Speed-Modulated Ironing: High-Resolution Shade and Texture Gradients in Single-Material 3D Printing

3D printed objects that result from FDM 3D printing typically only have a limited number of material properties since each extruder can only be assigned a single material at a time. To overcome this limitation, researchers investigated how to fabricate gradients of different materials. For instance, researchers showed how to create gradient color by interleaving black and white filaments to create grayscale images [11, 24]. A different approach to fabricating gradients of different materials is to use custom hardware that feeds multiple filaments into a single mixing nozzle at varying rates [7, 25]. Since different print temperatures and print speeds required by different materials limit what materials can be mixed together, Takahashi et al. proposed to pre-fabricate a filament that consists of individual segments made from different materials [15, 29]. While this allows for a more diverse range of materials, the filaments need to be prefabricated for each individual 3D-printed object. Addressing this issue, Mosaic provides a custom hardware add-on that can assemble a filament from different filament spools on the fly depending on the 3D model that is being printed¹. However, the properties of the 3D-printed object are still limited to the number of spools the hardware can process in one print job.

The 3D print quality, surface finish, and mechanical properties of FDM printed parts are dependent on process parameters (e.g., nozzle temperature, print speed) and slicing settings (e.g., layer height, line width, infill density, and type) [6, 23]. Thus, these parameters allow controlling certain properties of the printed part, such as strength and toughness [10, 27], or shape memory [20, 26].

While the previously mentioned works use common filaments to program material properties, there are also filaments with additives that can be triggered by heat to obtain a significant and visible response. For example, filaments with wood particles become darker when printed at a higher temperature and have been used to vary color shade in a print [16]. Another category is foaming filaments [9, 18]. These filaments contain a foaming agent that, when extruded, expands the material volume into a closed-cell foam structure. The properties of the 3D printed foam depend on the heat applied, which, when varied, can produce various visual and haptic [19], as well as mechanical properties [3, 4, 21]. While researchers have presented strategies to apply such filaments at different states in 3D printed objects, all of the strategies rely on using the relatively slow temperature change of the printing nozzle. This leads to limited freedom and resolution in locally varying properties. We discuss these challenges in more detail in Section 3.

(A) A blue nozzle prints a layer of material. (B) Then a red nozzle goes through the printed layer with different speeds. (C) The print process captured by a thermal camera, showing the heat of the nozzle and on the printed line. (D) A print captured by a regular color camera, showing a circle pattern in the middle of the object.

To expand the capabilities of 3D printers, researchers have developed a number of custom 3D printing strategies. For instance, WirePrint  [17] moves the print head along the z-axis to create toolpaths that 3D print the objects as a wireframe for fast prototyping. Furbrication  [13] uses rapid movements of the print head to create hair-like structures by leveraging the stringing properties of the filament. DefeXtiles [5] uses different rates of under-extrusion to produce textile-like structures with varying flexibility within 3D printed objects. Similarly, Velocity Painting uses under-extrusion due to rapid print speed changes to create visual patterns on the object's surface [32].Expressive FDM [28] modifies the height of the extruder in conjunction with the amount of deposited material, which results in different printed patterns, thereby forming various haptic structures. All of these works have in common that they enable new 3D printing capabilities without modifying the 3D printing hardware. Our work is in line with these related papers in that we leverage the existing unmodified hardware and embedded control systems to fabricate parts in an unconventional and new way. Unique in our approach is that we separate the printing from the programming step to achieve the desired visual and haptic properties, expanding the capability of the 3D printer.

Our method utilizes the ability of some materials to obtain varying color shade, texture, and translucency when exposed to different levels of heat. In this paper, we present our work on three off-the-shelf 3D printing filaments; however, in principle, our approach can be used for other temperature-responsive materials as well. The first filament is a foaming filament, which contains a foaming agent that causes the filament to convert into a closed-cell structure when heat is applied. The higher the applied heat, the more the foam expands, leading to larger cells. The cells in the foam structure scatter light, resulting in a brighter color appearance and increased opacity. In addition, the cells result in a coarser tactile feel of the surface. The type of foaming filament used in this work is LW-PLA². As the second and third material, we use PLA filled with wood fibers (Woodfill³) and with cork fibers (Corkfill ⁴) respectively. The last two filaments have a comparable temperature response, they obtain a progressively darker color shade when exposed to higher temperatures due to pyrolysis [16].

Our goal is to be able to create 3D prints with local property variations that are continuous and have a high resolution. While the temperature-response of filaments has been used before to locally vary appearance or other properties in 3D prints [16] [19], none of the existing methods is capable of reproducing high-resolution shade and texture details, since they rely on changing the printing nozzle temperature, which is very slow and challenging to control. For example, to obtain significant color change in Woodfill filament, the nozzle temperature must vary between 195° C and 300° C, which takes tens of seconds on most 3D printers. Since rapid changes in nozzle temperature are not possible, existing approaches rely on segmenting the 3D geometries into discrete regions, each printed at one temperature. The temperature changes are then done in-between printing the segments. Furthermore, the nozzle needs to be purged frequently when a filament that has been exposed to a high temperature is present in the nozzle, but the next part needs to be printed at a lower activation level. This results in long print times, and increased material waste. The segmentation imposes significant limits to the achievable resolution, as every segment is printed in a separate path, losing the continuous nature of extrusion 3D printing. The smoothness of graded transitions is also limited since gradients are divided into discrete steps and truly graded transitions are only possible in the z-direction of printing, the only case where the nozzle has sufficient time to achieve the required temperature. These limitations make the approaches that change printing nozzle temperature unsuitable for our purpose.

A process diagram, starting with input gcode and ending with output gcode. The mid-steps are the outer wall, ironing path, sampling points with varying speeds, ironing gcode, and merge, resulting in output gcode.

Working principle: We separate the 3D printing and re-heating for programming into two consecutive steps using an unmodified dual-nozzle 3D printer. The first nozzle is dedicated to the primary task of 3D printing the filament at a relatively low nozzle temperature that does not cause any temperature response of the material (see Figure 2 a). The second nozzle is empty (i.e., does not have any filament loaded into it) but is used for programming the properties by re-heating the already printed layer of material. The re-heating is done by moving the empty heated nozzle over the print while slightly touching, "ironing", the top of the printed layer. The temperature increase in the layer triggers a temperature response of printed material, such as change in color shade. The separation of the 3D printing and programming into two steps allows us to leverage the accurate motion control of the printing head since the speed of the ironing nozzle can be rapidly and precisely changed within a layer, unlike nozzle temperature. We keep the ironing nozzle temperature constant and modulate the applied heat through variations in ironing speed (see Figure 2 b). Slow ironing causes the layer to heat up more than when ironing at a faster speed (see Figure 2 c). We choose to use separate nozzles for printing and ironing. Doing both steps with the same nozzle is impractical as it requires long temperature changes at every layer. This increases print time significantly and exposes residual material in the nozzle to high temperatures for prolonged times, which will likely clog the nozzle. We derive the name of our method from the known concept in 3D printing, "ironing", where the top surface of a 3D model is smoothed by a nozzle at a constant speed.

Ironing workflow: The main input to our workflow is the gcode of a sliced geometry for a single-nozzle 3D print prepared by the user. Gcode instructions for the ironing by the second nozzle are inserted into this input gcode while all 3D printing instructions and features, such as inner walls, infills, and support structures, remain unchanged. This workflow also preserves the user preferences of slicing, machine, and material-related settings for 3D printing. A schematic overview of the workflow is shown in Figure 3.

Since we are interested in the color shade and textures of 3D objects in this work, our method applies the ironing step only to the outer contour in each layer. For each layer of the 3D print, we extract the contour geometry of the outer wall from the input gcode. To this contour, we apply sampling points at a distance that we tested as short as 0.2mm. For each point, we then sample the intended property value (color shade, texture, translucency). The property values can be defined via an image, such as the color texture image of a 3D mesh file, or the user can project an image onto a 3D geometry directly in our tool (see Section 4).

Each sampled value, for example, target color shade, is then mapped to an ironing speed that will yield the required re-heating and, therefore, corresponding color change in the material. Using these modulated speeds, we create ironing toolpaths and insert these into every layer of the input gcode, together with the printer instructions for the nozzle changes. The final gcode file can be sent to the 3D printer for fabrication.

Ironing parameters: For each of the used materials, we determined the ironing parameters that yield the desired results. We aimed for a maximum difference in the property (color shade, texture, translucency) while keeping the structural and 3D detail integrity of the 3D printed shape. For example, a too-high ironing temperature or too-slow ironing speed may be able to trigger a response of the material but at the same time damage the printed structure. Possible damages include an unintended roughening of the surface or even holes in the printed part. This means that the ironing parameters need to be well-balanced. The main parameters we investigated were ironing temperature, ironing speed range, and ironing height. Table 1 presents the ironing parameters we use in our method.

We tested different ironing nozzle heights, which were measured as the distance from the printed layer. When the ironing nozzle is too high, there is too little to no contact with the printed layer, which leads to insufficient re-heating and, therefore, no property change. An ironing nozzle that is too low will "dive" into the printed layer, causing visible damage. The best results were obtained when there was no height difference, i.e., the tip of the ironing nozzle was at the same height as the top of the printed layer. This requires a well-calibrated z-offset of the two nozzles in the 3D printer. Many modern 3D printers have an auto-bed leveling feature that can be used to address this issue.

A user interface. Left: The interface shows the material as LW-PLA, and the textured OBJ option is selected as the texture input mode. An image file, Einstein's face, is loaded. There is a text field for the output gcode name and a generate gcode button. Middle: The user can preview the result. The printed result on a cup.

The relationship between the ironing speed and the temperature obtained in the material has shown to be highly non-linear in our initial tests. To be able to control the temperature response of the material accurately, a better understanding of this relationship was needed. We therefore developed a theoretical model for ironing-induced heating. We use the reached material temperature as a proxy for color change in our model. For Woodfill, the relationship between temperature and color change has shown to be linear within a temperature range of 200° C and 300° C [16].

We model the system as a 1D conductive heating problem of a semi-infinite region with one-sided heating. The ironing nozzle acts as the heat source and is in close proximity or direct touch with the printed layer. The printbed is an adiabatic surface of fixed temperature. Since convection to air is minimal when ironing is actively occurring, we ignore its contribution in our model. Consequently, the heat transfer equation for our model is:

where T is the temperature of the material, z is the (vertical) distance from the heat source (ironing nozzle), t is the time, and α is the thermal diffusivity of the material. We solve the governing heat transfer differential equation (Equation 1) as a function of z and t:

where "erfc(x)" is "1-erf(x)" (error function). The boundary conditions are T(z, 0) = T_bed, constant heat flux at surface, and T(∞, t) = T_bed. Under these conditions, using constant values for PLA (k = 0.1W/mK, h = 50W/m²K, α = 10^{− 7} $\frac{m^{2}}{s}$) [22, 31],and substituting t with D_nozzle/V_iron (where D_nozzle = 1.3mm), the exact governing solution for the heat transfer equation is expressed as:

where T is the material temperature (°C), z is the distance from the heat source (mm), and V_iron is the nozzle's ironing speed (mm/s).T_bed is the printbed temperature, which is held constant at 60°C, and T_iron, is the ironing nozzle's temperature.

Figure 5 shows the modeled maximum temperature of the material as a function of different ironing speeds, assuming an ironing nozzle temperature of 340°C.

We validate our model with experimental data in Section 6 and use the insights of the non-linear relationship between ironing speed and locally obtained temperature for the mapping of the desired property change to ironing speed in our tool, as discussed in Section 4.

A plot showing the inverse relationship between temperature and ironing speed. The left version of the plot shows when the nozzle is at z=0mm, and the right version is at z=0.15mm.

Our approach offers precise control over shade variations within 3D-printed objects. For example, textures, decals, and text on 3D models allow for personalized designs. By carefully controlling the re-heating process, printed aesthetic patterns can be created, adding additional artistic appeal to objects such as vases, mugs, or decorative sculptures. Figure 6 shows a series of vases with different decorative patterns applied to the surface.

Our method can also be used to fabricate 3D-scanned objects. A textured mesh of a color-scanned owl figurine ⁷ was imported into our tool (Figure 7 a). The model with the texture details was reproduced using Corkfill filament,, (Figure 7 b and c), Woodfill filament (Figure 7 d), and green foaming filament (Figure 7 e).

These examples illustrate the capability of our method of enhancing 3D prints with decorative or functional visual features.

Five 3D-printed vases with different textures in various shades of brown.

Left: a digital model of an owl figure. Right: Two printed versions, one in darker brown, one in lighter brown.

Translucency is a desirable property in various objects, and our method enables the creation of 3D-printed objects with regions of varying translucency. As an example of local translucency differences, Figure 8 shows transparent liquid containers with opaque design features printed using "natural color" LW-PLA.

Two printed bottles. One says "TU Delft" in transparent style, and the rest of the bottle seems like opaque engraving. The other bottle has "water" written in opaque engraving style, and the rest of the bottle is translucent. Top: A sun-like pattern printed on a white-transparent piece.

Texture plays a crucial role in the tactile functionality and user experience of objects. Our method enables the incorporation of diverse textures into 3D-printed objects. It allows customization of texture variations according to custom comfort or personal requirements, making it suitable for a wide range of applications. This includes textured handles for tools and appliances or ergonomic grips for sports equipment, e.g., the bike handle example in Figure 9.

Figure 10 shows a speaker enclosure with touch buttons which are smoother to the touch compared to the rest of the housing. These areas also visually signal the tactile, "touchable" nature of the buttons. This part was printed using regular PLA (i.e. without special additives) and is, therefore, an example of our method being applied using a commonly used printing material.

A bike handle printed with a fine, honeycomb-like pattern with rough and smooth tactile details.

A green speaker with rough and smooth areas on its top. The touch buttons have a smooth finish.

We 3D printed samples of 80mm x 20mm to which a 2x2 rectangular grid of different regions was applied. Two of these were the reference regions that were ironed at a constant high speed at which no significant re-heating, and therefore no color change, occurs. The other two regions were ironed at lower speeds and varied per sample; these were the sampling regions (Figure 11 b). All samples were printed on an Ultimaker S5.

The samples were placed inside a diffusion box (Figure 11 a) for controlled lighting conditions and photographed with a DSLR camera. We used the CIELAB [8] color space to measure the colors of the samples.

To gauge the color difference our method can achieve, we use Δ E as the metric. Δ E (Equation 4) represents the Euclidean distance between two colors in CIELAB color space. A Δ E less than or equal to 1.0 is considered to be a non-perceivable color difference for the human eye [8].

The conversion to CIELAB and the calculation of the Δ E were done in an interactive Matlab script that we developed. Since our samples have two sampling regions and two reference regions within a single print, we calculate Δ E with respect to each reference, then report the global average (Figure 11 b). Because the color difference values are all measured relative to the reference zones, possible variations in lighting conditions are eliminated.

Left: A DSLR camera looking at a small brown piece. Right: The brown piece has light and dark details. The light region is marked as reference regions, and dark as sampling regions.

Woodfill Filament: We prepared 13 samples with fixed printing nozzle parameters (200°C printing temperature, 0.15 mm layer height, 0.4mm line width, no infill, wall-count 2). The ironing temperature was 340 °C, and the ironing speed was varied from 1mm/s to 13 mm/s for the sampling regions. All reference regions had an ironing speed of 40 mm/s.

The results of the printed samples are illustrated in Figure 12.

13 printed light brown checkerboard patterns stacked together, with decreasing contrast from left to right.

As shown in Figure 13, as the ironing speed increases, the Δ E value decreases exponentially until it plateaus. The maximum Δ E was measured in the 1mm/s sample with a value of 63.8 ± 5.06.

3 plots, each showing delta E plotted against ironing speed (mm/s). First one is for woodfill, second for corkfill, third for LW-PLA.

Because of the glossiness of the non-foamed material, the viewing angle and lighting direction affect the perceived color difference within the prints. For instance, in Figure 14 a, the test pattern is clearly distinguishable at no tilt. However, when the same sample is tilted forward, it appears darker, and when tilted backward, it appears lighter. Within our evaluation, we selected the orientation that yielded a stable color difference with minimal reflections, which is a vertical orientation with no tilt.

Finally, the range of speeds tested was from 1mm/s to 30 mm/s for the sampling regions, and 40 mm/s for the reference regions. The printed results are illustrated in Figure 16, where we generally observe an exponential decay of contrast values as a function of ironing speed. The maximum Δ E attained at this orientation under non-directional lighting is 28.66 ± 3.91 at 1mm/s ironing speed. The outliers between 4mm/s and 9mm/s can be attributed to the sample's changed glossiness at both ironed and reference regions, resulting in higher contrast or Δ E values, as can be seen in Figure 14 b.

Left: The same slab tilted over time, which shows the varying glossiness. Right: Multiple printed samples stacked horizontally.

13 printed dark brown checkerboard patterns stacked together, with decreasing contrast from left to right.

Similar to the measurements of the Woodfill and foaming filaments, the Δ E values for Corkfill decrease exponentially with an increasing ironing speed (Figure 13). The maximum Δ E value for Corkfill was found to be 55.33 ± 6.81, obtained by ironing at 1mm/s.

Measured Characteristic of Δ E: To get a characteristic description of the measurements in an equation, we first normalize the measured Δ E. We can describe the relationship between the measured Δ E and ironing speed in mm/s with the exponential equation:

where v is the velocity in mm/s. For normalized Δ E, the constant values we found for the three filaments are given in Table 2. All the values are within a $95\%$ confidence bound, while in the case of foaming filament, this confidence bound excludes the outliers at ironing speeds of 4,6,7,8 and 9 mm/s.

Validation of the theoretical model: The experimental results correspond to the general trend we expect from our theoretical model, which is an exponentially decaying graph as a function of increasing ironing speed.

By normalizing the data from the theoretical model between 1 and 0, we are able to compare the characteristic relationship between ironing velocity and measured Δ E and the modeled ironing velocity as a function of temperature at z= 0.15mm (layer height).

Figure 16 b shows this result for Corkfill filament. The theoretical model fits the experimental data well with a maximum normalized error of 0.0672 (i.e., 6.72$\%$ of the max value) and an average error of 0.0168 (i.e., 1.68$\%$ of the max value). We also generally conclude from the closeness of fit between the experimental and theoretical models how the relationship between the normalized temperature values and normalized Δ E values is linear.

Using the same method for the Woodfill printed samples, we find that the experimental model fits the theoretical model generally but with a larger error margin in comparison to the Corkfill model (max error of 0.16 or 16$\%$, and average error of 0.06 or 6$\%$). This is also true for the foaming LW-PLA filament with a max error of 0.1329 or $13.29\%$, and a mean error of 0.035 or $3.5\%$) excluding outliers.

3 plots, each showing normalized delta E I plotted against ironing speed (mm/s). The first one is for woodfill, second for corkfill, third for LW-PLA.

To evaluate the capability of our method to reproduce a variety of visual features and texture designs, we prepared samples that test rapid variations in color shade, thin features, and gradient transitions of color. All tests were carried out for LW-PLA, Woodfill, and Corkfill filaments.

Detail and gradient testing: A 60mm x 60mm x 60mm cube was designed, and a test image was applied to each of the four vertical faces. The test images are illustrated in Figure 17 a. The first test image is a QR code, which tests a practical application of rapid spatial transitions with significant color differences. The second image aims to evaluate the ability to produce smooth gradients. The third tests thin light lines on a dark background and dark lines on a light background at different angles. Finally, the last image evaluates rapid transitions between minimum and maximum reheating.

(A) Digital versions of a QR code, a circular gradient, sun-like icons. (B, C, D, E) 3D printed versions of these with different filament types.

All samples were printed successfully with Woodfill, Corkfill, black LW-PLA, and natural LW-PLA; see Figure  17 b, c, d, e. The prints allowed a closer inspection of the performance of our method with the test images. For all materials, the QR codes were of sufficient quality and contrast to be directly readable by a regular smartphone camera. Also, the circular gradient came out with an even and smooth transition with all materials. The main edge case is the test of thin lines printed with Woodfill. For these specific features, it might be necessary to reconsider the ironing parameters (i.e. nozzle temperature and speed). However, for Corkfill and LW-PLA, the lines were produced well and showed clear sharp transitions irrespective of line orientation.

Text legibility: To evaluate at which sizes legible text can be reproduced, samples of 50mm x 100mm were created, to which an image with text was applied. The text size varied from 40pt down to 5pt; see Figure 18 a. The resulting samples are shown in Figure 18 b, c, and d. We found that our method can reproduce the larger font sizes on all three materials. For LW-PLA, the smallest readable font size was 9pt, while for Woodfill and Corkfill, the smallest readable font size was 14pt.

Left: Digital text in different sizes “ABCD…” Right: Printed version of the text onto corkfill, woodfill, and black PLA.

3D printed high-resolution images of famous scientists, including Einstein. (A) printed with corkfill, and (B) printed with woodfill.