Low‐Roughness 3D‐Printed Surfaces by Ironing for the Integration with Printed Electronics

The roughness of 3D‐printed surfaces poses a challenge when integrating fused filament fabrication (FFF) printing with printed electronics, leading to inconsistencies and breaks in the circuit traces. To improve the surface roughness, an ironing toolpath is proposed. The ironing toolpath involves the hot nozzle going over the printed surface with finer line spacing, remelting the surface to fill gaps, and creating a smooth finish. For further optimization, various ironing parameters are investigated including flow, speed, line spacing, and temperature. A wide range of materials is tested, including commonly used low‐temperature filaments (polylactic acid, polyethylene terephthalate, acrylonitrile butadiene styrene) and high‐temperature filaments (polysulfone, polyetherimide, polyether ether ketone) suitable for integration with printed electronics and medical applications. To collect the extensive datasets, an automated measurement system is deployed. With this method, surface roughness reductions of up to 96.6% are achieved and significant trends are identified. Lastly, the integration of 3D printing with electronics is demonstrated by printing a high‐resolution strain gauge structure on top of an ironed surface and embedding it into fully printed tweezers which can be used in medical robotics. The insights on ironing extend beyond electronics and can also be valuable in other areas where low surface roughness of FFF‐printed parts is required.

[14] Alternative techniques for multimaterial photopolymerization printing involve the positioning and aligning of added conductive particles using magnetic or acoustic fields and then fixating the particles in the desired locations by curing the resin. [15]Since the optical nature of resin printers allows the production of fine details on the micrometer scale, they can also be utilized to print hollow channels in which conductive ink can be injected. [16]Another interesting optical approach is to use an engraving laser to transform the polymer surface of the print into graphene. [17,18]This laser-induced graphene (LIG) is limited to a few high-temperature materials such as polyetherimide (PEI), polyether ether ketone (PEEK), and polysulfone (PSU).The LIG is very flexible yet limited in conductivity and durability.
Valentine et al. have successfully fabricated flexible electronics with trace widths of 100 μm using direct ink writing techniques to print both the conductive and the dielectric elastomeric material. [19]Improving on this idea, nanodimensions has recently introduced their Dragonfly IV printer, which can print both conductive and dielectric inks via two inkjet printheads.This machine is capable of impressive resolutions of 18 μm in Â and y, 10 μm in z, and 75 μm trace widths with 100 μm trace spacing but comes at a high cost and is limited to materials that can be inkjet printed. [20]oerber et al. have shown an implementation of a dispensing unit and an aerosol jet printhead into a powder bed-based selective laser sintering (SLS) printer and were able to print and embed electronic tracks and connect them with surface-mount technology (SMT) components inside the printed mechanical parts. [21]Whitehead and Lipson have combined the SLS and SLA processes and also suggested that conductive channels could be printed with the metal powder and then fused with the dielectric polymer resin. [22]ll these approaches are very interesting, and each has its advantages and disadvantages.Here, we focus on the combination of FFF printing with extrusion (direct ink writing) printing.In extrusion printers, conductive ink consisting of metal particles, a binder, and an organic solvent is extruded through a fine nozzle by the force of a piston.Afterward, the traces are sintered under heat while the organic solvent evaporates.This method allows printing electronics with trace widths of roughly 100 μm with a large variety of inks.Since the xyz-motion system is very similar to the one in FFF printers the integration into one multimaterial system will be possible.FFF printers have also recently made big improvements in their slicing software with new slicing methods and better support structures that save time and material.Furthermore, the hardware has improved with more precise and lighter frames, sensors for input shaping and automatic bed leveling, new extruder designs, and light detection and ranging (LiDAR) scanners for automatic material calibration.All this has improved the userfriendliness, print speed, and overall quality of the prints while still being fairly low cost.As depicted in Figure 1a, our envisioned multitool head printer could have a regular FFF printer nozzle to (a) (b) Figure 1.a) Schematic image of a multitool head cartesian printer printing a skull implant with embedded electronics.i) High-temperature FFF printhead prints the 3D structure of the implant with medical-grade filaments such as PEEK.ii) Conductive pastes are extruded with a dispenser system.iii) Electrical components are positioned and placed into the wet paste with a pneumatic PnP tool head.b) Cross-section of the general ironing process to achieve a smooth FFF printed surface to reliably print electronics on top.In this process, the hot nozzle goes over the printed top surface in a back-and-forth path with fine line spacing while adding and remelting material.For further optimization of the top surface smoothness, four ironing parameters are studied, which include ironing speed, material flow, temperature, and line spacing, for three different sample sizes, which correlate to the ironing path length.
extrude polymers, a dispenser that extrudes conductive paste for the traces, and a PnP tool head to place electrical components on the contact pads.The circuit would be printed in a 2.5D fashion on the horizontal planes within the print and with vias connecting the layers.In the future, the system could be extended with more axes and nonplanar toolpaths to print on curved surfaces.
Problematically, the top surfaces of FFF-printed objects are very rough as a result of the line-by-line printing process with a spacing of usually 0.4-0.6 mm. [23]Hence, the dispenser nozzle that is printing the fine circuit patterns must maintain a sufficient distance to prevent collisions with the surface.With greater distance, however, the shape and corners of the lines cannot be controlled as well.Additionally, the rough surface will cause inconsistent line thickness and possibly breaks or shorts in the circuitry.Previous research has attempted to address this issue by implementing more steps such as polishing the surface, printing a thin epoxy layer on top of the surface, or milling channels into the surface for ink dispensing which adds more complexity to the process and requires additional tool heads. [24]Therefore, we aim to find a solution to this problem by using only the FFF tool head.The main research goal of this study is to study the relevant FFF parameters to optimize the surface roughness to the point where electronics can be printed reliably and precisely.
As the top layer is laid down in parallel printed lines, the flow of material plays a crucial role in ensuring that the area between lines is filled in without the presence of peaks caused by overextrusion or gaps caused by underextrusion.Therefore, the first part of this study focuses on analyzing the effect of the overall print flow on surface roughness.With an optimal general flow established, a method referred to as ironing is studied to achieve an even flatter and smoother surface.The hot nozzle is passed over the top layer again at the same print height as the last layer but with a finer line spacing while extruding small amounts of material, as shown in Figure 1b.This process remelts the surface and fills in the remaining gaps.Previous research has demonstrated the effectiveness of ironing in reducing surface roughness with top surface roughness reductions of 60%-70% yet these studies were unable to identify significant trends and only tested two settings per parameter and three materials (PLA, acrylonitrile butadiene styrene (ABS), and acrylonitrile styrene acrylate (ASA)). [25,26]Therefore, here, we employ a one-factor-at-a-time (OFAT) experimental approach for the most important ironing parameters to uncover trends and gain a deeper understanding of their effects.Additionally, the range of materials was expanded to include three low-temperature filaments that are commonly used (PLA, polyethylene terephthalate (PETG), and ABS) and three high-temperature materials (PSU, PEI, PEEK).These high-temperature materials were selected because they can withstand the temperatures used for sintering conductive inks and possess greater chemical resistance to the solvents used in printed electronics inks.Additionally, materials such as PEEK are biocompatible and have been used to print medical implants.Our research focuses on the ironing parameters of flow, print speed, line spacing, temperature, and sample size.
After good parameter values are found, conductive patterns are printed on the ironed surface using both extrusion and inkjet printing.To demonstrate the potential of this method, a "smart" tweezer is printed with a fully embedded strain gauge inside the printed body.By using PSU as the substrate, these tweezers could potentially withstand temperatures for sterilization and could therefore be useful in medical applications where soft tissue needs to be manipulated without damage.Finally, we assess the improvement of surface quality by ironing and what the possibilities, limitations, and future perspectives are for this approach of combining 3D printing with printed electronics.

3D Printing of the Test Samples
All samples to study the surface roughness were printed on a Hydra 16 A printer (Hyrel, United States) with a layer height of 0.2 mm.For the low-temperature filaments, a 0.5 mm brass nozzle was used in combination with an MK1-250 print head whereas a 0.4 mm Micro Swiss plated A2 hardened tool steel nozzle was used for the high-temperature filaments in combination with an HT3-450 print head.These nozzle sizes were chosen to be as similar as possible within the constraints of the nozzles available for the different tool heads.It is important to note that the use of distinct printheads and different nozzles affects direct comparability between the two material groups, and this choice should be considered within this context.For the printer bed, a high-temperature glass-ceramic was used which was additionally coated with nanopolymer adhesive (Vision Miner, United States).Table 1 lists the filament materials utilized in the studies.Hightemperature materials were stored in a vacuum desiccator and underwent a 4-hour drying process at 120 °C in a vacuum oven before printing.This precaution is necessary as these materials tend to absorb noticeable amounts of water under room atmosphere, leading to foaming during printing and adversely affecting surface quality.Despite the initial drying, extended print durations exceeding 90 min can still result in the filament reabsorbing moisture.While printing the high-temperature filaments, the nozzle fan was turned off to prevent warping.For all ironing experiments, unless stated otherwise, a standard 0.1 mm line spacing was chosen.While there are different options for the orientation and shape of the ironing pattern, we found that going in straight lines perpendicular to the layer below yields the best and most homogenous results.

Sample Design for Optimization
The initial step in enhancing surface roughness is to adjust the general flow, ensuring a satisfactory filling without any gaps or overextrusion.General flow refers to the printing of the layers that make up the 3D object as opposed to the ironing flow that is applied after the last layer has been printed.To study the general flow, a sample design consisting of sixteen cuboids, each printed with a different flow value, is used.For each material, the 100% flow rate is calibrated by printing a cube with only one perimeter, measuring the wall thickness, and adjusting the extrusion parameter accordingly until the wall thickness matches the line width.The 3D model of the samples is depicted in Figure 3a.The cuboids are printed using parallel lines along the full length of 50 mm and are characterized in the center to eliminate any effects from the turning points at the edges of the cuboids.Since each layer has the same printing path, the surface profile should accumulate with each added layer, making it easier to detect slight differences in roughness.All cuboids are printed in short succession to minimize variations from temporal humidity and temperature changes and are held together by a frame that serves as a guide to align the samples on the optical profilometer.Each cuboid on the surface is printed in a continuous meandering path before moving to the next cuboid.The second step in enhancing surface roughness is taking a similar optimizing approach for the ironing parameters.The results of ironing somewhat depend on the quality and filling ratio of the layer it is applied onto.Therefore, the samples are designed in a way such that all ironing paths are applied to the same uniform surface with a 5 mm distance to any borders of the print to allow for accurate comparisons.The uniform surface is created by alternating the layer direction to prevent the local accumulation of material that is intentionally brought about in the previous experiment.Each of the four ironing parameters (ironing flow, speed, line spacing, and temperature) is swept with sixteen values, as shown in Figure 3b.For every value of the ironing parameter, three different rectangle sizes were printed and characterized to test if the length of the ironing path had any effect.
To compare ironing with traditional smoothing methods, some samples were submerged in acetone for 10 min or carefully sanded on a straight surface using 1200 and 1800-grit sandpaper before being measured with the profilometer.

Path Generation
Generating the machine commands for the printer from the 3D models is a process known as slicing.In all tests, samples were sliced using Cura (Ultimaker, Netherlands) and printed with 100% solid infill.It is important to note that the ironing parameters changed from version 4.11 to 5.0.In the earlier versions, the ironing flow value is represented as a percentage of the material that is extruded for the regular lines (per millimeter of travel).This value remained constant regardless of the ironing line spacing, resulting in an overall increase in the total amount of material used during the ironing process when the line spacing was set to smaller values.With the introduction of Cura 5.0, which was used for all our experiments, the ironing layers are treated as another regular layer with their ironing flow value which is independent of the regular flow.Additionally, the total material consumption during the ironing process is no longer affected by line spacing variations, as the flow rate is now automatically adjusted accordingly.Consequently, the flow rate can get very low for finer line spacing or low printing speeds.Low flow rates are challenging to control as gravity can cause the molten material to extrude from the hot nozzle prematurely and irregularly.Additionally, low flow rates increase the likelihood of heat creeping up from the nozzle into the heatsink, softening the material too early, and causing the material to expand and blockade the metal tube when pressure is applied from the extruder gear, as shown in Figure 2.
To limit these effects, a high ironing flow was chosen for both experiments.

Characterization
All samples were scanned in two dimensions with an optical profilometer (Bruker, United States) in five different locations near the center and the roughness values were averaged.An image processing algorithm was designed using MATLAB (Mathworks, United States) that calculates the roughness values from profilometry scans for two frequency ranges, as illustrated in the flowchart in Figure 3c.These two frequency regimes could be interpreted as roughness and waviness.When printing (conductive) inks on top of these 3D-printed surfaces, different regimes could be more important depending on the printing method used and the desired feature size.The first step of this algorithm involves a Gaussian blur with a kernel size of 5 Â 5 pixels to remove initial noise from the profilometer scan.While this reduces the amplitude of very high-frequency noise with wavelengths lower than 10 μm, as shown in Figure 3d, it removes  most outliers and fills in missing data.Any remaining areas of missing data are mostly due to being outside the z-scanning range of the optical profilometer, e.g., where the ironing layer was not printed completely because too little material was extruded or due to small deep holes that the profilometer could not illuminate and retrieve data from.Next, the tilt is removed from each sample and a 2D fast Fourier transformation is carried out.Subsequently, low-pass and high-pass Butterworth filters of the first order with a cutoff wavelength of 0.2 mm are used to filter the scans into their high-and low-frequency features.This wavelength was chosen as it is situated between the line spacing for the regular lines (0.4-0.5 mm) and the line spacing of the ironed lines (0.1 mm).In this way, the roughness value of the low-pass filtered image should allow studying if the wavy pattern of the regular lines has been removed by the ironing step without having to worry about newly introduced high-frequency noise.Analogously, we can also compare if new high-frequency noise has been introduced by the ironing operation.Then, all roughness values are calculated using the average absolute deviation and are shown for the two example samples in Figure 3e.
The pixel size of each scan is 0.75 Â 0.75 μm, thus, wavelengths smaller than double this cannot be detected.Similarly, the maximum scanning area of 0.96 Â 0.72 mm does not allow much bigger wavelengths to be detected.Scans that had a high percentage of missing points or where no material was deposited during ironing causing the unironed surface underneath to be measured instead are grayed out in the graphs.In addition, the ironed and unironed surfaces were viewed with a scanning electron microscope (SEM) (Thermofisher Quanta 3D, United States).

Printing and Characterizing the Electronics
After the examination of ironing parameters, conductive inks were printed on top of the ironed substrates using two different printing methods.First, a Nova extrusion printer (Voltera, Canada) was used to print several patterns in silver paste (CM120-7, Creative Materials, United States) with a nozzle size of 125 μm.Furthermore, a custom-built inkjet printer with a 60 μm diameter nozzle (MJ-ATP-01-60-8MX, Microfab Technologies, United States) was employed to print smaller line traces with less viscous silver ink (ANP DGP 40LT15 C, Advanced Nano Products, South Korea).A resistive strain gauge was designed and printed onto an ironed PSU substrate and was consequently fully embedded with another PSU layer which would then form the tip of a pair of printed tweezers.To achieve this, the components were transferred back and forth between the FFF and extrusion printer.
PSU was chosen as it can withstand the sintering temperature of the conductive ink and the temperature needed for sterilization in medical applications.The heated build plate of the FFF machine was utilized for sintering silver on layers in proximity to the base at 170 °C.PSU also exhibits the best low-frequency roughness out of all the high-temperature filaments studied here.Finally, the strain gauge was tested.To track the elongation, we used the video recording from a stationary smartphone and tracking markers in Blender (Blender Foundation, Netherlands).The force was determined using a load cell and the resistance was measured on a 4200 A-SCS Parameter Analyzer (Keithley, United States) with a constant current of 1.0 A.

Parameter Study: General Flow without Ironing
An important parameter for 3D printing is the general flow rate to build up the layers before any ironing is applied.This greatly affects the actual linewidth and height of all printed lines of the printed object.This flow rate significantly influences the actual width and height of all printed lines within the object, thus exerting a notable effect on surface roughness when no ironing is involved.Identifying the optimal flow rate can lead to a threefold enhancement in surface roughness.This improvement in roughness is primarily observed in the longer wavelengths' regime, while the roughness remains relatively constant for smaller wavelengths.A comparable trend can be observed in every material in Figure 4a but with an offset in flow values.At low flow values, gaps can be observed between the extruded lines.These gaps, which are not desirable, have a depth outside the range of our scanning method, resulting in missing data, and therefore do not greatly affect the calculated roughness values even though they lead to large surface roughness.As shown in the progression of the larger PLA profiles in Figure 4b, these gaps are filled with increasing flow, and the confined material spills over to the next line, resulting in ridges on the surface.This surface structure has also been observed in a previous study and has been used for lasering in and demolding flexible LIG circuits from the corrugated PEEK surface. [27]With further increased flow, the ridges grow larger, and the average surface height approaches the midpoint between the highest and lowest points of the ridges and trenches.This is where the standard deviation used to measure roughness reaches its peak as indicated in the overall trend in Figure 4c.Beyond this point, roughness starts to decrease again as the overspills grow wider and the trenches smaller.With the lines becoming taller, they get closer to the nozzle and are more in contact with the previous line and under higher pressure.All these factors contribute to good fusion and fewer sharp edges between the lines.However, this higher flow value is unstable for larger prints as the excess material piles up with each added line, and the geometric accuracy of the parts suffers from the induced tilt.The ABS and PSU (Figure S18, Supporting Information) flow study also shows that increasing the flow further will cause roughness to increase again.
In conclusion, the data suggest that there are two options for general flow values to achieve a smooth surface for most materials.The first option is to find the flow value just before lines start overlapping, and the second is to print with significant overlap.However, the first version has gaps that could potentially result in breaks in the conductive ink printed on top, and the second option may become unstable for larger samples.Therefore, neither option is ideal for printing small features on top.

Parameter Study: Ironing Flow
To further reduce surface roughness beyond what is possible by optimizing the general flow, an ironing tool path is added and the effect of the ironing flow on the surface quality is investigated.The term "ironing flow" only refers to the amount of material that is extruded during the final ironing pass.A range from 17.4% to 83.4% was chosen for the ironing flow values while keeping the general flow of the underlying layers constant.Throughout the ironing process, the z-position of the nozzle remains unchanged from the last layer.If the regular layer before the ironing step is filled with the appropriate amount of material (as discussed in the previous study), the average distance between the printed part and the nozzle should be close to zero.Therefore, an ironing flow of 50% results in half of the material volume from a regular layer being extruded into the top layer again.This close proximity between the nozzle and the completely filled surface enables the application of pressure and smoothing.
Interestingly, the ironing flow did not cause as much variation on the surface roughness compared to the general flow, and the roughness values remained consistently low even for high ironing flow values, as illustrated in Figure 5a.A drastic increase in surface roughness is only observed when there is an insufficient extrusion of material.In such cases, the limited material is unevenly deposited in small heaps on the higher regions of the surface, causing a significant rise in surface roughness.The limited impact of the ironing flow on surface roughness can be attributed to the behavior of excess material, which is akin to how a snowplow pushes aside any surplus snow it cannot compress.This excess material is displaced to the sides of the heated nozzle, where it accumulates neatly, as illustrated in Figure 5b.The ironing flow predominantly influences the height of these material stacks without significantly altering the actual topography or roughness of the ironed layer.In fact, the sample thickness increases linearly with higher ironing flow, which must be considered for components that require precise geometric dimensions (Figure S33, Supporting Information).As shown in Figure 5b, a ridge is formed on the left side of the nozzle due to the accumulation of excess material.This ridge is progressively carried along and grows throughout the entire ironing process.With a higher flow rate, a larger "lip" is left at the final ironed line.While this lip does not affect the primary surface, it may pose clearance issues when subsequently printing electronics on top of it and should be minimized.
With low flow values, the ironing lines of PETG are more visible, and for high values, small ripples perpendicular to the lines appear.In the transition between these two regimes, optimum smoothness and flatness are achieved.For ABS, the smaller  The peak of this general trend is found at around 112% for PLA, 115% for PETG, and 104% for ABS.flow values while a downward trend in roughness was observed in the two larger sample sizes (Figure S14, Supporting Information).This observation suggests that overextrusion is more problematic for smaller samples, and thus requires finer tuning.In contrast, when analyzing the PEI samples (Figure S24, Supporting Information) it was found that the surface roughness of the largest sample drastically increases with the ironing flow.This implies that further adjustments are necessary for the larger samples in this case.Upon close inspection of the scans, the 0.4 mm wide lines of the layer beneath could still be seen for low flow values (especially noticeable for ABS) but became less noticeable with a higher flow as more material is compressed into the surface and valleys are filled out better.In summary, the surface roughness stays mostly constant for a wide range of ironing flow values and only has one significant peak when underextrusion occurs.By avoiding easy-to-detect underextrusion and excessive build-up around the edges of the ironed area, it should be straightforward to find good values for this parameter.

Parameter Study: Ironing Speed
To investigate the effect of ironing speed, samples were tested using a range of values from 1 to 75 mm s À1 , which is the maximum print speed of our printer.Ironing slower than 1 mm s À1 would result in an extensive print duration and, although the obtained results are interesting at 1 mm s À1 , they lack usability and do not justify further expenditure of time.
In the low-speed regime, ranging from 1 to 10 mm s À1 , a maximum for the overall roughness, a maximum for the lowfrequency roughness, and one for the high-frequency roughness can be found at different speeds in the graphs in Figure 6a.This common trend is depicted in Figure 6c.At the lowest speed of 1 mm s À1 , the nozzle is almost stationary.This allows heat to transfer into the material in all directions around the nozzle, resulting in the remelting of the entire surrounding area and the removal of small features and lines, as illustrated in Figure 6b.While this produces a very glossy surface finish, it also results in high low-frequency roughness (waviness).The reason for this waviness could be a combination of poor temperature stability of the nozzle and low flow rate.Depending on the tuning of the proportional-integral-derivative controller for the heating element and thermistor in the printing tool head, the temperature can fluctuate by a few degrees over a period of a few seconds or minutes.The temperature variation can affect the flow rate, causing each ironed line to have a slightly different amount of material and height.With slower print speeds, this temperature fluctuation has an effect within smaller distances.Another factor that could be responsible for the large waviness is the low material flow rate that is associated with low speeds.The hot molten material can leak from the nozzle due to gravity if the feeding speed is not fast enough and the heat can also travel upward along the filament path softening the material too early.Additionally, the extrusion precision is inherently limited by the step resolution of the extruder stepper motor.At very low flow rates, this limitation becomes more apparent.For the higher-temperature materials, Proportions are not to scale.The height of the ironed layer can be regulated by adjusting the ironing flow, even though the position of the nozzle in the z-axis always remains the same as the last layer.The results indicate that increasing the ironing flow has only a small impact on the surface roughness compared to the general flow, as the primary effect is only a shift of the surface profile in the z-direction.
this method of slow printing and remelting appears more ineffective.A better heating chamber might reduce the dissipation of heat and allow for better remelting.Overall, further tuning of other parameters, such as ambient and nozzle temperature and a better extruder with high extrusion precision and good temperature stability might make this method suitable for cases where a glossy surface finish is desired.
At slightly higher speeds in the range of 5-20 mm s À1 , a decrease in the low-frequency roughness is observed, while a significant increase in the higher frequencies is found.The heat does not have enough time to transfer into the surrounding area and remelt the material.While the deposited material leaves a small trail of heat, it cools down too quickly before the nozzle can return to the warm material in the next line.Therefore, newly added material does not fuse well, resulting in a defined profile of parallel lines with high roughness, especially at higher frequencies.
In the high-speed regime from 20 to 70 mm s À1 , all roughness values are overall lower and remain mostly constant with speed for the low-temperature materials.For the high-temperature Figure 6.Effect of the ironing print speed on surface roughness.a) Roughness data that were calculated from profilometry scans with our algorithm for PLA, PETG, and ABS.Selected scans are shown for every second ironing speed value below the graphs.The results for the three high-temperature materials PSU, PEI, and PEEK, and the other sample sizes are presented in the Supporting Information.b) Schematic top view of heat gradient for different ironing speeds with selected sample scans of PLA.Low speeds allow the heat to disperse deeper and wider into the material causing almost complete remelting of the surface.A slight increase in speed can result in a longer heat gradient, but the overlapping lines will still be cooled down by the time the tool head moves to the adjacent line.High speeds will cause sufficiently larger areas of material to still be hot by the time the nozzle returns to the same area on the adjacent line, which allows better fusion between the last line and the newly printed line.c) General trend of roughness found for the ironing flow resulting from this.materials, a slight upward or downward trend is observed depending on the material.For PSU (Figure S20, Supporting Information) and PEEK (Figure S30, Supporting Information), a transition from lines to perpendicular ripples is observed, with the optimal print speed being found at the transition point (20 mm s À1 for PSU and 30-35 mm s À1 for PEEK).The appearance of ripples could be linked to the increased flow rate at higher speeds.At these flow rates, it is possible that PSU and PEEK might not be heated fast enough inside the nozzle, causing it to extrude in irregular bursts.Conversely, the higher thermal conductivity of PEI may enable quicker heating, which could explain why the surface roughness of PEI is optimal at a higher speed of 70 mm s À1 (Figure S25, Supporting Information). [28]n conclusion, higher ironing speeds (70 mm s À1 and upward) produce a surface with low roughness for most materials while also being time efficient.For ABS, PSU, and PEEK, the speed should be lowered to 20-30 mm s À1 for more consistent results.While very smooth, these surfaces still appear matte (Figure S1, Supporting Information).A glossier surface finish can only be achieved with very low speeds in the range of 1 mm s À1 .However, these are very time-consuming and would require more fine-tuning to make them less wavy.

Parameter Study: Ironing Line Spacing
In terms of line spacing, an intuitive hypothesis would be that the surface roughness improves with finer line spacing.Indeed, a downward trend in high-frequency roughness is observed in Figure 7a as the line spacing is decreased from large values.However, when line spacing is reduced to a very small value, the flow rate becomes very low and the two issues discussed in the speed section arise.The first issue is inconsistent extrusion, which is particularly apparent in ABS and PEEK.The second issue is that temperature oscillations and the resulting variations in flow rate become more noticeable as the lines are brought closer together.These oscillations are visible in the profilometer scans of PLA and PETG for a line spacing of 0.01-0.03mm and affect the low-frequency roughness as the wavelengths of the planar waves now fall within the 0.4-1 mm range.Another issue is that line spacing of 0.01 mm is close to the precision limit of the printer, which has an xy-resolution of 0.005 mm.For these reasons, the finest line spacing of 0.01 mm does not produce the best surface roughness, except for PEI where an impressively low surface roughness (compared to the unironed PEI surface) of 701 AE 19 nm can be achieved on the larger samples.This might be possible because the temperature fluctuations of the high-temperature hot end (HT3-450) are not as severe compared to the low-temperature hot end (MK1-250) with which the PLA, PETG, and ABS samples were printed (Figure S37, Supporting Information).Also, PEI appears to exhibit a high level of precision in dosing and extrusion at low flow rates.For PSU, the effect of line spacing is dependent on the size of the sample (Figure S21, Supporting Information).Whereas the smallest PSU sample exhibits the smoothest surface with a fine 0.02 mm line spacing, the other two sample sizes show the lowest roughness at 0.15 mm.For PEEK, line spacing has almost no effect on roughness, unless it is too fine causing underextrusion (Figure S31, Supporting Information).ABS shows the smoothest surface for a wider line spacing of 0.15 mm while for PLA and PETG, a medium line spacing of 0.04-0.05mm was found to produce the best results.In summary, line spacing is a vital factor in controlling the ironing process.Our findings contradict the expectation that a very fine line spacing of 0.01 mm would consistently produce the best smoothness because the limitations that come with low flow rates are already reached at this line spacing.Instead, the results of this study indicate that the optimal line spacing falls within the range between 0.05 and 0.15 mm.However, with improvements in the temperature stability of the nozzle and improvements in the filament path of the hot end, the waviness caused by the temperature oscillations and problems with low flow rate could be mitigated.This would allow for a further reduction in line spacing until the maximum resolution of the printer is reached, potentially leading to very low surface roughness in the higher frequencies, which may justify the additional printing time that comes with a finer line spacing.

Parameter Study: Ironing Temperature
The temperature range for this study was chosen based on the manufacturer's recommended printing temperature range.The lower limit was reduced by 10-20 °C to investigate the lower limit.For PLA, PETG, and ABS, the upper limit is constrained by the MK1-250 printhead's maximum temperature of 250 °C.The influence of ironing temperature on the surface roughness was found to be mostly negligible except at extremely high or low temperatures.At low temperatures, the extruder is unable to sufficiently soften and extrude the material, resulting in poor surface roughness.Conversely, at very high temperatures, the material can degrade and the moisture within the filament may lead to increased foaming and poor surface quality.Furthermore, high temperatures can result in heat creeping into the heatsink of the extruder, which can cause the material to become soft too early, resulting in expansion and blockage of the feeding tube when pressure is applied from the extruder motor.Both extremes of the temperature range and their effect on roughness can be seen in the ABS graph in Figure 8. Between these extremes, no significant change in the surface roughness values can be found for almost all materials.However, a common trend can be observed across various materials where the surface structure transitions from irregular ripples to straight lines with increasing temperatures.This is especially noticeable for PSU where the roughness decreases steadily with higher temperatures.For PETG, the best roughness values are found in the transition between the two regimes.

Parameter Study: Sample Size
All the ironing parameter studies have been performed for three different sample sizes (15, 30, and 45 mm) corresponding to three different lengths of the ironing lines.All detailed results for 15 and 45 mm can be found in the Supporting Information.For shorter lines, the nozzle returns sooner to the already printed area, and the heat concentration is higher.This either allows for better merging of the material or causes unpredictable reflow of the soft material.Another factor to consider is the decrease in the average speed with smaller samples since more time is spent with acceleration and deceleration.For very small samples, the desired speed might not even be reached while printing the lines.Also, the waves caused by the temperature fluctuations are spaced differently depending on the width of the sample, which is especially noticeable in the PLA line spacing study.For further analyses, the roughness values from all our ironing studies are averaged based on the sample size and presented in Figure 9.The whiskers and quartiles in the boxplot indicate the magnitude of the effect that each parameter has on roughness in that study.The outliers are mostly due to samples where the parameter values did not allow for a defect-free ironed top surface.While the overall shape of the roughness trends in Figure 4-8 was usually similar for all three sample sizes, noticeable variations in magnitude and scaling become apparent in the averaged roughness values presented in the boxplot.Therefore, optimal settings remain constant regardless of sample size, but differences in the overall surface quality can be expected for different sizes and shapes of the sample.Therefore, to generate the best result for a given geometry, the ironing path could be rotated or split intelligently so that the optimal path length is used.In some cases (e.g., ironing flow of PEI), the sample size Figure 9. Boxplot of the roughness values grouped by sample size and averaged for each parameter experiment.The whiskers and how much variation in was achieved by varying the parameter of that study.For PLA and PSU, the average roughness decreases slightly with smaller samples while for the other materials roughness increases with smaller samples.The sample size directly translates to the path length in the ironing process and therefore is an important factor to consider when designing the geometry of the ironing path.Box plots are also shown for the unironed samples averaged over the general flow study.Note the scales of the y-axes are different in different graphs to improve readability.greatly increases the standard deviation toward higher roughness values.This suggests that optimal settings need to be chosen especially carefully if the path geometry cannot be altered.In some rare cases, opposite trends are found within a given material-parameter combination, such as the ironing flow in ABS and PSU the line spacing in PSU (Figure S21, Supporting Information), where the slope of these graphs changes from positive to negative for different sample sizes.In such cases, the optimal surface finish on irregularly shaped samples can only be achieved by adjusting the print settings for each line based on its length.Alternatively, the machine path could be designed to always have a similar ironing path length, for example, by adding slight curves.

Microscale Characterization with SEM
Based on the collected datasets, optimized ironing settings were selected (Table S36, Supporting Information) and the resulting samples were analyzed under the SEM (Figure 10).While the SEM also revealed significantly reduced large-scale waviness in the ironed samples, at the microscale, the ironed PETG and PSU samples exhibited a rougher texture compared to their unironed counterparts.This fine microscopic roughness could not be correctly captured by the profilometer due to its limited resolution and the prefiltering method in the image processing algorithm.This fine roughness may be attributed to the hygroscopic nature of these materials, as the expansion of trapped moisture during the ironing process could be a possible cause.As ironing is performed with a much lower flow rate than the regular lines, the filament is pushed more slowly through the orifice of the nozzle.This increased contact time allows for more heat to be transferred from the nozzle to the material, potentially leading to an increase in foaming.Additionally, the nozzle drags parts of previously laid lines because the outside surface of the nozzle comes into greater contact with the material during the ironing operation causing fine peaks and even tiny hair-like structures which are particularly noticeable in the ironed PEEK sample.Furthermore, the SEM images reveal that the ironing lines on the PEI surface are very consistent, confirming its high level of controllability and steadiness, which PEI has shown for tight line spacings and low flow rates.Similarly, the ironed ABS sample also demonstrates straight and steady lines.

Comparison with Other Smoothing Methods and Substrates
In Figure 11, the smoothness of 3D-printed samples with and without ironing is compared to the smoothness of traditional smoothing methods.For the treatment in acetone, the samples were submerged inside acetone for 10 min and then left to dry before being measured with the profilometer.While this method is usually only used on ABS prints, it also shows an effect on PLA and PETG.Interestingly, for PLA and PETG, the low-frequency waviness of the samples is reduced while the smaller features and lines remain mostly intact.In contrast, for ABS, the lowfrequency waviness remains almost the same while all highfrequency noise is reduced drastically, confirming that a glossy surface can be achieved on ABS with acetone.Still, the roughness values of the ironed samples surpass the improvements by acetone.The overall roughness after ironing is comparable to a Figure 10.SEM images of ironed and unironed surfaces.While PETG and PSU show a more roughened-up surface on the micrometer scale compared to the unironed version, PLA, ABS, and PEI remain mostly the same after the ironing step but with less waviness.In the case of PEEK, the large profile of the wide lines is replaced with finer lines but with an inferior high-frequency roughness.
surface that was sanded with sandpaper somewhere between 1200 and 1800 grit.The waviness of the sanded samples is slightly less than that of the ironed samples, indicating that the sanding process results in good flatness, as expected.However, the improvement does not justify the effort and added complexity that an integrated sanding tool head would entail in a multitool head printer.Compared substrates used for printed electronics, the roughness of the ironed samples is comparable to an FR1 PCB board but higher than a flexible PET substrate.This means that ironed 3D-printed surfaces can likely be used for printed Figure 11.Comparison of different methods to improve surface quality.The histograms and surface plots share the same scale in the first two rows and a different scale bar is shared for the third to fifth row.The ironed samples exhibit similar roughness as traditional PCBs and sanded 3D-printed surfaces, but higher roughness than flexible PET substrates.
electronics, but there is still room for improvement to match higher-quality substrates.

Final Assessment of Surface Roughness Improvement by Ironing
A summary of the improvement in roughness by ironing is shown in Table 2.It should be noted that, for a fair comparison, we use the lowest roughness that we found in the samples with and without ironing.Therefore, improvements are even larger compared with unoptimized 3D printing parameters.These results show that ironing has a very substantial impact on the overall roughness of the commonly used materials PLA, PETG, and ABS with reductions in roughness ranging from 83 to 89%.Most of this improvement is found for the low-frequency roughness.Therefore, we can conclude for these materials that ironing is very effective at eliminating the line patterns greater than 0.2 mm in width while there is only a minor improvement if not worsening of the microscopic surface as confirmed by the SEM images.
In contrast, the results for engineering-grade filaments such as PSU, PEI, and PEEK exhibit more varied outcomes.PEI demonstrates the highest level of controllability throughout our studies, allowing for a remarkable roughness reduction of approximately 95% when optimal settings are employed.Conversely, with PEEK, we observed a more modest reduction in roughness, around 26%, and an increase in high-frequency roughness when using ironing.An alternative approach for enhancing PEEK's surface quality could involve the use of carbon fiber-reinforced PEEK, as it has been shown to yield a superior surface roughness. [29]PSU falls somewhere in between, with an improvement that is slightly less than that of PLA.
The challenges observed with higher glass transition temperatures of PSU, PEI, and PEEK are further explored in Figure S35 (Supporting Information), correlating material properties with observed roughness trends.While establishing a direct and simple relationship is intricate, lower glass transition temperatures and thermal conductivity generally appear to be beneficial for optimal ironing results.Lower glass transition temperatures may enhance material reflow, while thermal conductivity concentrates heat, further enhancing remelting behavior.
Some of the parameter boundaries in our tests are tied to the limitations of our printer.Recent advances in printer technology, such as higher print speeds up to 500 mm s À1 and improved extruders that can apply more feeding force on the filament and have sharper temperature gradients, may be able to push these boundaries and achieve more precise control of flow rates and print speeds.Additionally, the integration of LiDAR scanners into the printhead of consumer printers has the potential to enable autotuning of flow and ironing values without the need for expensive equipment in the future.

Integrating the Ironed Substrates with Printed Electronics
Planar inductor and capacitor patterns are designed and printed onto ironed PLA to assess the controllability of the ink.For comparison, the same circuit design was printed onto a regular 3Dprinted PLA sample without ironing.To prevent collisions between the fine nozzle and the rougher unironed substrate, the dispensing height had to be raised from 20 to 80 μm.The results are presented in Figure 12.Visible breaks and an irregular trace width are found on the substrate without ironing, which will render the printed devices nonfunctional.Conversely, the ironed substrates exhibit more consistent lines without breaks.By using an ironed surface, the dispensing height can be reduced further, resulting in sharper corners and more consistent traces.This enables the creation of tighter wound coils, highercapacitance interdigitated capacitors, and denser and more compact circuit boards.As a result, the height of the conductive traces is also reduced, which can be compensated for by stacking multiple layers.
The process of dispense printing also requires drying and sintering.For this, the heated build plate of the FFF machine was utilized for sintering silver on layers in proximity to the base at 170 °C.For higher layers, the sintering step could also be accomplished by increasing the ambient temperature of the printer enclosure or by embedding the circuit with another hot layer.
Furthermore, to demonstrate the feasibility of extrusion printing electronics on or inside printed parts, a resistive strain gauge was printed on all the ironed materials.The average resistivity of the gauges varies between the materials and prints, but all are capable of detecting uniaxial load.A good linear response is observed with increasing load, but a less linear response with decreasing load.Additionally, the total resistance increases irreversibly with each cycle.The first cycle of the printed sensor shows the largest irreversible increase in resistance.A similar degradation has been observed other researchers and is likely due to the connected silver particles being irreversibly torn apart. [30]inally, a smart-printed tweezer was printed with a fully embedded meander structure strain sensor inside the material, as shown in Figure 13.By choosing PSU as the structural and isolator material, the embedded sensor should be suitable for applications that require sterilization and exposure to many chemicals.Additionally, the process of embedding significantly reduced oxidation and aging in comparison to unembedded circuits that were exposed to the lab environment.The dimensions of the 3D model are somewhat altered by the addition of an ironed layer, and while smooth on the microscopic level, it is slightly curved along the edges of the model.Therefore, it is necessary to remeasure the surface after ironing to ensure that the electronics are printed correctly.The NOVA printer used here can create a surface profile using a leveling probe.This process is mandatory for good results and could also allow for printing on purposely curved surfaces, such as those created by nonplanar printing.

Conclusion
In the first part of this study, the potential of ironing for creating smooth top surfaces on parts printed by FFF printers was quantified.Activating the ironing feature results in significant improvements in the surface roughness of PLA, PETG, and ABS.Optimization of each ironing parameter yields only a minor improvement for these low-temperature materials, suggesting that ironing can remain effective across a broad range of settings.The lowest roughness value was observed for PLA, where an absolute roughness of 464 AE 14 nm was achieved, which is much lower than previously reported values, although this may be partly due to differences in sample design, measuring tools, and data processing.Based on the comparison with other smoothing methods, it can still be concluded that the ironed samples in PLA, PETG, and ABS are similar in smoothness to surfaces that were sanded with 1200-grit sandpaper or FR1 boards.In other research the lowest roughness of the ironed samples is compared to the average roughness of unironed samples and a reduction of 72.2% is reported. [26]Doing the same comparison for our research, we find reductions from 93.7% to 95.8% for the three low-temperature materials and a very high reduction of 96.6% for PEI, 84.8% for PSU, and 61.4% for PEEK.If we compare the mean values of both the ironed and the unironed samples, we still find a reduction ranging from 83.7 to 92.9 % for PLA, PETG, and ABS, 46.2% for PSU, and 75% for PEI.In the case of PEEK, the roughness increases on average by 29.4% with ironing.This is because these average values also include samples where the parameter values led to suboptimal results.Therefore, we additionally compared the lowest values of both the unironed and ironed samples and found reductions in roughness ranging from 83.1 to 89.4% for the three low-temperature materials and 75.5% for PSU, 95.2% for PEI, and 26.2% for PEEK showing that with the correct settings, the ironed samples always surpass the . Microscope images of dispenser-printed silver ink on 3D-printed PLA substrates.In the first image of each series, the PLA is not ironed, and the silver ink shows periodic breaks and varying line widths due to the surface profile of the unironed substrate.In the second image, the PLA has been ironed.In the third image, the same ironed substrate is used but the conductive ink is printed at a lower dispense height.Various shapes are tested for consistency: a) lines perpendicular to the printing/ironing orientation, b) fingertips and wider lines, and c) corners of a square coil.d,e) The entire printed capacitor and square coil on an ironed substrate with a dispense height of 80 μm.
unironed samples even for PEEK.However, the trends and results for PSU, PEI, and PEEK varied more, therefore tuning of ironing settings is crucial for these materials, and more extreme values were found to be Overall, the collected data show that there is a higher potential in ironing than previously reported and the recorded trends can guide a fine-tuning process for each material.Further optimization for different materials could be achieved by characterizing the material properties of 3D printing filaments at the relevant temperatures and modeling the temperature distribution during printing.Since ironing speed, line spacing, and temperature should be comparable across different FFF printers, these parameters could be set first and then be followed by the adjustment of the printerspecific ironing flow so that no under-or overextrusion occurs.Currently, ironing is only designed to be used on top surfaces of models.[33][34][35][36][37] Besides the obvious aesthetical reasons, this could also be useful to reduce friction and wear between printed parts or improve the outer surfaces for aero and fluid dynamic applications. [38]Future work could also explore whether ironing affects the mechanical properties and crystallinity of the materials, potentially yielding similar effects to annealing.This research on ironing was driven in particular by the goal of reliably printing small electronics on top of 3D-printed parts.To evaluate this, ironed and unironed substrates were tested by printing several patterns using silver ink and comparing their features.We found that the circuit patterns on the ironed substrates showed significantly better pattern consistency than on the unironed surfaces.Based on these findings, a functional strain gauge was printed and embedded inside the tip of fully The first four layers were printed on the FFF printer in PSU and the top layer was ironed.For the second step, the specimen was removed from the FFF printer and aligned on the Voltera NOVA extrusion printer by using the outer frame.Then a strain gauge design was printed with conductive silver ink.In the third step, the part was realigned on the FFF printer, and the electronics were fully embedded by another PSU layer while leaving the contact pads exposed.The finished tip (c) was cut out and inserted into an FFF-printed tweezer body, as shown in (d).e) Change in resistance under uniaxial load slowly applied for one cycle.f ) The resistance of the sensor over 500 cycles.
3D-printed tweezers.While strain and damage sensors, heating elements, capacitors, and inductors are simple structures that can be printed using a single material, printing more complex structures that incorporate organic semiconductors and dielectrics to produce diodes and transistors should be possible in the future. [39,40]By also adding bio-or chemically sensitive materials, a variety of advanced sensors could be directly printed onto or embedded within geometric structures that guide fluids or serve other purposes.
In conclusion, this article demonstrates that ironing is a viable method to facilitate the integration of FFF and extrusion-printed electronics, which have both made significant advancements in recent years.By combining these technologies into a single, affordable system, individuals from industrial and research sectors may be able to fabricate complex, custom, biocompatible, flexible, and fully functional electromechanical components in a rapid prototyping fashion.

Figure 2 .
Figure2.Cross-section of an extruder hot end affected by heat creep.The combination of slow extrusion speed, constant pressure, and heat creeping upward in the filament causes the filament to expand in colder areas which causes higher friction inside the feeding channel or even total clogging of the hot end.In that case, the molten polymer can leak irregularly from the nozzle due to gravity.

Figure 3 .
Figure 3. a) Sample design used to sweep the overall printing flow from left to right with 16 different settings.All layers have the same printing orientation to amplify the buildup of the material over all layers.Each of the sixteen rectangles is scanned on five different spots in the middle as shown on the right.b)Sample design used to sweep each of the ironing parameters from top to bottom.To prevent artifacts from the edges of the model and the lower layers, the ironing is performed in the middle of one uniformly printed surface.To get this base layer as evenly as possible, the printing direction of the layers beneath is alternated.Each rectangle represents the areas that are ironed perpendicular to the top layer orientation.Additionally, the width of the ironed area is varied in 15 mm steps to study whether the length of each ironing path line has an effect.c-e) Overview of the image processing algorithm.c) Flowchart of the main processing steps for two selected samples from our study.d) Frequency response from values returned by the algorithm with ideal sine waves of different wavelengths as input.e) Resulting roughness values for the two samples in (a).The two samples show high roughness values for different frequencies: sample 1 is mostly flat but contains ripples with a shorter wavelength.Meanwhile, sample 2 has a smooth surface but is wavy with a much bigger wavelength.Depending on the application, one or the other roughness could be more important.

Figure 4 .
Figure 4. Effect of the general flow on surface roughness without ironing.a) Roughness data that were calculated with our algorithm for PLA, PETG, and ABS.Selected scans are shown for every second flow value below the graphs.The results for the three high-temperature materials PSU, PEI, and PEEK, and the other sample sizes are presented in the Supporting Information.b) Larger cross-section of the PLA surfaces with varying flow values.As the flow is increased, lines start overlapping and material spills over and accumulates.c) Schematic general trend of roughness for the general flow with some reference points from the scans in (b).The peak of this general trend is found at around 112% for PLA, 115% for PETG, and 104% for ABS.

Figure 5 .
Figure 5.Effect of the ironing flow on surface roughness.a) Roughness data that were calculated from profilometry scans with our algorithm for PLA, PETG, and ABS.Selected scans are shown for every second flow value below the graphs.The results for the three high-temperature materials PSU, PEI, and PEEK, and the other sample sizes are presented in the Supporting Information.b) Schematic cross-section of material build-up during ironing.Proportions are not to scale.The height of the ironed layer can be regulated by adjusting the ironing flow, even though the position of the nozzle in the z-axis always remains the same as the last layer.The results indicate that increasing the ironing flow has only a small impact on the surface roughness compared to the general flow, as the primary effect is only a shift of the surface profile in the z-direction.

Figure 7 .
Figure 7. of the ironing line spacing on surface roughness.The graphs show the roughness data that were calculated with our algorithm for PLA, PETG, and ABS.Selected scans are shown for every second line spacing value below the graphs.The results for the other two high-temperature materials PSU and PEEK and all other sample sizes are presented in the Supporting Information.While a large line spacing is preferable for ABS, in the case of PEI a very smooth surface can be created by choosing the smallest line spacing.

Figure 8 .
Figure8.Effect of the ironing temperature on surface roughness.The graphs show the roughness data that were calculated with our algorithm for PLA, PETG, and ABS.Selected scans are shown for every second temperature value below the graphs.The results for the other two high-temperature materials PEI and PEEK and the other sample sizes are presented in the Supporting Information.For the PETG and PSU samples, a transition from ripples to straight lines can be observed in the surface plots with increasing temperature.

Figure 13 .
Figure 13.Overview printed tweezers with integrated printed strain gauge.a) Computer-aided design (CAD) drawing of the tweezer tip with an integrated strain gauge.b) Manufacturing process for the tip of 3D-printed compliant tweezers.The first four layers were printed on the FFF printer in PSU and the top layer was ironed.For the second step, the specimen was removed from the FFF printer and aligned on the Voltera NOVA extrusion printer by using the outer frame.Then a strain gauge design was printed with conductive silver ink.In the third step, the part was realigned on the FFF printer, and the electronics were fully embedded by another PSU layer while leaving the contact pads exposed.The finished tip (c) was cut out and inserted into an FFF-printed tweezer body, as shown in (d).e) Change in resistance under uniaxial load slowly applied for one cycle.f ) The resistance of the sensor over 500 cycles.

Table 1 .
List of 3D printing materials used in this study.

Table 2 .
The lowest roughness values found in the studies for the samples with and without ironing.The last column shows the reduction that can be achieved by ironing for the absolute, high-frequency, and low-frequency roughness.