)$20 changes from $F_a(x)$21–$F_a(x)$22 GPa at RH $F_a(x)$23 to $F_a(x)$24–$F_a(x)$25 GPa at RH $F_a(x)$26; adhesion rises from $F_a(x)$27 nN at $F_a(x)$28 RH to $F_a(x)$29–$F_a(x)$30 nN at $F_a(x)$31 RH; dissipated energy increases from $F_a(x)$32 J to $F_a(x)$33 J; and swelling at $F_a(x)$34 RH reaches up to $F_a(x)$35 in “loose” ROIs and $F_a(x)$36 in “tight” ROIs.

Humidity sensitivity is corroborated in nanofibrillated cellulose films. Simão et al. report dry, local water-free Brillouin Light Scattering values $F_a(x)$37 m s$F_a(x)$38 and $F_a(x)$39 m s$F_a(x)$40, corresponding to $F_a(x)$41 GPa, $F_a(x)$42 GPa, bulk modulus $F_a(x)$43 GPa, Poisson ratio $F_a(x)$44, and $F_a(x)$45 GPa. Under QNM-AFM, the same material shows $F_a(x)$46 GPa at $F_a(x)$47 RH and $F_a(x)$48 GPa at $F_a(x)$49 RH, while adhesion changes from $F_a(x)$50 nN to $F_a(x)$51 nN, with repeatability of $F_a(x)$52 and $F_a(x)$53 better than $F_a(x)$54 over three RH cycles (Simao et al., 2015). The stated modulus change, $F_a(x)$55, emphasizes that RH is not a secondary test variable but a primary state variable in FACT mechanics.

At the cell-wall scale, Amando de Barros et al. combine micropillar compression and DIC on Norway spruce tracheids to resolve MFA-dependent stiffness and failure. Pillars with height $F_a(x)$56m and diameter $F_a(x)$57–$F_a(x)$58m are tested at MFA $F_a(x)$59, $F_a(x)$60, $F_a(x)$61, and $F_a(x)$62 under displacement-controlled compression to $F_a(x)$63m at nominal strain rate $F_a(x)$64 s$F_a(x)$65. DIC-based measurements at $F_a(x)$66 kV imaging yield $F_a(x)$67 GPa and $F_a(x)$68 GPa for MFA $F_a(x)$69, $F_a(x)$70 GPa and $F_a(x)$71 GPa at $F_a(x)$72, $F_a(x)$73 GPa and $F_a(x)$74 GPa at $F_a(x)$75, and $F_a(x)$76 GPa and $F_a(x)$77 GPa at $F_a(x)$78. Low-MFA pillars show fibril-aligned kink bands, whereas high-MFA pillars show shear-related catastrophic failure. Continuous $F_a(x)$79 kV exposure causes extensive surface shrinkage and more than $F_a(x)$80 drop in $F_a(x)$81 and $F_a(x)$82 versus “no-beam” measurements (Barros et al., 12 Jun 2025). This result directly connects imaging protocol to apparent cell-wall mechanics.

5. Chemical disassembly and interfacial interactions

FACT also includes chemically driven fibril-state control. In TEMPO-mediated oxidation of sugarcane bagasse cellulose pulp, the catalyst system consists of $F_a(x)$83 mmol TEMPO and $F_a(x)$84 mmol NaBr per gram of cellulose, with three NaClO loadings: SC-5 at $F_a(x)$85 mmol NaClO/g cellulose, SC-25 at $F_a(x)$86 mmol NaClO/g, and SC-50 at $F_a(x)$87 mmol NaClO/g. The reaction is maintained at $F_a(x)$88, temperature $F_a(x)$89C, and total reaction time $F_a(x)$90 h, then quenched with ethanol and washed until conductivity plateau. The degree of oxidation rises from $F_a(x)$91 mmol COO$F_a(x)$92/g for SC-5 to $F_a(x)$93 mmol/g for SC-25 and $F_a(x)$94 mmol/g for SC-50, with zeta potentials $F_a(x)$95 mV and $F_a(x)$96 mV for the more oxidized states. AFM analysis of $F_a(x)$97 individual nanofibrils shows aggregated bundles $F_a(x)$98 nm in SC-5, but individualized nanofibers of average length $F_a(x)$99–$E_{diss}(x)$00 nm with a major width peak at $E_{diss}(x)$01 nm in SC-25 and SC-50, plus sub-elementary populations at $E_{diss}(x)$02 nm and $E_{diss}(x)$03 nm assigned to single-chain and double-chain nanofibers (Silvestre et al., 2020). First-principles calculations quantify the accompanying energetic changes: in pristine CNFs, $E_{diss}(x)$04 eV/unit-chain and $E_{diss}(x)$05 eV/unit-chain, whereas at $E_{diss}(x)$06 carboxylation and $E_{diss}(x)$07, $E_{diss}(x)$08 weakens from $E_{diss}(x)$09 to $E_{diss}(x)$10 eV and $E_{diss}(x)$11 from $E_{diss}(x)$12 to $E_{diss}(x)$13 eV. The stronger relative weakening of interchain O–H$E_{diss}(x)$14O hydrogen-bond interactions is presented as the mechanistic basis for liberation of single and double chains.

Interfacial adsorption assays based on cellulose nanocrystals extend FACT to colloidal formulation science. Cotton linters hydrolyzed with $E_{diss}(x)$15 (w/v) H$E_{diss}(x)$16SO$E_{diss}(x)$17 at $E_{diss}(x)$18C for $E_{diss}(x)$19 min yield lath-shaped particles with length $E_{diss}(x)$20 nm, width $E_{diss}(x)$21 nm, thickness $E_{diss}(x)$22 nm, and $E_{diss}(x)$23 mV at $E_{diss}(x)$24 wt\% and pH $E_{diss}(x)$25. The cationic surfactant TEQ has $E_{diss}(x)$26 mV, self-assembles into unilamellar vesicles of $E_{diss}(x)$27–$E_{diss}(x)$28 nm diameter at $E_{diss}(x)$29 wt\%, multivesicular structures of $E_{diss}(x)$30 nm–$E_{diss}(x)$31m at $E_{diss}(x)$32 wt\%, and large bilayer stacks above $E_{diss}(x)$33 wt\%. Continuous Variation plots of Rayleigh ratio $E_{diss}(x)$34 and hydrodynamic diameter $E_{diss}(x)$35 show maxima at $E_{diss}(x)$36, whereas electrophoretic mobility crosses zero near $E_{diss}(x)$37, indicating charge-driven neutralization. The binding constant $E_{diss}(x)$38 extracted from a Langmuir-type isotherm is stated to fall typically in the $E_{diss}(x)$39–$E_{diss}(x)$40 M$E_{diss}(x)$41 range for strong electrostatic adsorption (Oikonomou et al., 2017). Here FACT functions as a sensitive surrogate assay for deposition on cotton.

Ion-specific perturbation of cellulose provides a further chemical axis. Replica-exchange MD of cellotetraose in NaCl solution and NPT-LD simulation of a cellulose I$E_{diss}(x)$42 fibril show preferred Na$E_{diss}(x)$43 contacts at O$E_{diss}(x)$44, O$E_{diss}(x)$45, and O$E_{diss}(x)$46. For the solvated tetramer at $E_{diss}(x)$47 K, the reported probabilities are $E_{diss}(x)$48, $E_{diss}(x)$49, and $E_{diss}(x)$50, with O$E_{diss}(x)$51–O$E_{diss}(x)$52–Na$E_{diss}(x)$53 bridging at $E_{diss}(x)$54. For the fibril surface at $E_{diss}(x)$55 K, $E_{diss}(x)$56, $E_{diss}(x)$57, and $E_{diss}(x)$58, while core hydroxymethyl populations shift from tg/gt/gg $E_{diss}(x)$59 in pure water to $E_{diss}(x)$60 with NaCl (Bellesia et al., 2013). The paper interprets this as disruption of native intrachain hydrogen bonds and promotion of alternative intersheet interactions.

6. Process analytics, limitations, and technological significance

FACT is not limited to static imaging or ex situ assays; it also includes process-relevant orientation dynamics. In the flow-stop POM method for semi-dilute CNF dispersions, a thin channel is placed between crossed polarizers at $E_{diss}(x)$61 and $E_{diss}(x)$62 relative to the flow axis, illuminated by a $E_{diss}(x)$63 nm laser over a $E_{diss}(x)$64 mm spot, and imaged at $E_{diss}(x)$65 fps while fast three-way valves arrest the flow quasi-instantly. The birefringence-derived signal obeys

$E_{diss}(x)$66

and, under a purely Brownian dilute-rod model,

$E_{diss}(x)$67

Piecewise fitting reveals two rotary-diffusion regimes, with $E_{diss}(x)$68–$E_{diss}(x)$69 over $E_{diss}(x)$70–$E_{diss}(x)$71 s and $E_{diss}(x)$72 over $E_{diss}(x)$73–$E_{diss}(x)$74 s. The fast process depends on prior deformation history, and de-alignment is faster in shear-dominated flow than in pure extensional flow (Rosén et al., 2018). The same report explicitly proposes FACT quality criteria such as maximum $E_{diss}(x)$75 at design shear below $E_{diss}(x)$76, $E_{diss}(x)$77, and steady-state $E_{diss}(x)$78 curves within $E_{diss}(x)$79 of a validated reference dispersion.

The methodological breadth of FACT also creates recurring limitations. In ML-based NegC-SEM analysis, reliable performance requires high-contrast images and at least $E_{diss}(x)$80 px across the narrowest fibril; segmentation errors in low-contrast or highly overlapped regions produce extraneous skeleton spurs, junction-point handling depends on SST and SSF tuning by trial-error, and width resolution is limited to $E_{diss}(x)$81 the pixel size (Baez et al., 8 Sep 2025). In cell-wall micropillar compression, uncontrolled electron-beam exposure degrades pillar integrity and can explain data scatter and mechanical underestimation in earlier studies, leading to the recommendation that continuous SEM imaging be limited to $E_{diss}(x)$82 kV with beam currents $E_{diss}(x)$83 pA and short acquisition windows (Barros et al., 12 Jun 2025). In neutron dark-field tomography, only three scattering directions are probed, attenuation can bias dark-field unless corrected, and the spatial resolution is limited by detector pixel size, so complementary SAXS or SEM remains necessary below the $E_{diss}(x)$84 nm regime (Busi et al., 2024). In humidity-sensitive mechanics, moisture strongly modulates $E_{diss}(x)$85 and adhesion, so RH must be standardized or at minimum reported alongside mechanical values (Simao et al., 2015).

These limitations do not diminish the integrative value of FACT; rather, they define the conditions under which its descriptors are interpretable. In flax-fibre defects, localized porosity spikes and MFA misalignment are linked to stress concentrators and weakened fibre-matrix adhesion in composites (Quereilhac et al., 2023). In single-fibre AFM–CLSM–SEM mapping, the output is a multi-parametric fibre “fingerprint” that feeds improved finite-element or network models (Auernhammer et al., 2020). In oxidation-controlled disassembly, measured $E_{diss}(x)$86 and $E_{diss}(x)$87 are presented as inputs for process-simulation modules and for targeting diameter distributions under given ionic strength and oxidation conditions (Silvestre et al., 2020). Collectively, the reported literature suggests that FACT is becoming a general framework for converting fibril-resolved measurements into design rules for composites, membranes, paper networks, foams, and formulation-relevant cellulose interfaces.