Rheological and rheo-optical behaviors of nanocellulose suspensions containing unfibrillated fibers

We investigated the relationship between the concentration of nanosized fibers and the rheological and the rheo-optic behaviors of CNF suspensions containing unfibrillated fibers prepared by a wet refinement system (Water Jet System). The CNF suspensions behaved as a gel in which entropic elasticity was dominant, and the entanglement among fibers would be enhanced with the nanosized fiber volume fraction. The aggregates of the nanosized fibers elongated in the flow direction and deformed into elliptical shape with the applied shearing. However, the effect of these shape changes of the aggregates was barely observed on the viscosity curve. We speculate that this is due to Newtonian flow of the unfibrillated fibers (> 25% by volume) playing a significant role in the overall flow behavior of the CNF suspensions. In conclusion, for the CNF suspensions containing poorly nanosized fibers, we speculated that the nanosized and unfibrillated fibers were largely responsible for the linear response under micro-deformation and the nonlinear response such as under shear flow, respectively.


Introduction
Recently, cellulose has been reevaluated as an alternative biomass material to fossil materials because of its recyclability. In addition, the cellulose nanofibers (CNFs) have attracted attention due to recent advances in nanotechnology. CNFs are literally cellulose fibrillated to a nanoscale, and is expected to be used in a wide variety of fields, such as reinforcement materials for composites, heat insulators, optical and electronic devices, and medical materials, due to its light weight, high strength, and low linear thermal expansion coefficient (Eichhorn et al. 2010;Isogai et al. 2011;Klemm et al. 2011;Siqueira et al. 2010).
The manufacture of CNFs can be roughly divided into chemical and mechanical treatments. As a chemical treatment, sulfuric acid has long been used in their manufacture. In this method, cellulose fibers are treated at 30 ~ 40 °C for a long time to hydrolyze and remove the non-crystalline portions in order to obtain microcrystalline cellulose with the degree of polymerization of about 200 ~ 250. The product is called cellulose nanowhiskers. Recently, a catalytic oxidation method using TEMPO (2,2,6,6-tetramethylpiperidine-1-oxyl radical) has been proposed; the TEMPO catalytic oxidation was reported in the 1990s as a reaction that can regioselectively convert the C6 other from opposite nozzles (injection speed: about 700 m/s, collision pressure: about 200 MPa), and the cellulose fibers are miniaturized only by the penetration force of the water between them. Although the force of collision is weaker than that of a covalent bond, the interaction between fibers is cleaved, thus the fibers can be nanosized without damaging the fiber surface. In addition, since all of these methods use only water and raw materials, the inherent properties of the raw materials, such as the degree of polymerization and crystallinity, can be maintained.
In addition, cellulose nanofibers have been produced by chemo-mechanical methods, which is a combination of chemical and mechanical treatment. For example, in the case of mechanical fiber disintegration process alone such the high-pressure homogenizers, pretreatment with enzymes is sometimes used because the fibers can clog the homogenizer and the energy consumption is high (Charani et al. 2013;Hayashi et al. 1998;Henriksson et al. 2007;Pääkkö et al. 2007). Hayashi (2008) added endoglucanase to cellulose when mechanically disintegrating it under conditions where the enzyme could act. The results showed that the mechanical disintegration alone, such as ball milling, produced cracks and partial fibrillation along the axial direction of the cellulose, but many aggregated parts remained, while enzymatic treatment together resulted in fibrillation down to the microfibril unit. In addition, a chemical treatment method has been developed to modify cellulose by carboxymethylation (Wågberg et al. 1987(Wågberg et al. , 2008) as a pretreatment for mechanical disintegration, which can reduce the energy consumption of the disintegration process to an industrially acceptable level. The recovery of the solvents is pricy, however, because these chemical treatments are efficiently performed in non-aqueous medium. More recently, a method of physically attaching carboxymethylcellulose chains onto wood pulp has been developed (Ankerfors and Lindström 2009). Fibrillation of their pulps in the microfluidizer-type high-pressure homogenizer was found to result in the same consistencies as many of the commercially available cellulose nanofibers (Chen et al. 2015;Su et al. 2019). The manufacturing process was also deemed to be energy efficient, as it lacked the need for mechanical pre-treatment, which is often a prerequisite for the production of many existing preparations of nanocellulose.
CNFs prepared from various types of cellulose sources using the above-mentioned process have been deployed in various industries due to their valueadded properties, and studies have been conducted focusing on the rheological properties of the CNF suspensions (Bercea and Navard 2000;Boluk et al. 2011;Charani et al. 2013;Iotti et al. 2011;Iwamoto et al. 2014;Naderi et al. 2014;Pääkkö et al. 2007;Tanaka et al. 2020;Yamagata and Miyamoto 2021;Yamagata et al. 2020). In other words, it is important to understand how to control the rheological behavior of the CNF suspensions in order to fully exploit the potential of the material.
Factors that affect the rheological behavior of the CNF suspensions include the chemical composition of the fibers, including the surface functional groups, as well as the fiber shape and length distribution. Different starting materials have different chemical compositions, which results in a different fiber stiffness and interaction forces between the fibers. Therefore, even if the shape and the length distribution of the fibers are the same, the rheological behavior will be different depending on the fiber type. In addition, the viscosity and elastic modulus increase with the nanofiber concentration even if the starting material is the same (Charani et al. 2013;Iotti et al. 2011;Pääkkö et al. 2007;Yamagata et al. 2020).
Although the viscosity curve of nanocellulose suspensions exhibits the shear thinning behavior, the behavior is not monotonically decreasing, but rather shows complex behavior where the plateau of the apparent viscosity appears in the intermediate shear rate regions (Iotti et al. 2011;Li et al. 2015;Takai-Yamashita et al. 2021;Yamagata and Miyamoto 2021). According to the work of Li et al. (2015), the reason for the appearance of the apparent viscosity plateau in the intermediate and high shear rate regions is that the former is due to the formation of a more entangled network structure of oriented fibers by the applied shearing, and the latter is due to the disruption of most of the entangled network structure and the formation of a welloriented structure. Iotti et al. (2011) also reported that the reason for the appearance of the viscosity plateau in the intermediate shear rate region is the shear-induced structure formation, i.e., since CNFs are long and thin fibers with a very high specific surface area covered by hydroxyl groups, the fibers are organized by being very close to each other in the structure, helping to increase the viscosity. Karppinen et al. (2011Karppinen et al. ( , 2012 observed the change in floc structure in the suspensions containing of microfibrillar cellulose or in which polymers were added to them under shear flow, using a rheometer equipped with the concentric cylinder measuring geometries and the camera. They concluded that the plateau region appearing on the flow curve was caused by the change in the floc structure. We also discussed the reason for the appearance of the viscosity plateau of the CNF suspensions with fiber length of about 830 nm based on the results of the rheo-SALS measurements. CNFs form an almost isotropic circular (three-dimensionally spherical) aggregated structure in the low shear rate region. As the shear rate increases and reaches a critical value, the aggregation elongates in the flow direction and deforms into an anisotropic elliptical shape in order to reduce the flow resistance. We speculated that transient shear stress is generated at this time, suppressing the shear thinning behavior and appeared in the viscosity plateau region. In other words, the appearance of the viscosity plateau is considered to be caused by the organization and macroscopic structural changes of the nanosized CNFs.
However, there seems to be only a few quantitative and detailed studies on the rheological behavior of nanocellulose suspensions containing some amount of unfibrillated fibers. In particular, in the case of mechanical treatment, unlike chemical treatment, the shape, the diameter and the length of the produced CNFs are not uniform. From the viewpoint of reducing the production cost of CNFs, it is important to find practical applications for the CNFs that contain unfibrillated fibers. For this purpose, it is necessary to evaluate the physical properties of CNFs containing unfibrillated fibers. In this study, we investigated the relationship between the CNF nanosized volume fraction and the rheological behavior of the CNF suspensions, which contain unfibrillated fibers, produced using a water-jet type wet refinement system. Furthermore, macroscopic structural changes of the CNF suspensions under shear flow were also discussed based on rheo-optic measurements.

Preparation of fibrillated cellulose
Low-substituted hydroxypropyl cellulose (LODI-CEL®, Shin-Etsu Chemical Co., Ltd., hereinafter referred to as HPC) as the starting material, was added to ion-exchange water to prepare a dispersion of 4 wt%. A water-jet type wet refinement device (Ultimaizer HJP25005, Sugino Machine Limited, Japan) was used as a mechanical refinement device, and the prepared dispersion was injected through 2 diamond nozzles (nozzle diameter: 0.13 mm) facing and collided with each other under a processing pressure of 180 MPa at room temperature (about 20 °C). The number of collisions was increased from 20 to 60× to obtain the nanosized CNF suspensions.

AFM observation
CNFs diluted with water to about a 0.001 wt% concentration was dropped onto a mica plate in a volume of about 1 mL, heated at 100 °C for 10 min, then airdried at room temperature for a day and night. Fibers that adhered to the mica surface were obtained by this process. The samples were observed by an AFM system (Tosca 400, Anton Paar GmbH) in the tapping mode.

Particle size distribution measurement
The nanosized fibers were considered as spherical particles, and the particle size distribution was measured by a laser diffraction/scattering particle size analyzer (LS 13 320, Beckman Coulter, Inc.) equipped with a universal liquid sample module. In the actual operation, the suspensions (fiber concentration: 4 wt%) immediately after mechanical treatment were diluted 20× with distilled water, followed by ultrasonicated for 2 min to prepare uniform suspensions. Then, the measurement was started by dropping the diluted suspensions into the module containing distilled water in the apparatus until a measurable intensity was obtained. The fiber concentration in the module was approximately 0.01 ~ 0.05 wt% when the diluted samples were dropped.
The cumulative particle in the normal measurement range of 0.017 ~ 20,000 µm were set as 100%. As described below, among them, particles with a diameter of less than 1 µm were defined as "apparently nanosized fibers", and the volume fraction of them was automatically analyzed by the software, and the obtained value was defined as the nanosized fiber volume fraction (Vn).

Rheometry
Rheological measurements were carried out at 20 °C by a stress-controlled rheometer, MCR-102 (Anton Paar GmbH). Parallel plates of 25 mm diameter with a 0.5 mm gap between them were used for the viscoelastic measurements. A quartz plate of 43 mm diameter with a gap of 0.5 mm was used to measure the steady flow. The measurement program is shown in Fig. 1, where the apparent viscosity (̇) was measured (1 ~ 5 min) until the steady state was reached while increasing the shear rate in a stepwise manner. The shear rate varies with the radius (distance from the center) when using a parallel plate geometry. The shear rate at the outermost circumference of a parallel plate was often used, but in this case, the shear rate at 2/3 of the way from the center to the circumference, i.e., (43/2) × (2/3) = 14.3 mm, was used. This reason is the laser beam is irradiated at a position 15 mm from the center to the periphery in the SALS measurement. The effects of the sample evaporation were minimized by using a protective hood. A small angle light scattering (SALS) system was attached to the above-mentioned rheometer and light scattering measurements were synchronously performed along with the viscosity measurements. Figure 2 shows a schematic diagram of the apparatus. A laser beam with a wavelength λ = 658 nm was incident parallel to the axis of rotation from the top of the rheometer onto a sample sandwiched between a transparent quartz plate and a glass plate. θ is the scattering angle, ϕ is the dimension of the scattering image. The polarization of the incident light and the detected light were both perpendicular to the flow direction, the exposure time was 8.5 ms, and the scattered light was directly recorded on the CCD camera chip at the bottom of the apparatus. The projection distance (camera length) l was 15 cm, the diameter of the beam stopper was 1 mm, and the gap between the plates was 0.5 mm.

Results and discussion
Size of CNFs Figure 3 shows AFM images of the samples obtained after 20 and 60 collision times. The 20 collision times sample ( Fig. 3(a)) showed many fibers longer than several tens of microns, whereas the 60 collisions ( Fig. 3(b)) showed few fibers longer than 10 μm. In addition, fibers with a width of more than 1 μm were observed in the 20 collisions sample.
The most common method to measure the length and the width of nanocellulose fibers is to use the transmission electron microscopy (TEM) or the atomic force microscopy (AFM) observation. This method is based on the measurement of the length and the width of each of several hundred fibers observed. We also tried to measure the fiber length and width from the AFM images, but it is difficult to calculate them easily because it requires a lot of labor and the fiber length distribution of our sample is assumed to be very wide, which may lead to arbitrary evaluation depending on the observation site. Several methods have also been proposed to estimate the length and spectral ratio from the viscosity of dilute dispersions of nanocellulose (Tanaka, et al. 2014;Iwamoto et al. 2014), but it is not clear whether they can be adapted to a system where both nanosized and unfibrillated fibers are mixed, such as our samples. Recently, the shape of the fibers is considered as spherical and the size of the fibers was calculated for convenience using a laser diffraction/scattering particle size analyzer (Kojima et al. 2013(Kojima et al. , 2015. Therefore, we decided to estimate the size of CNFs for convenience using the laser diffraction/scattering system as shown below. It is difficult to distinguish whether the calculated CNF size represents a single fiber or the aggregation of nanocellulose fibers, however, since nanocellulose fibers tend to aggregate in water. The laser diffraction/scattering method is a measurement method to obtain the particle size distribution as the spherical equivalent diameter from the diffraction and scattering theories, using that the angular dependence of their diffracted and scattered light intensities is a function of the particle size. Therefore, it is not possible to accurately determine the length of a sample such as CNFs, but possible to obtain information, as a particle size distribution, based on the behavior (spread and size) in water reflecting the fiber length.
In order to obtain the actual light intensity, the Fraunhofer diffraction theory (Eq. (1)) is applied for r >> λ and the Mie scattering theory (Eq. (2)) is applied for r ≤ λ, based on the relationship between the incident wavelength λ and the particle radius r.
where I(θ) is the scattering intensity at scattering angle θ, I 0 is the intensity of the incident light, λ is the wavelength of the incident light, d is the particle size, and J 1 is the Bessel function of the first kind.
where R is the distance from the scattering particle, i 1 is the intensity function of the s-polarized component (perpendicular to the polarization direction of the incident light or the scattering plane), and i 2 is the intensity function of the p-polarized component (parallel to the polarization direction of the incident light or the scattering plane). (1) The instrument used in this study incorporates the two measurement principles described above. This instrument also employs "Polarization Intensity Differential Scattering; PIDS", which solves the problem that the light scattering pattern becomes Rayleigh scattering and the scattering intensity decreases when r << λ, making the measurement difficult. This is a measurement technique used on Mie scattering theory, which is based on the property that light scattering by small particles has different scattering intensities with respect to polarization. In other words, the main light source (780 nm) is used for particles with large size distributions, while the PIDS light sources (450, 600 and 900 nm) are used to measure particles smaller than 1 μm, and both detect particles from 0.04 to 2000 μm. Since the PIDS measurement data are added to the same deconvolution matrix used for particle size distribution measurement by separating and detecting the polarization components of the Mie scattering theory, the relative volume of particles in each size fractionation channel is specified by the solution of this matrix. Thus, the 2 measurement results are completely integrated through a calculation process that compensates for each other's weak detection range in a single matrix equation, therefore enabling the measurement of a wide range of particle size distributions from tens of nm to thousands of μm. Figure 4 shows the particle size distributions of the CNF suspensions with the different number of collisions. The particle size distribution of the unfibrillated fiber suspensions without mechanical treatment is also shown in Fig. 4. The average particle size of unfibrillated fibers was a single peak centered at about 100 μm, and no particles size smaller than 4 μm was observed. On the other hand, all of the mechanically treated samples showed a bimodal distribution, with a single peak centered at about 0.3 μm and several peaks coexisting between 5 and 200 μm. The Application Note of a trading company specializing in particle size analyzers (Nikkai-Bios) is reported that the particle size distribution of cellulose nanofibers suspension containing nanocrystals smaller than 1 μm is bimodal, with a single peak topping out at 50 nm due to the nanocrystals, and several peaks coexisting between 15 and 400 μm due to the cellulose nanofibers by measured the laser diffraction/ scattering system with on the PIDS method. It is also reported that the particle size distribution of cellulose nanofibers tends to be detected as a particle size distribution with shoulders and multiple peaks rather than a clear normal distribution due to the high aspect ratio of them. Considering them, it can be inferred that the single peak centered at about 0.3 μm, which appeared only in the samples treated with mechanical treatment, due to the nanosized fibers, and the group of peaks from 5 to 200 μm due to the unfibrillated HPC. The peak (volume fraction) near about 0.3 μm  Then, fibers with a particle size of less than about 1 μm were defined as "apparently nanosized fibers" and their volume fractions were calculated as nanosized fiber volume fraction (Vn). The relationship between the nanosized fiber volume fraction (Vn) and the number of collisions is shown in Fig. 5. Vn was 32.7% for the 20 collisions, however, Vn increased with the number of collisions, reaching 74.8% for the 60 collisions. Paradoxically, this means that a little more than 20% of the unfibrillated fibers remain even after 60 collisions. At the same time, the cumulative medium diameter (D 50 ) of the particles decreased by more than one order of magnitude, from about 10 μm at 20 collisions to about 0.3 μm at 60 collisions. These results indicated that a higher number of collisions is more effective for the nanosizing of the fibers. These trends were consistent with the results of Kojima et al. (2013), that the longer the milling time and the higher the rotation speed when cellulose powder suspension (median particle size: 41.4 μm) was treated by the ball mill, i.e., the more severe the mechanical treatment processing conditions, the smaller the median particle size measured by the laser diffraction/scattering measurement.

Viscoelastic behaviors of CNF suspensions
All of the nanosized CNF suspensions prepared by mechanical collisions from 20 to 60× were qualitatively considered to be gels, because they had poor flowability and retained their shape for a while even when the suspensions were left to rest on a horizontal surface. In order to evaluate the properties of the suspensions more quantitatively, dynamic viscoelasticity measurements were performed.
Before measuring the frequency dependence of the viscoelasticity of the CNF suspensions, we performed the strain sweeps to confirm the linear region. The results are shown in Fig. 6a. Both the storage modulus Gʹ and the loss modulus Gʺ increased with the number of collisions, i.e., with the nanosized fiber volume fraction Vn. The critical strain for the transition from linear to nonlinear was about 3%, which was almost independent of Vn. Based on these results, the frequency sweeps were measured with the strain fixed at 1%. The results are shown in Fig. 6b. All the CNF suspensions have Gʹ >> Gʺ over the entire measurement frequency range, and both Gʹ and Gʺ are almost independent of the frequency (Gʹ ≈ ω 0 , Gʺ ≈ ω 0 ), thus it is considered that they are an elastic-dominated gel, and the elasticity is attributed to the nanosized fibers. The nanosized fiber volume fraction, Vn, can be converted into the weight concentration In other words, if more than 1.3 wt% of the nanosized fibers are dispersed, the system shows elastic properties. The rheological behaviors of the CNF suspensions may be related to chemical properties such as surface functional groups of fibers or physical properties such as entanglement among fibers. In general, the bonds in cellulose fibers include β-1,4 bonds and intramolecular hydrogen bonds, and the water-jet type refinement system that we used can selectively cut only the hydrogen bonds, so it is possible to nanosize cellulose without damaging its crystal structure or surface state (Ogura 2017). As a result of the zeta potential measurement by electrophoretic light scattering method using Litesizer 500 (Anton Paar GmbH) for samples diluted 100× with 1 mM KCl solution, zeta potentials of all samples treated with 0-60 collision times showed about − 5 mV, and no significant difference was observed among them (Fig. S1). Zeta potential values are generally comparable to that of pulp (Onabe and Nakano 1970), which has a large number of hydroxyl groups on its surface, so it is considered that the chemical properties of the fiber surface are not changed by mechanical treatment. Therefore, it can be inferred that the rheological behavior of CNFs are due to the physical interplay rather than the chemical properties of the fiber surface.
Then, the concentration of nanosized fibers in the sample with 30 collision times (Vn = 47.4%) is about 1.9 wt%, and the elastic modulus Gʹ is about 10 3 Pa. These nanosized fiber concentrations and modulus Gʹ are comparable to those of a 1.5 wt% suspension of CNFs (fiber length: 827 nm, aspect ratio: 243) prepared by TEMPO oxidation (Yamagata et al. 2020) and a 2% dispersion of microfibrillated cellulose (MFC) prepared by high-pressure homogenization by Iotti et al. (2011). Therefore, although qualitative, the nanosized fiber volume fraction, Vn, estimated by laser diffraction/scattering measurements is might be a reasonable value.
The relationship between the elastic moduli of CNFs and the concentration of nanosized fiber converted from Vn was discussed. As already mentioned, since Gʹ and Gʺ are almost independent of the frequency, the elastic moduli at ω = 1 rad/s are considered as pseudo-elastic moduli (G p ʹ, G p ʺ). Both G p ʹ and G p ʺ increased with the concentration of nanosized fibers as shown in Fig. 7. Many papers have been reported the dependence of the elastic modulus on the concentration in fiber suspension systems (Guenet 2000;Tamai et al. 2004;Tatsumi et al. 2002). In most cases, a scaling law is observed between the modulus G and the concentration c, as shown in Eq. (3).
Here, A is a constant that depends on the modulus and the aspect ratio of the fiber itself, independent of the concentration c. α is an index that reflects the network structure, and is reported to be about 2.25 in cellulose dispersion systems with a few μm fiber length regardless of the fiber type (Guenet 2000). It was found that the elastic moduli of the suspensions increased with the concentration of nanofibers, i.e., depended on the concentration of them. The values of α for G p ʹ and G p ʺ were 1.5 and 1.7, respectively, which were lower than 2.25. Ramzi et al. found that the elastic modulus of the agarose cosolvent system can be represented by 2 straight lines with different exponents, α = 1.5 for the agarose concentration above 20 g/L and α = 2.25 for the agarose concentration below 20 g/L (Ramzi et al. 1998). According to morphological observations, agarose gels are composed of nearly linear fibrillar arrays and are known to be intrinsically rigid materials (Sugiyama (3) G = Ac Fig. 7 Pseudo-plateau moduli G p ʹ and G p ʺ at 1 rad/s and yield stress σ y of CNF suspensions as a function of concentration of nanosized fibers et al. 1994). In light of Jones and Marque's theory (Jones and Marques 1990), it is concluded that the agarose gel is rigid and has an enthalpic elasticity in low concentrations, but exhibits an entropic elasticity due to the binding sites between the fibers being disordered and flexible at high concentrations (Ramzi et al. 1998). Considering these facts, the gelation of suspensions of the mechanically fibrillated CNFs is attributed to the increased entanglement of the nanosized fibers, and the entangled junctions are disordered and flexible, suggesting that the system as a whole is gel dominated by an entropic elasticity.
The stress σ obtained by strain sweep measurement of dynamic viscoelasticity was plotted against strain γ (Fig. S3), and the stress at the point of deviation from the straight line of gradient 1 in the low strain region was calculated as the apparent yield stress value σ y . The obtained yield stress value σ y was plotted against the concentration of nanosized fibers as in the case of pseudo-elastic moduli (Gpʹ, Gpʺ) (Fig. 7). As a result, the scaling law shown in Eq. (4) was found to observe between the yield stress value σ y and the concentration of nanosized fibers (α = 1.47, A' is a constant).
In other words, the entanglement of fibers increases and the three-dimensional network structure becomes stronger with the concentration of nanosized fibers, and as a result, the stress (yield stress value) required for flow also increases. Figure 8 shows the viscosity curves as a function of the shear rate for CNF suspensions prepared by 20 to 60 collisions. The apparent viscosity (̇) increased with the number of collisions in the entire measured shear rate range. (̇) at ̇ = 0.1 s −1 was higher than 10 5 mPa.s for all samples, but (̇) linearly decreased with the increasing ̇ . At ̇ = 1000 s −1 , (̇) decreased by three orders of magnitude to 10 2 mPa.s. These viscosity curves can be approximated by the power law shown in Eq. (5).

Shear flow behaviors of CNF suspensions
where, k and n are constants that have no physical meaning, but n is also called the power exponent coefficient, and past papers have shown that n ≤ 0.8 for polymer solutions and melts, and n ≈ 1 for cohesive particle dispersion systems and liquid crystals. The n-values of the CNF suspensions ranged from 0.84 to 0.93. When the n values were plotted versus the concentration of nanosized fibers, the n values increased with the concentration as shown in Fig. 9. This suggested that as the number of nanofibers increased, the aggregation progresses with the entanglement of the  fibers and the viscosity in the near-stationary state also increased, but when the aggregated structure is destroyed by the applied shearing, the viscosity decreases with the shear thinning behavior becoming more pronounced.
On the viscosity curves of CNF suspensions containing unfibrillated fibers (more than 20% relative to the cellulose fiber concentration) was not observed the plateau region of the apparent viscosity, but only a monotonous decreasing behavior. This is clearly different from the flow behavior of almost or completely nanosized cellulose fiber suspensions. Thus, the change of the macroscopic structural of fiber aggregates applied shearing is also expected to be different. Therefore, we decided to perform Rheo-SALS measurements of the CNF suspensions contain unfibrillated fibers to confirm whether there is any structural change in the CNF aggregations under shear flow.

Rheo-SALS measurements of CNF suspensions
Small-angle light scattering (SALS) is one of the most widely used techniques to visualize changes in the internal structure of systems under shear flow by synchronizing it with rheological measurements. The intensity distribution of the scattered light caused by the incident primary laser beam is detected by a CCD camera during the light scattering. Various polarization components can be observed by changing the polarization directions of the incident light and the detector when making the measurements. The case where the polarization directions of the polarizer and the detector are orthogonal is called HV, and the case where they are parallel to each other is called VV. Based on the HV and VV scattering, we can obtain information about the optical anisotropy and density fluctuations and the optical anisotropy, respectively. The scattering images are omitted, but only the VV scattering results are shown below, since HV scattering, which indicates anisotropy, was rarely observed in the CNF suspensions. Figure 10 shows the scattering images obtained by the applied shearing from 0.1 to 1000 s −1 to the CNF suspensions prepared by 20 to 60 collision times. The flow direction is from left to right in the scattering image. In our measurement, the position of the laser beam irradiation was shifted upward (toward the center of the geometry), and the alignment was somewhat insufficient. Because all the scattered images were deviated in the same way, however, we judged the relative comparisons were possible and there was no problem in discussing the degree of orientation (anisotropy) of the macroscopic scatterers.
The scattering images showed an almost isotropic circular regardless of the number of collisions at a shear rate of 0.1 s −1 as in the near static state. However, the scattering images transformed from a circular shape to a longitudinal elliptical shape extending perpendicular to the flow direction with the increasing shear rate. At the high shear rate, the elliptical scattering images became smaller with the number of collisions.
The shape and size of the scattered image are inversely related to those of the scatterers. In other words, a larger scattering image means smaller scatterers, and an elliptical scattering image extending along the y-axis means an elliptical scattering image extending along the x-axis (flow direction). We decided to calculate the size of the scatterer from the scattering image, however, since it is too qualitative to discuss the change in the structure of the scatterers from those of the scattering image.
Internal structure change of CNF suspensions under shear flow by Rheo-SALS analysis The observed scattering images were imported into Anton Paar's software, NewSALS ver.2.01, and the scattering intensity was calculated from the scattering vector q, which corresponds to the distance from the center of the transmitted light, and the average value of the light intensity in a specific angular range (± 15°). The scattering intensity of CNFs I(q) was calculated by subtracting those of water calculated in the same way. The scattering vector q can be calculated using Eq. (6), and the effective range of q for this device is 0.3 to 4 μm −1 (Läuger 2006).
where λ is the wavelength of the laser beam (658 nm) and the scattering angle θ was calculated by Eq. (7).
l is the camera length (15 cm) and ϕ is the size (length) of the scattering image. Figure 11(a) shows the scattering curves for the flow direction of 0° to the x-axis with 60 collisions, and parts of the curves at the shear rate of 0.1 ~ 1000 s −1 are extracted. In addition, the y-axis is shifted upward and downward by multiplying the scattering intensity by an appropriate factor (0.01 ~ 50), since the scattering curves overlap and are difficult to understand. A plateau appeared around q = 0.4 ~ 0.7 μm −1 at the low shear rates of less than 100 s −1 . However, the plateau disappeared with the increasing shear rate, a maximum was observed Here I 0 is the scattering intensity when extrapolated to q = 0, and R g is the radius of gyration which is a measure of the size of the scatterers. R g can be calculated from the slope of the straight line (red dashed line in Fig. 11(b)) when lnI(q) is plotted versus q 2 from Eq. (8), and the slope increased with the shear rate. The Guinier approximation is said to be effective when the product of the scattering vector (8) q max at the right end of the linear approximation region of the lnI(q) − q 2 curve and the calculated R g is approximately 1.3 or less (Putnam et al. 2007;Zheng and Best 2018). In our experimental systems, q max ·R g ≈ 0.8 ~ 1.7, which is a rather high value, but the measured values slightly deviate upward from the approximate line, indicating that R g may be underestimated. We believe that the variation of R g with the shear rate and the number of collisions can be well discussed, however, because the characteristic known as the Guinier region in which the scattering intensity rapidly decreases from the plateau to the wide-angle region are observed on the scattering curves. The shear rate dependence of the radius of gyration of the major and minor axes of the scatterers, R g , obtained from Eq. (8) is shown in Fig. 12. The • and ○ plots in the Fig. 12 show the R g of the scatterers obtained from the scattering intensity in the angle range of ± 15° with respect to the x-and y-axis, respectively. In other words, • corresponds to the magnitude of the short axis of the scatterer oriented perpendicular to the flow direction, and ○ corresponds to the magnitude of the long axis of the scatterers along the flow direction.
R g of the scatterers at the low shear rate was 1.2 μm when the number of collisions was 20 (Fig. 11a). This size is smaller than that of the starting material HPC and larger than that of the fibrillated fibers. R g of the aggregates at low shear rates slightly increased with the number of collisions, but did not significantly change.
The particle size distributions for the CNF suspensions shown in Fig. 4. Shows that the particle size of 1 μm corresponds to the valley between the particle size distributions of nanosized fibers and that of HPC or unfibrillated fibers, and that particles of this size are rarely found. Therefore, it can be inferred that R g is not the size of a single nanofiber, but the size of the cluster-like aggregates (or flocs) formed by the entanglement of nanosized fibers. In other words, we speculated that the nanosized fibers entangle to form aggregates of about 1 mm in size and these aggregates might exist as apparently isolated particles.
The range of scattering vectors in the Guinier region is approximately q = 0.5 to 1 μm −1 , which means that micron-sized particles can be observed in this region. The scattering of unfibrillated giant particles should appear at smaller angles, but cannot be observed because the range of scattering vectors that can be measured by the SALS instrument is q = 0.3 ~ 4 μm −1 . Therefore, we speculate that the contribution of unfibrillated particles to the SALS measurements is negligible.
We now discuss the relationship between the size and shape of the aggregates and shear rate. In the case of 20 collisions, the size of the aggregates in the flow direction (0°) and the vertical direction (90°) were almost equal and its shape seemed to be spherical below 10 s −1 . As the shear rate increased, however, the size in the flow direction increased to 1.7 μm while the those in the vertical direction did not significantly change. Thus, the CNF aggregates elongated in the flow direction and deformed into an ellipse with the applied shearing. This shape change became more pronounced with the number of collisions, and the size of the aggregates in the flow direction grew to 2.5 μm for the 60 collisions. Figure 13 shows the dependence of the average aspect ratio of the CNF aggregates at 0.1 ~ 0.25 and 400 ~ 1000 s −1 on the concentration of nanosized fiber. As can be seen from the figure, the average aspect ratio of the aggregates showed a constant value of about 1, which is a small concentration dependence, at the low shear rate, but increased with the concentration of nanosized fiber at the high shear rate. In other words, in the high shear rate region, the aggregates elongate relatively more in the flow direction and the anisotropy increases with the increasing of the concentration of nanosized fiber.
Based on the above results, the mechanically fibrillated nanofibers formed an isotropic spherical agglomerated structure in the static state. In addition, its structure becomes slightly larger with the nanofiber concentration. The aggregates deformed into an elliptical shape elongated in the flow direction with the applied shear rate above 10 s −1 , and its deformation is more pronounced for the nanofiber concentration. Therefore, it is not surprising that the shear thinning behavior is suppressed due to the transient shear stress induced by the shape change of the aggregates around 10 s −1 , and a plateau region appears on the viscosity curves. However, no changes were observed on the viscosity curve. This is probably due to the presence of unfibrillated long fibers (> 25% by volume fraction in the fibers). Fig.  S3 shows the viscosity curve of a 4 wt% suspension Fig. 13 Relationship between aspect ratio of aggregates and concentration of nanosized fibers of HPC, the starting material before the fibrillization. The suspension of the unfibrillated long fibers exhibits a Newtonian flow, and the shear stress σ monotonically increases with the shear rate γ. We speculated that the unnanosized long fibers, which remained more than 25% volume fraction in fibers even after 60 collisions, absorbed the transient stresses caused by the deformation of the nanosized fiber aggregates, resulting in the monotonous shear thinning behavior of the CNF suspensions.

Conclusions
The relationship between the concentration of nanosized fibers and the rheological behaviors of CNF suspensions containing poorly nanosized cellulose fibers produced using a water-jet type wet refinement system was discussed. Based on the frequency sweep, we found that the CNF suspensions behaved like an elastic-dominated gel, and the nanosized fibers were responsible for development of the elasticity. The elastic moduli increased with the concentration of the nanofiber volume fraction (= concentration of nanosized fibers), suggesting that the entanglement of the fibers was enhanced. The pseudo-plateau modulus G p ʹ is proportional to the concentration of nanosized fibers, and its constant α = 1.5, indicating that the entropic elasticity is dominant.
The viscosity curve of the CNF suspensions showed a shear thinning behavior, in which the viscosity monotonically decreased with the shear rate. From the Rheo-SALS simultaneously measured, we found that the aggregates of the nanofibers elongated in the flow direction and deformed into an elliptical shape with the applied shearing. The shape change became more pronounced with the increasing nanosized fiber concentration, but the influence of those changes was hardly observed on the viscosity curves. We speculated that the unfibrillated fibers (> 25% by volume fraction in the fibers), which exhibit a Newtonian flow, are mostly responsible for the flow behavior of the CNF suspensions.
In conclusion, for the CNF suspensions containing poorly nanosized fibers, we speculated that the nanosized and unfibrillated fibers were largely responsible for the linear response under micro-deformation and the nonlinear response such as under shear flow, respectively.