Selective Recrystallization of Cellulose 1 Composite Powders and Microstructure 2 Creation through 3D Binder Jetting 3

Binder jetting is an additive manufacturing technique in which powdered material is sequentially laid down 13 and printed on by an ink binder, in a selective manner, to form a 3D object. Unfortunately work in this area 14 relevant to food materials is largely unpublished, however a typical application of this technique is sugar 15 powder bound by a water and alcohol based ink with optional colour or flavour demonstrated by commercial 16 ventures. In this work we demonstrate the use of a small scale powder layering device under an ink jet printer 17 to test prototype powders prior to producing quantities typically used in commercially available binder jetting 18 machines. Powders comprising predominantly of ball milled, amorphous cellulose were successfully used to 19 create 3D structures when interacting polysaccharides were present in the ink (xanthan gum) and as a 20 proportion of the powder component (glucomannan) by inducing selective recrystallization. These ingredients 21 are categorized as dietary fibre, thus such formulations can be used to create low-calorie 3D printed food 22 designs to be used within food products. 23


25
The first record of applying additive manufacturing (AM) to food materials can be found in a patent by Yang et 26 al.,(2000) describing the creation of a complex 3D cake using an extrusion based layering mechanism. Though 27 filed in the millennium year there is no further record of such a food-specific printer being created by the 28 inventors. In more recent years the interest in applying food materials to additive layer manufacturing 29 techniques from businesses and researchers has boomed, evidenced by the recent Journal of Food Engineering 30 Special Issue on 3D Food Printing (3D Printing and AM are often used as interchangeable terminology). A short 31 time ago it was difficult to find published work on 3D printed food (Lanaro et  published regarding powder processes. In any case, this is indicative of a greater understanding of food 37 materials related to different AM processes forming within the field, but there is still much more that we can 38 learn. 39 As with all AM techniques powder-based AM first involves slicing a 3D computer design into 2D segments to 40 be printed. A powdered substrate is sequentially layered in the print bed and the 3D design is constructed by 41 binding each powder layer as the 2D slices. Binding these layers in horizontal and vertical directions may occur 42 through thermal sintering or application of a binding 'ink'. Food specific examples include 3D sugar hot air 43 sintering on the 'CandyFab' Printer (EvilMadScientist Laboratories: Oskay, 2007), laser sintering of Nesquik TM , 44 sugar and other powders (TNO: Sol et al. 2015) or binding ink-mediated sugar creations with the ChefJet Pro 45 (3D Systems, n.d.). In addition, two patents exist describing potential food-grade powder bed printing 46 methods, with accompanying 'recipes' or formulations which may be used to create a variety of textured 47 products (Von Hasseln et al., 2014; Diaz et al., 2017). Powders used are typically pure sugar or sugar-based, 48 which will have implications on how they effectively bind together. These short chain carbohydrates (like 49 sucrose) or dextrans (such as maltodextrin) are naturally 'sticky' materials and are often used in the food or 50 pharmaceutical industries as binding agents, making them well suited to a binder jetting AM approach 51 (Chumnanklang et al. 2007; Cuq et al. 2013; Diaz et al. 2017). It is of interest to the food industry to look at 52 alternative food powders for these processes as official bodies are advising consumers and producers to limit 53 sugar intake (Public Health England, 2017) but also as a means of expanding the number of potential end-use 54 products that could be obtained. 55 Cellulose, the most abundant polymer on Earth, is a building block of plant cell walls. Humans do not possess 56 the necessary enzymes to digest it, therefore it contributes to the diet as 'dietary fibre' (Cui and Roberts, 2009; 57 Wüstenberg, 2014). Cellulose is composed of β(1-4) diequatorially linked D-glucose molecules that associate 58 through strong hydrogen bonding and pack together in a hierarchical fashion. Within the plant cell wall other 59 polysaccharides are associated with cellulose, such as hemicelluloses, pectin and lignin (Gibson, 2012). It is 60 logical, then, that researchers have found that interactive effects exist between glycans with stereochemically 61 similar β(1-4) linked backbones such as galactomannans, glucomannans and the extracellular polysaccharides 62 xanthan gum (XG) ( presence of ethanol and modification of XG used in the present study this will be discussed briefly in the 70 context of work presented here, however more insight will be given in an intended future publication from our 71 research group. The abundance of cellulose in nature and potential health benefits conferred through 72 consuming dietary fibres make it an interesting material to study as a candidate for use in a powder AM 73 process. 74 By creating 3D structures based upon cellulose using an AM process we envision the possibility of mimicking 75 assemblies relevant to food products, such as the gluten networks present in breadsticks or cookies. The 76 printed structures could act as ingredients in a food manufacturing process by providing a crumbly or brittle 77 textured 'skeleton' onto which coatings are added to enhance palatability. This could be a viable route to 78 create low calorie snack products utilising novel and exciting technology (3D printing) for the food industry. 79 The previous experimentation which led to the 3D application described in the present work (Holland et 80 al. 2018) involved formulating both a usable cellulose powder (CP) and XG based 'ink' component, then testing 81 the interaction between these in a 2D scenario. The results from powder and ink characterisation plus 82 preliminary tests in 2D could then be extrapolated into a 3D layering process with intended layer height of 83 100 µm. Our previous work showed that high speed, short time mechanical attrition of cellulose reduced the 84 particle size (from D[3,2] = 16 to D[3,2] = 13), crystallinity (from 25 % to 6 %) and viscosity average degree of 85 polymerisation (from 1339 to 411) of the original sample. The native cellulose fibrillar structure was also lost, 86 3 rendering a powder with semi-spherical particles of polydisperse size which was passed through a 100 µm 87 sieve to remove any aggregates larger than the intended layer height (see SEM images in Holland et al. 2018).

88
Recrystallization of the amorphous powder could be achieved by imposing different moisture and temperature 89 regimes on the sample. Therefore it was hypothesised that layer adhesion of amorphous cellulose particles 90 through deposition of an aqueous binder and a post-process heating step could be used to selectively induce 91 recrystallization and create controlled structures in a binder jetting AM process. Not discussed in our previous 92 work was the inclusion of other β(1-4) glycans in the powder component to evoke synergistic interactions. 93 During the present work, it became clear that composite powders would be favourable in creating 3D 94 structures, therefore this will be analysed in detail below. 95 Suitability of ink formulations was determined in terms of their 'printability', characterized by the Z number 96 (Derby, 2010). Viscosity (ɳ), surface tension (γ) and density (ρ) parameters were measured for each ink to 97 assess whether they fell within a theoretically 'printable' range (Z = 1-10) out of a nozzle, whose diameter 98 (L) = 21 µm by using the equation: = √ ɳ . Inks tested attained Z numbers higher than this range, falling into 99 the region were satellite droplets may form. Upon experimental testing in an ink jet printer it was found that 100 the satellite droplets produced re-formed with the main droplet during flight out from the nozzle, thus not 101 negatively impacting the printing process. Unmodified and modified XG was included in the ink formulations 102 presented in the previous work, but at relatively low concentrations (0.25 %wt). Similarly to cellulose, ball mill 103 treatment of XG was shown to reduce its molecular weight, signified by a viscosity decrease at any given shear 104 rate for 0.5 %wt aqueous solutions. Simultaneously the ball milling increased the molecular weight distribution 105 within the sample, shown as an elongation of the Newtonian plateau region, whilst giving similar viscosities at 106 high shear for a given concentration of XG. The advantage of this was that the viscosity at printing shear could 107 be maintained without the potential for long polymer chains clogging the printing nozzle. Nor would they be 108 subject to disadvantageous molecular scission as a result of high shear rates experienced at the nozzle during 109 printing. Additionally, a higher concentration of XG was able to be included in ink formulations, thus 110 presenting a higher number of accessible chains which could interact with the cellulose powder. The inks with 111 ball milled XG also attained lower Z numbers. Subsequently, given the evidence for component stoichiometric 112 ratio to impact on synergistic interactions, further formulation of these inks has been carried out and will be 113 detailed below. 114 Building on the outcomes of our previous work (summarised above), the results discussed in this paper 115 translate findings from 2D powder-ink interaction tests into a 3D binder jetting equivalent system. In addition, 116 interactive glycans have been incorporated into the powder component and xanthan gum loading of the ink 117 has been increased to enhance binding and structure creation.  per 12 mL ZrO2 pot was used to mill powders. Powder weight remained at 1 g per pot and milling was 131 undertaken as six cycles of 5 min milling followed by 10 min pause to give 30 min total milling at 800 rpm. 132 Ball milled (BM) samples produced and tested in the printing scenario comprised of pure cellulose (SF300) and 133 composite powders of cellulose with LBG or KGM in admixture prior to milling at 1:1, 7:3 and 9:1 ratios. 134 water as a reference, ≈800 mg of sample was loaded into the sample cell at 20 °C then cooled to 10 °C. A 143 temperature ramp to 95 °C and back to 10 °C at 1 °C min -1 was conducted twice on the sample and data taken 144 from the second scan to ensure the samples had the same thermal history. 145

146
The equipment set up was identical to section 3.2. However, due to the smaller starting particle size of XG, 3 g 147 of powder was added per pot to retain the desired fill volume. In addition to those outlined in the previous 148 work a further milling cycle of 120 min at 800 rpm was introduced to give four potential XG samples to use in 149 ink formulations: Native, 400 rpm 60 min, 800 rpm 60 min and 800 rpm 120min. 150  for all inks tested. The drop watcher software function was used to assess stability and repeatability of 177 droplets during printing. 178 3.6.2 2D and 3D Capability 179 Two bespoke powder substrate holders were designed for use in the Dimatix printer. The first allows powder 180

Ink Formulation and Characterisation
and ink compatibility testing in 2D (single layer printing) and determination of the required ink : powder 181 saturation due to the inclusion of 100-400 µm deep indents. 182 The second enabled printing in 3D (multi-layer printing). A holder was designed in CAD and laser sintered using 183 Polyamide 12. The internal section housed magnets and acted as the print bed whereas the outside border 184 was used to collect excess powder after layer spreading. Aluminium Shims 100 µm thick were cut to create 185 basic 40x20 mm frames. These frames were stacked one by one on the magnetic centre (process 1) with 186 powder spread across the gap with the long edge of a glass microscope slide (process 2) and selected areas 187 printed on in between (process 3), thus creating a 3D system with 100 µm layers. The magnetic base enabled 188 accurate superimposition of subsequent layers and prevented movement of each metal layer during the 189 spreading process. The 2D recessed plate, an annotated diagram of the 3D set up and schematic of the printing 190 process can be seen below in Figure 1. After the printing and layering process was complete, the whole system 191 was place in a convection oven to deliver the thermal energy required for recrystallization of ink-saturated 192 powder. The 3D object was removed from the surrounding unbound powder with tweezers. Samples were mounted on 12.5 mm Φ stubs with carbon tape before being platinum coated using an Emitech 203 SC6740 Sputter Coater (Polaron Ltd, UK). 204

X-Ray Micro Computed Tomography (MicroCT)
205 Samples were mounted on plastic rods using epoxy resin and scanned in 3D using the GE Phoenix Nanotom 206 180NF (GE, Germany) X-ray Computed Tomography System. The scan consisted of collecting 1200 projection 207 images over 360° at an electron acceleration energy of 50 kV and a current of 240 μA. Each projection image 208 was the average of three images (to reduce image noise) using a detector timing of 500 ms, resulting in a total 209 scan time of 1800 secs (30 min). The sample spatial resolution varied from 1.25-1.5 μm/voxel depending on 210 sample. Data was visualised and prepared for analysis using VGSTUDIO MAX V.2.2 Software (Volume Graphics, 211 Germany). A median filter was used to reduce image noise, then regions of interest (ROIs) were created and 212 reconstructions rendered in 3D. 213 Fiji, an open source image processing package by ImageJ (see Schindelin et al., 2012), was used to analyse 215 stacks of 2D images. Three stacks of 200 images each were selected from different regions running through the 216 3D object. Each stack underwent a thresholding step to create a binary image and both light and dark noise 217 below 3x scan resolution was removed from the images. ROIs were selected running through a given stack so 218 that the area being analysed always contained particles and was not on the periphery of the object. 219 When processing images for the maltodextrin sample the air bubbles and crack features were analysed using 220 the 'Analyse Particles' ImageJ plugin, whereas for the cellulose sample the LUT was inverted after thresholding 221 to allow analysis of the particles. Computed values for average size and percentage area were not significantly 222 different between regions of interest (ROIs) and stacks for either sample. ImageJ also allows analysis of particle 223 aspect ratio and average roundness 4 [ ] 2 with 1 being a perfect sphere and 0 an infinitely elongated 224 polygon (ImageJ User Guide, Rasband). was observed at shear rates higher than 10 s -1 . In Figure 2 this can be compared to the relatively short zero-235 shear plateau of native XG at equivalent concentration. However at high shear rates, relevant to ink jet 236 printing, the solution viscosity at 0.5 %wt is very low. This is actually advantageous, as increasing the solution 237 concentration to 1 %wt XG a printable ink viscosity is maintained whilst allowing a higher effective 238 concentration of XG to be available for interaction with the polysaccharides in the powder constituent. As 239 discussed in the previous work, ball mill treatment of XG lowered the sample molecular weight, indicated by a 240 solution viscosity decrease at a given concentration. Ball milling for 60 min at both 800 rpm and 400 rpm in the 241 previous work showed little change to the viscosity at high shear for 0.5%wt solutions compared with native 242 XG. Using the higher of these speeds and doubling the milling time, tested in the current work, has had a much 243 greater effect which will be discussed below. Lower molecular weight polymers are less likely to cause nozzle 244 blockage compared with the native high molecular weight, long chain polymer. The zero shear plateau also 245 becomes longer for samples ball milled for a longer time or at a higher speed, indicative of a wider molecular 246 weight distribution. This is expected, considering the non-specific destructive nature of the ball milling process 247 and equivalent results observed for other polymers, such as cellulose. 248 8 The 0.5 %wt solution of XG ball milled at 800 rpm for 120 min was printable without adding ethanol and 249 Tween 20, but droplet formation was not consistent. With added water, ethanol and tween 20 1 %wt of the 250 newly milled XG could be incorporated into the ink whilst maintaining the appropriate printing characteristics. 251 Therefore, through milling, a higher weight percentage of XG could be incorporated into the ink, increasing the 252 interaction with the cellulose powder substrate as described above. The range of flow curves (A) in Figure 2  253 show how modulation and tailoring of ink formulation flow properties can be achieved through the addition of 254 mechanically modified XG. Ethanol is a low viscosity, low surface tension substance which is miscible with 255 water. It is often used as a co-solvent in edible and non-edible aqueous ink formulations for its favourable 256 properties ( transition occurring between 65-50 °C as well as reduction in peak sharpness and slope gradient are evident.

262
This indicates that the ball milling process is affecting the ability of XG molecules to form helices in solution.

263
The proposed mechanism for this effect is that the reduction in molecular weight is sufficient to drop below 264 the required critical chain length for helix formation and association in some instances, a phenomenon seen 265 with other polymers (Aymard et al., 2001) as well as XG with respect to solvation in molecular dynamic 266 simulation studies (Ong et al. 2018). Due to the larger molecular weight distribution and lower average 267 molecular weight observed through rheological studies it is likely some of the sample has been broken down 268 to this extent during the intense milling regime. The material that survives milling forms helices at the same 269 temperature as the unmilled samples, which indicates that the milling process itself does not change the 270 chemical composition of XG (i.e. acetate and pyruvate content), as this is known to shift the thermal 271 transitions. 272 independent of the initial cellulose source used. Regardless, recrystallization was always performed on the 284 entirety of a sample either through equilibration at selected relative humidities or drying with or without first 285 saturating with an anti-solvent. 286

Selective Recrystallization in 3D
We previously described the production of an amorphous cellulose powder by ball mill treatment and defined 287 a moisture and temperature dependency on its recrystallization with thermal analysis after equilibration over 288 selected saturated salt solutions. It was concluded that extrapolation of this knowledge into a binder jetting 289 scenario would lead to the selective recrystallization of powder printed on with the ink, leaving the 290 surrounding unbound powder in an amorphous state as is detailed in the results obtained below. 291 Using the standard ball milled cellulose powder and both ink 0.25 %BM400 rpm, 60 min and 1 %BM800 rpm, 292 120 min 10x10 mm 3D square blocks were created of various heights using the 3D set up described in section 293 3.6.2. Preliminary testing revealed that the perimeter of an ink droplet on a model porous substrate was 294 slightly larger than the nozzle diameter of 21 µm (results not shown). To ensure a good overlap of droplets and 295 sufficient saturation of powder a 10 µm spacing was selected for printing. This drop spacing equates to 2540 296 drops per inch (DPI). 297 Ideally, through use of a binder jet system with a heated bed capacity, the bulk powder would be kept a few 298 degrees below the target recrystallization temperature (determined on relating thermal properties of the 299 particular powder composition to the ink : powder saturation ratio) and recrystallization would be 300 'instantaneous' as the ink is jetted onto the substrate. However for the proof of principle set up used here this 301 was not possible. Instead, the entire system was placed in an oven after printing and layering had been 302 completed to provide the necessary heat for recrystallization. The printed structure and surrounding unbound 303 powder were collected separately and analysed with WAXD. From Figure 3 it is clear that the powder which 304 had been printed on had partially recrystallized whereas the bulk powder surrounding it had not, despite both 305 phases being subjected to the same heat treatment for the same time. The amorphous powder remaining 306 could be collected and re-used in such a process, with automation, thus allowing structure development 307 without material wastage. 308 The crystallinity values determined from peak deconvolution indicated that the resulting partially crystalline 309 sample had a mixture of type I and type II cellulose (see table 2 below). Visually, this is apparent by the sample 310 which is usually at its maximum at 21-23° (Ghaffar, 2016), and the 040 peak just below 35° re-emerging as a 312 sign of cellulose Iβ crystallites (Park et al., 2010;Ju et al., 2015). This result is expected as more recent studies 313 on ball milling and recrystallization of cellulose, mentioned above, suggest that with longer milling times 314 recrystallization to type II is favoured, whereas type I (or a mixture) is favoured for short to medium milling 315 times. The milling regime of 800 rpm for 30 min produces a sample on what the authors refer to as the 316 'amorphous threshold' i.e. this was the first time period at the given rpm that the diffraction pattern exhibited 317 a broad, featureless Gaussian-like shape indicative of a sample lacking in order. Deconvolution of this indicated 318 that a small amount of residual type I crystals were present in the milled sample. Thus on recrystallization 319 during the printing and heating process it is likely that these contributed a 'seeding' effect for more type I 320 crystal propagation. In localised areas of the sample where these seeds were not present (which could be truly 321 described as 'amorphous' regions) the more thermodynamically stable type II crystal lattice would be 322 favoured. 323 A range of temperature and time combinations were tested to ascertain which allowed sufficient heat transfer 324 through the powder bulk. It was found that 10 min in an oven set to 75 °C achieved the desired selective 325 recrystallization of powder which had been printed on without causing detrimental effects, such as surface 326 cracking, that occurred at heat treatment of higher temperature/shorter time. 327

Effect and concentration of additional powder components 328
Structures comprising of pure cellulose were very brittle and difficult to handle, even though they had partially 329 recrystallized. Solid objects such as the squares described above in section 4.2 could be removed easily from 330 the powder bed but more complex designs were less successful. As discussed in the introductory section other 331 stereochemically similar polysaccharides are known to interact with cellulose, including galactomannans (e.g. 332 LBG) and glucomannans (e.g. KGM). The DSC traces in Figure 4 confirm changes to the thermal behaviour of 333 such mixtures with the system recrystallization event highlighted by an arrow. The other peak observed at 334 ~50-60 °C relates to a generic polymer relaxation peak, described by Appelqvist et al. (1993). This transition is 335 seen for a broad range of polymers; its enthalpy but not temperature position is affected by water content of 336 the sample. Initially 1:1 admixtures of cellulose with LBG and KGM were ball milled as described in section 3.2. When 339 tested in the 2D recessed plate with all ink formulations the resulting films were noticeably porous, rather than 340 cohesive, and had a tendency to 'lift' from the powder surface, rather than absorbing ink in a uniform manner, 341 to create warped films. Translated to a 3D process this is a major issue as layering more powder on top would 342 disturb the printed pattern underneath, causing misalignment and destroying the intended 3D structure. Both 343 LBG and KGM are hydrophilic polymers. As ball milling is both a destructive and non-selective process the 344 molecules are randomly broken up. This may expose reactive groups which may not ordinarily be available by 345 disrupting ordered molecular conformations and hyperentanglements (detailed below). As mentioned in the 346 introductory remarks, an optimal 1:1 ratio was identified for β(1-4) glycan synergistic interactions (Goycoolea 347 et al., 1995b). This stoichiometric relationship has not specifically been discussed in literature for 348 galactomannans or glucomannans with cellulose, but there is strong evidence that this is the case (Abbaszadeh which is not apparent within the cellulose only sample (see Figure 5). Warping in 3D binder jetting is a known 353 issue typically caused by delayed solidification and in-homogenous shrinkage during printing (Schmutzler et al.,354 2016). Within the system presented here this could be explained by a preference of ink moisture absorption by 355 the hydrophilic LBG and KGM fragments present within the sample compared with cellulose fragments. Thus 356 moisture distribution in the sample would not be homogenous and directly affect drying kinetics. This, 357 however, would imply competition rather than synergism between the polymers present. What is more likely, 358 given information available in literature and the observed impact of LBG and KGM on the cellulose 359 recrystallization transition in Figure 4, is that the synergistic molecular interaction of components is greatest at 360 this 1:1 ratio but macroscopically this does not translate to linear effects. Molecular reordering and 361 heterogeneous moisture release on drying during recrystallization, as well as film formation of dried gel 362 through specific interactions of XG with LBG or KGM, cause warping at the corners of single layer printed 363 squares. What is evident in Figure 5 is that increasing the relative content of cellulose in the powder compared 364 with LBG and KGM slightly mitigates this negative warping effect (a 'failure' for translation to 3D) whilst also 365 giving enhanced single layer cohesion in 2D printing. 366 Therefore the cellulose fraction in admixture was further increased relative to the LBG and KGM sequentially, 367 with the most promising results being observed with 9 parts cellulose to 1 part added polymer. 3D 10x10 mm 368 squares were successfully printed with pure cellulose and 9:1 ball milled mixtures of cellulose and LBG or KGM.

369
The crystallinity profiles are compared with that of a printed 100 % cellulose sample in table 2. 370 Though XG acts synergistically with both LBG and KGM, the synergism is greater for glucomannans due to the 371 higher proportion of unsubstituted sections of the glucomannan backbone (Stephen et al., 2006). Even with 372 the reduced LBG fraction the powders exhibited undesirable properties for 3D structure creation such as lack 373 of cohesion and warping (see Figure 6), thus more complex structures in 3D were created with the 9:1 374 cellulose KGM ball milled powder, shown in the following section. interactions between neighbouring galactomannan chains above a critical coil overlap concentration, termed 380 "hyperentanglement" (Goycoolea et al., 1995a). These hyperentanglements are not restricted to 381 unsubstituted regions of the mannan backbone, along the axes of the a crystal lattice plane, and can occur 382 along regions with galactose residues to create stable associations over time (Doyle et al., 2009). Cellulose 383 chains pack together along the b axis to form crystalline, ordered regions (Wüstenberg, 2014). Therefore, the 384 destructive nature of a high speed short time ball milling regime is likely to disentangle some 385 hyperentanglements as well as breaking up the cellulose crystalline lattice, whilst leaving some type I crystal 386 seeds as previously discussed. Upon application of the ink molecular mobility increases and this is further 387 enhanced by the post-process heating step required to induce recrystallization. During the printing and 388 heating processes the kinetics of recrystallization onto the type I seed templates will compete with the 389 thermodynamic drive of fully amorphous cellulose segments to recrystallize in a type II crystal lattice structure 390 (evidenced by the mixed lattice structures of the resulting printed objects). In amongst this, reformation of 391 hyperentanglements between LBG fragments and their incorporation into the semi-crystalline cellulose lattice 392 will increase spacing between chains along the b axis. Therefore it is unlikely that the resulting crystal structure 393 is able to form linearly and that steric hindrance by LBG will cause directional changes and kinks on a molecular 394 level which appear as warping on a macroscopic level. KGM, on the other hand, is unbranched and contains 395 diequatorially linked glucose along with mannose in the backbone, which enhances its interaction with 396 cellulose (Abbaszadeh et al., 2014). Low molecular weight KGM (such as that used in this work) has been 397 shown to crystallize readily into the antiparallel mannan I lattice and interact with cellulose (Chanzy et al.,398 1978; Chanzy et al., 1982). Therefore as recrystallization ensues in this system steric hindrance will not inhibit 399 chain packing along the b axis, mannan I KGM will certainly interact with the type I lattice formed through the 400 seeding effect and perhaps also with the thermodynamically favourable type II lattice due to the antiparallel 401 nature of chains in this configuration. 402  Table 2 Crystallinity of cellulose only and mixed powder systems after printing and recrystallization. As the powder layering process in this work (see section 3.6.2) was designed as a low volume, experimental set 404 up, with a heavily manual emphasis on its ability to function, maltodextrin (MD) was used as a model powder 405 to determine the capability of the system. The manufacturer lists relevant material applications such as 406 excellent binding and diluent properties during compression and a suitable carrier in spray drying, thus it was a 407 suitable model material choice. 408 In 2D, patterns were created as a series of individual drops facilitated by a pattern editing function in the 409 Dimatix software. To move to 3D more complex bitmap files were imported for printing, such as star shapes to 410 observe the resolution of sharp corners. MATLAB script was also used to create two different series of bitmap 411 images which could be alternately printed as layers to build designs. The first script coded for two square 412 images; one larger with a defined border thickness and hollow centre, the other a set of four solid squares 413 whose sides were of equal length to the thickness of the border which could be used as struts on top of the 414 larger square (A in Figure 7). This enabled the two images to perfectly superimpose and create hollow boxes or 415 'lattice' type structures when multiple layers of each were printed. The second script coded for a series of 10 416 circles whose diameter decreased by one unit per time. These could be printed in sequence to produce a 417 variety of circular based structures such as spheres, round bottomed 'pyramids' or 'spinning top' type 418 structures (B in Figure 7). Length scales are not specifically mentioned in the descriptions as these could be 419 changed in the script, depending on the desired size of the final structure. 420 In Figure 8 printed pieces made entirely of MD are shown, including an SEM micrograph of the surface and 421 internal structure, visible by fracturing the sample prior to Pt coating. 422 Figure 9 shows successfully printed square, star and 'spinning top' 3D pieces, and SEM of their microstructure, 423 using 9:1 cellulose/KGM ball milled powder and an ink formulation containing 1 %wt milled XG 800 rpm 424 120 min. More complex designs with thin 'struts' holding square boxes were not successfully printed without 425 also incorporating maltodextrin into the design, for example alternately layering the cellulose and KGM 426 powder with MD powder. In addition to confirmation of recrystallization by XRD (section 4.2) DSC of these 427 printed samples (results not shown) still exhibited the generic polymer peak (described by Appelqvist et al., 428 14 1993) but no longer showed the recrystallization event highlighted by arrows in Figure 4, thus indicating that 429 this had occurred during the printing and heating process as intended. 430 Scans of the printed pieces using MicroCT correlated SEM with respect to structural observations. The smooth 431 interior, cracks, bubbles and comparatively rough exterior of the printed MD seen in Figure 8 is also apparent 432 in image A of Figure 10. Similarly, the needle-like particulates contributing to a highly porous structure, seen in 433 the top images of Figure 9, is observable throughout the structure in images B and C of Figure 10.   native particle morphology is semi-spherical, but not high aspect ratio like native, crystalline cellulose particles. 453 In Figure 10 the selected ROI appears to be comprised, in part, of particles with a higher aspect ratio which 454 may be a result of the selective recrystallization event occurring during the post-print heating step. to provide selective recrystallization of ball milled cellulose facilitating the production of 3D structures has 458 been confirmed. Though solid, filled structures (such as squares) could be printed with cellulose alone, the 459 structure cohesiveness and intricacy of structures that were possible was greatly enhanced by the addition of 460 konjac glucomannan. This is due to the synergistic effect between the three polymers used resulting from 461 stereochemical similarities in their backbone; cellulose, xanthan gum and glucomannan. The recrystallized 462 cellulose structures obtained comprised of a mixture of type I and type II celluloses due to the persistence of 463 type I crystallite seeds in the ball milled starting material. It was confirmed that only powder which had been 464 printed on was recrystallized upon heating, the surrounding unbound powder could be collected and reused to 465 create a no waste, sustainable system. Ball milling of xanthan gum enabled higher concentrations to be 466 incorporated into inks whilst maintaining acceptable printing qualities. This processing of xanthan lowered the 467 intensity of the coil to helix transition but not the temperature range over which it occurs, therefore the 468 milling was not so severe that the chemical structure was altered. 3D structures were successfully created 469 ranging from simple square designs to more complex stars and 'spinning top' geometries using a bespoke 470 manual layering substrate system. These samples were highly porous but were still able to be handled, 471 transported and analysed by various methods. A small scale layering system was produced and shown to be 472 useful in testing experimental powders, allowing for flexibility in that large quantities were not required.

473
Texture analysis of these printed materials in the future would give further insight into the product hardness 474 and breakdown profile of the structures in relation to intended food analogues. The next stage would be to 475 test these experimental powders in a commercial, automated binder jetting machine which would allow for a 476 controllable powder bed temperature as well as powder spreading and layering less subject to human error 477 than the experimental rig presented here. 3D printed designs could range from recreation of natural food 478 structures, such as strong, brittle gluten matrices in breadsticks or cookies to mimicking porous, hard materials 479 in the biomedical sphere and other industries.